Compare commits

...
Author SHA1 Message Date
Brent Vatne e5a6626bde Release 0.21.0 2026-10-23 12:03:55 -07:00
Brent Vatne c9ab89025b Screens: a filmstrip layout beside the grid, switched at the top right; the sidebar is headed by its find field (#64) 2026-10-23 12:03:24 -07:00
Brent Vatne 72bdb8772c Ideas: capture base and head under the same conditions (#63) 2026-10-21 14:21:54 -07:00
Brent Vatne 4ed584c990 Release 0.20.7 (#62) 2026-10-20 17:17:14 -07:00
Brent Vatne f2de6c9002 Repack sharing: list the EAS environment without --non-interactive, and check every eas flag against the pinned eas-cli (#61) 2026-10-20 17:13:34 -07:00
Brent Vatne d992899292 Release 0.20.6 (#60) 2026-10-20 16:42:02 -07:00
Brent Vatne 8e86e6a14e Argent capture waits for a new session's tool server, and a side keeps its light captures when only its dark session fails (#58) 2026-10-20 16:38:49 -07:00
Brent Vatne ebbcbb62c4 A failed hosted repack starts again, and a run's matrix jobs share one repack of each commit (#59) 2026-10-20 16:34:15 -07:00
Brent Vatne 29f1f0721f Release 0.20.5 (#57) 2026-10-20 14:18:37 -07:00
Brent Vatne f620577e30 A review waits beside head for a baseline of its base commit that is still capturing (#56) 2026-10-20 14:15:25 -07:00
Brent Vatne 6c927b1f22 Release 0.20.4 (#55) 2026-10-20 13:50:17 -07:00
Brent Vatne b72a70f7d8 Capture asks again when simulator:control --list fails, and a session that can't switch to its appearance stops before folding (#54) 2026-10-20 13:46:16 -07:00
Brent Vatne fecf040ce5 Release 0.20.3 (#53) 2026-10-20 12:22:14 -07:00
devin-ai-integration[bot]andbrent e47399440f Store opened iPhone Duo screenshots upright in explore:binary/session (#51)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-20 12:18:17 -07:00
devin-ai-integration[bot]andbrent 46b46bb184 Add --replay to explore:binary and explore:session (#50)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-20 12:11:32 -07:00
Brent Vatne 942bdaad40 Capture waits for a focused field's keyboard, and the comment names off-screen changes (#52) 2026-10-20 12:09:40 -07:00
Brent Vatne ed31b0c767 Release 0.20.2 (#49) 2026-10-19 19:50:28 -07:00
Brent Vatne b1edd7ea81 Reviews survive one target's flaky session: resend a tool whose service didn't start, and report the targets that captured (#48) 2026-10-19 19:47:06 -07:00
Brent Vatne 50230b7f93 Viewer previews: wait 20 s for the alias, then link it with a going-live note (#46) 2026-10-19 19:18:48 -07:00
Brent Vatne 33f31da08e CI: retry a failed simulator:control action and a GitHub lookup the Actions API hasn't caught up on (#47) 2026-10-19 19:18:43 -07:00
Brent Vatne 7fc0f34c9b Release 0.20.1 (#45) 2026-10-19 17:23:51 -07:00
devin-ai-integration[bot]andbrent 6daf0a4391 Phone header: the pill leaves out the app's name, then the PR link, before anything overlaps (#43)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-19 17:20:22 -07:00
Brent Vatne 1796a9b810 Viewer: screenshots fade in, go blank on a target switch, and Compare reads the next screen ahead (#44) 2026-10-19 16:17:56 -07:00
Brent Vatne 87aec5b820 Screens, search and the counts leave out routes a public report never captured, as the Map does (#42) 2026-10-19 12:51:11 -07:00
Brent Vatne 89f18cad05 README: Screens and the PR comment open full size when clicked (#41) 2026-10-19 10:38:10 -07:00
Brent Vatne bb8b73e16e README: refresh for 0.20.0 (#40) 2026-10-19 10:25:49 -07:00
Brent Vatne 7360502e86 Release 0.20.0 (#39) 2026-10-19 09:41:41 -07:00
Brent Vatne 6539b5c002 Source planning: follow every single-directory tsconfig paths alias (#38) 2026-10-19 09:28:17 -07:00
Brent Vatne 422cf8caf4 Viewer: theme, full screen and shortcuts in a "…" menu; the top right keeps only the Map's camera (#37) 2026-10-19 09:24:13 -07:00
devin-ai-integration[bot]andbrent b281c734f4 Viewer: make the Inspector resizable from its left edge (#34)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 22:39:36 -07:00
Brent Vatne c99c26f9bb Release 0.20.0-next.0; prereleases publish under the next tag (#36) 2026-10-18 22:09:03 -07:00
Brent Vatne 9150a24cd9 Map: ring the selected screen instead of fading the rest; Return opens the screen the arrows reached (#35) 2026-10-18 22:05:21 -07:00
devin-ai-integration[bot]andbrent 2c5f8f85a3 Compare: outline changed areas on After only, linked to Before on hover (#33)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 21:52:45 -07:00
devin-ai-integration[bot]andBrent Vatne 23227ca55c capture.maxSessions: a review-wide session ceiling, spent on separate light/dark sessions when that pays (#32)
Co-authored-by: Brent Vatne <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 21:42:28 -07:00
devin-ai-integration[bot]andBrent Vatne ed1bf34434 Faster capture: parallel compare with early-abort region search, one Argent bridge thread per capture (#29)
Co-authored-by: Brent Vatne <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 21:29:58 -07:00
Brent Vatne bdf9e597d3 Large moves read as one moved region instead of fragments (#18) 2026-10-18 17:03:07 -07:00
devin-ai-integration[bot]andbrent 9063de5c52 Inspector: stack Before over After for landscape screenshots (#24)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:55:35 -07:00
devin-ai-integration[bot]andbrent 21ed8c06f6 Text that differs between a build's own light and dark captures is runtime state, not a change (#28)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:49:03 -07:00
devin-ai-integration[bot]andbrent 6795f19d04 Viewer: ⌘K is a command palette (#22)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:47:31 -07:00
devin-ai-integration[bot]andbrent 990e2ecc8d Ignore a foldable's rail status-bar pill like a phone's status bar (#23)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:44:10 -07:00
devin-ai-integration[bot]andbrent 95be8b5248 Viewer: calmer Map hover — no screen fading, Open in Inspector as a tooltip (#30)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:40:12 -07:00
devin-ai-integration[bot]andbrent 6fa7cba087 Viewer: switch targets in place, glide [ ] and +/- from the key down, keep screenshots up while they swap (#20)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:38:35 -07:00
devin-ai-integration[bot]andbrent f728eb187b Quiet PR comment: count + report link heading, per-device subtitle, changed screens as a list, previous run kept while a new run works (#19)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:36:23 -07:00
devin-ai-integration[bot]andbrent d069a892dc Clip diff regions that cross a system band; Android status-bar band 7 % (#25)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:33:21 -07:00
devin-ai-integration[bot]andbrent 4499a999d1 Viewer: / finds screens on the Screens page (#21)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:32:31 -07:00
devin-ai-integration[bot]andbrent 929ac82765 A comment or whitespace-only edit to a source file changes no screen (#26)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:31:59 -07:00
devin-ai-integration[bot]andbrent d5d6a5ff72 Inspector says why an excluded or redirect route has no screenshot (#31)
Co-authored-by: brent <[email protected]>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
2026-10-18 16:31:26 -07:00
devin-ai-integration[bot] 5c19061b0f Excluded routes leave the public map instead of showing as "Not compared" (#27) 2026-10-18 14:45:06 -07:00
Brent Vatne ec75e37e45 Release 0.19.8 2026-10-18 10:00:42 -07:00
Brent Vatne 1254a99841 Viewer preview fixtures: one report per target (#16) 2026-10-18 10:00:26 -07:00
Brent Vatne c2bae00f31 Fix flaky CI: stray SIGTERMs, port-lock race, fixture colour collisions, mid-review downloads, cold-Vite reloads (#17) 2026-10-18 09:56:13 -07:00
Brent Vatne 65b79684f3 Hosted previews serve the report as static files under the token, with no encryption and no Explore expiry (#14) 2026-10-18 09:45:24 -07:00
Brent Vatne b8b43ee2c8 Hosted previews open from a manifest, load screenshots on demand, and switch targets in place (#11) 2026-10-18 09:45:23 -07:00
Brent Vatne c59ccce32b PR comment map: opt in with report.comment.map; README: Compare at full width; release 0.19.7 (#13) 2026-10-17 21:54:20 -07:00
Brent Vatne c24049c443 README: animated hero and Compare, a map of changed screens in the PR comment (#12) 2026-10-17 21:31:58 -07:00
Brent Vatne cefeb8059a Release 0.19.6 2026-10-17 20:44:52 -07:00
Brent Vatne 078902710b Capture iPhone Duo on EAS Simulator with eas simulator:control, for both controllers (#9)
Capture iPhone Duo on EAS Simulator with eas simulator:control, for both controllers

- eas-cli 25.2.0: app.approveUrlScheme after install, appearance.set for Argent on
  iOS EAS sessions, and springboard.restart as the last fold recovery step. The
  controller's own tools come first; these only fill what neither controller has.
- iPhone Duo on EAS: both controllers' folds are verified (Argent by its tree,
  agent-device by its screenshot) and recovered by restarting SpringBoard and
  folding again, once; never by stopping the session's tool-server.
- agent-device opens the app before folding on EAS, as it does locally.
- agent-device pinned to 0.21.29-nightly.20261018.74 for callstack/agent-device#3776,
  which stops @expo/ui sliders on a regular-width display from wedging its runner.
- A session EAS fails to start is replaced, up to five times within 30 minutes,
  with a growing wait, only when the failure is provably EAS's.
- doctor warns about Argent's single appearance only for Android EAS sessions.
2026-10-17 20:44:41 -07:00
Brent Vatne 46ff99aad1 Viewer: held keys repeat and glide, arrows enter the map, held modifiers show their keys, T and ⇧T (#10) 2026-10-17 18:52:55 -07:00
Brent Vatne a75cbf84ea Link the deployment when the example alias still serves an earlier one
The example report reuses one token and the same fixture reports, so an alias
still pointing at the previous deployment serves the token page and the same
report bytes; only its reports.json (an older Viewer's listing) differed, and
that mismatch failed the job as PREVIEW_VERIFY_FAILED. The alias now counts as
moved only once it serves this deployment's listing and reports, and an alias
that never does is PREVIEW_ALIAS_BEHIND: the verified deployment is linked
with a warning. The deployment's own origin is still checked strictly.

The README's "Run it yourself" now opens with the two ways to use Explore and
which one needs an Expo account.
2026-10-17 16:26:40 -07:00
Brent Vatne 6c10477c5b Run explore:binary on a local simulator, and bring the README up to date (#8)
explore:binary --simulator local explores a build on a simulator or emulator
it creates on this Mac, with no EAS login: the same devices capture uses,
recorded for the startup sweep, driven by agent-device with its own daemon or
by a local Argent tool-server. The report opens in the local Viewer; --upload
hosts it.

The README now covers local repack and local simulators, a single-path quick
start that reviews a change or explores a build, targets, and requirements
split by what each path needs. The explore-binary, explore-setup and
readme-refresh skills follow.
2026-10-17 15:53:21 -07:00
241 changed files with 20535 additions and 3316 deletions

No files matched your search

+7 -1
View File
@@ -61,7 +61,13 @@ jobs:
env:
# npm cannot generate provenance for a private repository.
NPM_CONFIG_PROVENANCE: "false"
run: npm publish --access public
VERSION: ${{ steps.version.outputs.version }}
shell: bash
# A prerelease (0.20.0-next.0) goes to the next tag, so latest keeps the last release.
run: |
set -euo pipefail
if [[ "$VERSION" == *-* ]]; then tag=next; else tag=latest; fi
npm publish --access public --tag "$tag"
- name: Tag the release
if: steps.version.outputs.publish == 'true'
env:
+13 -3
View File
@@ -67,19 +67,29 @@ jobs:
shell: bash
env:
COMMENT_ID: ${{ steps.comment.outputs.id }}
# How long an alias can keep serving the previous push: an EAS Hosting edge that served it keeps it for about
# five minutes (FIXTURE_ALIAS_WAIT_MS in src/report/preview-deploy.ts).
ALIAS_CATCH_UP: 5 minutes
run: |
set -euo pipefail
jq -r --arg marker "$MARKER" --arg sha "${SHA:0:7}" --arg run "$RUN_URL" '
jq -r --arg marker "$MARKER" --arg sha "${SHA:0:7}" --arg run "$RUN_URL" --arg catchup "$ALIAS_CATCH_UP" '
.previewUrl as $url
| ([.reports[] | select(.platform == "android")] + [.reports[] | select(.platform != "android")]) as $reports
| ($reports[0].source // "") as $source
| [$marker,
"**Viewer preview** of \($sha)\(if $source != "" then " · reports from \($source)" else "" end)",
"",
"| Platform | Changes | Open |",
"| Report | Changes | Open |",
"| --- | --- | --- |",
($reports[] | "| \(if .platform == "ios" then "iOS" else "Android" end)\(if .device then " · \(.device)" else "" end) | \(.changes // "–") | [Map](\($url)?platform=\(.platform)) · [Screens](\($url)?platform=\(.platform)#/screens) · [Compare](\($url)?platform=\(.platform)&view=compare) |"),
# One row per target, named by its label; a target named after its platform keeps ?platform= links.
($reports[]
| (.target // .platform) as $id
| "\(if $id == "ios" or $id == "android" then "platform" else "target" end)=\($id)" as $query
| ((.label // (if .platform == "ios" then "iOS" else "Android" end)) | gsub("\\|"; "\\\\|")) as $name
| "| \($name)\(if .device then " · \(.device)" else "" end) | \(.changes // "–") | [Map](\($url)?\($query)) · [Screens](\($url)?\($query)#/screens) · [Compare](\($url)?\($query)&view=compare) |"),
"",
# The alias still served the previous push when the deploy stopped waiting for it: say so, and link the new deployment.
(if .aliasBehind then "> **Going live:** this link can take up to \($catchup) to show \($sha); until then it shows the previous push, or nothing on a new pull request. [This push'"'"'s deployment](\(.deploymentUrl)/p/\($url | split("/p/")[1])) is live now.\n" else empty end),
"The link stays the same on every push. [Deploy run](\($run))"]
| join("\n")' "$RUNNER_TEMP/preview.json" > "$RUNNER_TEMP/comment.md"
if [[ -n "$COMMENT_ID" ]]; then
+82 -9
View File
@@ -14,7 +14,7 @@ Source of truth: `templates/explore-review.yml`, `templates/explore-baseline.yml
| Pipeline | `capture:review` → `runCiReview` | The same `runCiReview`, called directly |
| EAS credentials | `EXPO_TOKEN` secret | Local `eas login` session or `EXPO_TOKEN` |
| Model credentials (agent mode) | `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` | Claude Code or Codex subscription login; API keys are removed from the subprocess |
| Base cache | A baseline workflow artifact for the exact base SHA, when one exists | None; both sides are captured |
| Base cache | A baseline workflow artifact for the exact base SHA, when one exists, or the baseline run still capturing it, waited for alongside head | None; both sides are captured |
| Output | Report artifact + sticky PR comment with **Open** / **zip** | `runs/<id>/artifacts/report.zip` + `PR-comment.md`, opened with `ci:local:view` |
## 1. Installing Explore in an app repository
@@ -49,16 +49,16 @@ Each workflow starts with a `targets` job: it installs the pinned CLI and runs `
| | checkout `app-base`, `app-head` | Checks out the exact base and head SHAs with full history and without persisted credentials. |
| | Verify deterministic source planning | `source:verify` builds the capture plan from source alone in two fresh processes and compares the bytes. Runs before any build or simulator work and needs no credentials. |
| | Install app dependencies | `npm ci --ignore-scripts` in both checkouts (`APP_INSTALL_DIR` for monorepos). |
| | Find / Download optional exact-base evidence | `ci:find-baseline` looks for a successful `explore-baseline-<base sha>-<target>` artifact (target `ios` also accepts the older unsuffixed `explore-baseline-<base sha>`) from the baseline workflow on the default branch and downloads it as `EXPLORE_CAPTURE_CACHE`. Missing evidence is not an error. |
| | Capture review | `capture:review` (§3). `EXPO_TOKEN`, the provider key and the app's `{{secret:NAME}}` values reach only this step. The sticky PR comment is updated while it runs. It writes `preview=<hosting>` to the step outputs, which decides whether the deploy job runs. |
| | Find / Download optional exact-base evidence | `ci:find-baseline` looks for a finished run's `explore-baseline-<base sha>-<target>` artifact (a run another target's baseline failed in counts: a baseline job uploads its artifact only after its own steps passed) (target `ios` also accepts the older unsuffixed `explore-baseline-<base sha>`) from the baseline workflow on the default branch and downloads it as `EXPLORE_CAPTURE_CACHE`. With none, it names a baseline run of the base commit still queued or capturing (`pending`), held to the same checks: the baseline workflow, the default branch of this repository, a `push` or `workflow_dispatch`, exactly the base SHA; nothing from the PR head selects it. Missing evidence and a failed lookup are not errors. |
| | Capture review | `capture:review` (§3). With `EXPLORE_BASELINE_RUN` (the find step's `pending`) it waits for that run's baseline of this target (§3.1). `EXPO_TOKEN`, the provider key and the app's `{{secret:NAME}}` values reach only this step. The sticky PR comment is updated while it runs. It writes `preview=<hosting>` to the step outputs, which decides whether the deploy job runs. |
| | Stop capture sessions and check the checkouts | `ci:after-capture`, whenever the PR was resolved, success or not: stops the EAS Simulator sessions named in `head-session` / `base-session`, nothing else, then fails with `CHECKOUT_MODIFIED` if capture modified an app checkout. |
| | Upload captured review / diagnostics | `explore-capture-<run>-<attempt>-<target>` feeds the report job. On failure `explore-diagnostics-<run>-<attempt>-<target>` keeps plans, exports and agent diagnostics for 7 days. |
| report (matrix: target) | Install Explore, download the capture | Runs without Expo or model credentials. |
| report (matrix: target) | Download the capture, install Explore | Runs without Expo or model credentials, after every capture job, whether or not each succeeded (`!cancelled()`). Each leg downloads its own target's capture by an exact-name pattern: a target whose capture failed uploaded none, so its leg packages and publishes nothing, and the other targets' reports still reach the comment. |
| | Package hosted review bundle | `ci:report` writes the report directory. |
| | Upload hosted review bundle | `explore-review-<run>-<attempt>-<target>`, kept 30 days. |
| | Publish current PR hosted links | `ci:publish` re-checks that the artifact belongs to this PR, then updates the sticky comment with **Open** (the hosted Viewer) and **zip**. The comment is rendered from one hidden record per target (`<!-- expo-explore-record:<target>:… -->`, plus one history line for earlier runs); each job folds its own record into the ones the comment already carries and re-renders the whole comment, so parallel targets do not overwrite each other. Rows, report links and validation columns are labelled with the target's label (`iOS` and `Android` without `targets:`) and follow config order. |
| `deploy` (only when `report.preview` is set; after every report job) | Deploy the hosted preview | `ci:deploy --inputs --reports` downloads every target's capture and report artifacts, zips and encrypts each report, deploys it with the prebuilt preview Viewer to a dedicated EAS Hosting project (the only post-capture job holding `EXPO_TOKEN`), verifies that only `/p/<token>` answers, then updates the comment with **Explore report** and the link's expiry. |
| failure | Report failure only on the current PR | Downloads whichever diagnostics exist and posts the failure comment to this PR alone, replacing only the failed target's record when the run has several; the target is the id the failed job's diagnostics artifact name ends with, and its label comes from the record the capture job already posted. |
| `deploy` (only when `report.preview` is set; after every report job, once any of them published) | Deploy the hosted preview | `ci:deploy --inputs --reports` downloads every capture and report artifact the run has (a target whose capture or report job failed has none and is left out), writes each report's manifest and screenshots as static files under the link's token, deploys them with the prebuilt preview Viewer to a dedicated EAS Hosting project (the only post-capture job holding `EXPO_TOKEN`), verifies that only `/p/<token>` answers, then updates the comment with **Explore report**. The link works as long as the EAS Hosting deployment. |
| failure | Report failure only on the current PR | Downloads whichever diagnostics exist and posts the failure comment to this PR alone, replacing only each failed target's record when the run has several; the target is the id the failed job's diagnostics artifact name ends with (or, when download-artifact extracted a lone artifact without its directory, the `target` its `error.json` records), and its label comes from the record the capture job already posted. Once the other targets published, the comment shows their reports and this failure, with no new-run line. |
A reviewer downloads the report ZIP and drops it on the hosted Viewer; downloading needs read access to the repository.
@@ -68,7 +68,7 @@ Validate baseline source → Install Explore → Verify deterministic source pla
### 2.4 `explore-compare.yml`
The same jobs as a PR review, for any `base` and `head` without a pull request. The `targets` job resolves both refs once with `ci:compare-refs` (a tag wins over a branch of the same name; the same commit twice fails with `COMPARE_SAME_COMMIT`) and passes the comparison to the capture jobs, which run `capture:review --compare` with no GitHub requests. The optional `scope` and `expires-in` inputs override `exploration.scope` and `report.preview.expiresAfter` for that run. The report job's `ci:publish` writes each target's **Open** and **zip** links to the run summary, and the deploy job adds the hosted link there; nothing comments on a pull request.
The same jobs as a PR review, for any `base` and `head` without a pull request. The `targets` job resolves both refs once with `ci:compare-refs` (a tag wins over a branch of the same name; the same commit twice fails with `COMPARE_SAME_COMMIT`) and passes the comparison to the capture jobs, which run `capture:review --compare`; its only GitHub requests are the wait for a baseline run still capturing the base (§3.1), with the job's `actions: read` token. The optional `scope` input overrides `exploration.scope` for that run; `expires-in` is accepted and ignored. The report job's `ci:publish` writes each target's **Open** and **zip** links to the run summary, and the deploy job adds the hosted link there; nothing comments on a pull request.
## 3. Inside `capture:review` (`runCiReview`)
@@ -77,7 +77,7 @@ The same jobs as a PR review, for any `base` and `head` without a pull request.
| 1 | `plan` | `buildVerifiedPlan`: snapshot both sources, compute the two-dot diff, run `plan-worker` twice with reversed input order and compare the canonical plans byte for byte. Summarises `sourceChange` per surface into `execution-plan.json`; the PR comment shows the planned capture count and warnings. | `SOURCE_ANALYSIS_FAILED`, inconsistent plan |
| 2 | `preflight` | Checks `EXPO_TOKEN`. `prepareCiExploration`: nothing for deterministic mode; for agent mode, confirms the instructions file is committed, creates the provider session and makes one short tool-free model request. | `AGENT_API_KEY_MISSING`, `AGENT_INSTRUCTIONS_UNCOMMITTED`, `AGENT_MODEL_NOT_FOUND` |
| 3 | `capture-head` | `runCiMap(head)` (§4). Records head's runtime profile and build. | `CAPTURE_*`, `SETUP_*`, `EAS_*` |
| 4 | `capture-base` | Tries the cache first: a baseline is reused only when its recipe matches (tools, config, instructions, device, profile, observer hash, runtime profile). Only routes missing from the cache run through `runCiMap(base)`. | same |
| 4 | `capture-base` | Tries the cache first: a baseline is reused only when its recipe matches (tools, config, instructions, device, profile, observer hash, runtime profile). Only routes missing from the cache run through `runCiMap(base)`. With no cache but a baseline run still capturing the base, base starts only once the wait for it ends (§3.1). | same |
| 5 | `compare` | `assessPublicEvidence` compares pixels and accessibility for route pairs verified on both sides; `prepareCaptureReview` assigns verdicts (§6). | |
| 6 | `capture-validation` | Every selected route was captured on both sides. | `CAPTURE_INCOMPLETE` |
| 7 | `comparison-validation` | Every pair was compared visually and for accessibility. | `COMPARISON_INCOMPLETE` |
@@ -86,6 +86,17 @@ The same jobs as a PR review, for any `base` and `head` without a pull request.
A failure at any stage is recorded in `state.failure`, a partial report is written when a plan exists, and the PR comment switches to the failed state.
### 3.1 Waiting for a baseline still being captured
Right after a merge, the next pull request's review starts while the baseline workflow is still capturing the new base: in brentvatne/euxy's review run 39348956443 it had started 15 s before, and every target captured its base too (10–14 minutes per target instead of about 6, on twice the EAS Simulator sessions). A baseline job takes 6–8 minutes and head capture 5–8, so `capture:review` waits for the baseline beside head instead of before it (`src/ci/baseline-wait.ts`):
1. The wait starts after preflight, only when `EXPLORE_BASELINE_RUN` names a run, no downloaded baseline matches, base has routes to capture, and `capture.baselineWaitMinutes` (default 15, 0 never waits) is not 0. Head starts capturing at once.
2. Every 20 s it reads the run (checked again by the same trust rules as `ci:find-baseline`), this target's artifact (`explore-baseline-<sha>-<target>`) in that run, and the run's jobs, logging `waiting for main's baseline run <id> of <sha> (<target>: in progress, 3 min)` once a minute or when the state changes. Each read retries GitHub's 5xx and network failures (`retryGitHubLookup`).
3. When the artifact is listed, it is downloaded in-process (the token goes to GitHub, never to the storage URL it redirects to), unpacked into `<output>/baseline-download` within the cache's size limits, and restored as a downloaded baseline once head has its runtime profile. Another target's failed job does not matter; each target's job uploads its own artifact.
4. The wait ends without a baseline when this target's job fails, is cancelled or is skipped, the run ends without listing the artifact for a minute, the time runs out, GitHub refuses the token (`actions: read` missing) or keeps failing, or head fails. Base then captures at once, beside head when the sides capture concurrently. The wait never fails the review.
Head capture is never delayed. When the baseline arrives while head captures, base costs only its restore; when it does not, base starts the moment the wait gives up, so a failed baseline delays base only by the time it took to fail. Sessions are planned as without a wait: with a share of two or more, head keeps its half while base waits, so base can start in the rest; with one session, base follows head as before, and the wait may go on past head, up to its limit.
## 4. Capturing one side (`runCiMap`)
| # | Stage | Work |
@@ -93,11 +104,73 @@ A failure at any stage is recorded in `state.failure`, a partial report is writt
| 1 | cache | If a reuse directory exists, its bundle and map are validated and copied. With no routes left, skip straight to export. |
| 2 | `configuration` | `eas config --profile` confirms the app identity and scheme, `ios.simulator: true` and the bundle ID; `eas simulator:availability`. |
| 3 | `build` | `eas fingerprint:generate`; a finished simulator build with the same fingerprint is reused, otherwise `eas build --wait` (up to 45 minutes). |
| 4 | `instrumentation` | A hosted EAS `repack` workflow job (`eas workflow:run --wait --json` from a staged copy of the checkout) rebundles the current JavaScript with a read-only navigation observer (the same observer on both sides), and the repacked app is downloaded and installed into the session; `capture.repack: local` (the default with `capture.simulator: local`) runs `@expo/repack-app` on the runner, and its app is installed without an upload. Generated files stay outside the checkout. |
| 4 | `instrumentation` | A hosted EAS `repack` workflow job (`eas workflow:run --json` from a staged copy of the checkout, then its run read until it ends) rebundles the current JavaScript with a read-only navigation observer (the same observer on both sides), and the repacked app is downloaded and installed into the session; `capture.repack: local` (the default with `capture.simulator: local`) runs `@expo/repack-app` on the runner, and its app is installed without an upload. Generated files stay outside the checkout. Preparing the app has one 30-minute budget, a failed hosted repack starts again, and a run attempt's matrix jobs make one repack of each commit between them (§4.1). |
| 5 | `session` | `eas simulator:start`; the session ID is written to `<side>-session`; the app is installed. |
| 6 | route capture | Mode-specific (§5). `agent-device` drives the simulator and records a screenshot plus the accessibility tree. A capture is accepted only when the observer's reported destination matches the requested route. |
| 7 | `export` / `cleanup` | Validate the map, write `<side>-export/` (`bundle.json`, `execution.json`, PNGs, `capture-diagnostics/`), stop the session. |
### 4.1 The hosted repack: one budget, retries, and one repack per commit for a run's jobs
`prepareHostedRepack` (`src/capture/eas-build.ts`) runs the repack job. Three things keep it from failing a target, running long, or repeating work.
**One budget.** Preparing the app gets 30 minutes (`HOSTED_REPACK_BUDGET_MS`), counted from the moment the EAS session starts beside the repack. That covers waiting for another job's repack, following it, and every repack this job starts, including each run's wait. Before this, each run had a 30-minute wait of its own, so a share wait plus retries could prepare for over an hour. The rules:
- A repack starts only when at least 5 minutes of the budget are left.
- A run still going at the end of the budget is cancelled, and the target fails with `EAS_REPACK_FAILED` (`failure: timeout`).
- The session was asked for the capture's minutes plus a 20-minute allowance (`SESSION_PREPARATION_MINUTES`). If the app took longer than the allowance, the session is stopped and a fresh one started before the install, so the capture keeps all its minutes. This counts as one of the session's replacements.
**Explore waits on the run itself.** `workflow:run` runs without `--wait`. Explore reads the run's id from the JSON, or from the "See logs:" link if the JSON is missing, then reads `eas workflow:view <id> --json` every 20 s. It does not use `--wait` because eas-cli 25.2.0's `--wait` exits with code 13 after six failed status reads and leaves the run going, and Explore then started another one beside it. Now:
- A run whose state stays unreadable six times in a row is cancelled (`eas workflow:cancel`). So is a run that outlasts the budget, or whose capture is stopped. None of them is replaced.
- If the start times out, `defaultRunner` keeps what it printed. When that names the run, Explore waits on it. When it doesn't, Explore starts no other, because one may be running.
- `capture:review` and `capture:baseline` turn the runner's SIGINT and SIGTERM, which GitHub sends to a cancelled job, into an abort. The capture then stops its session and cancels the run it waits on. A second signal quits at once.
**A failed repack starts again.** In brentvatne/euxy#176 (review run 39506405687), EAS ended a repack's build while its "Upload artifact" step was saving the app ("Turtle build is not in progress", reported as `Failed to upload application archive.`), and the target failed with `EAS_REPACK_FAILED`. Only a run EAS reports finished can be followed by another. `hostedRepackFailure` reads the build's error, the job's errors, the failed step's logs (`eas workflow:logs <job> --json`) and the CLI's output, in this order:
| Failure | Read from | Fresh repacks |
| --- | --- | --- |
| `canceled` | a run someone cancelled | 0 |
| `configuration` | `Not logged in`, `Unauthorized`, `not authorized`, `E401`/`E403`, `401 Unauthorized`/`403 Forbidden` (a private registry's included) | 0 |
| `infrastructure` | a file of EAS's worker missing (`'/opt/eas/…'`) | 2 |
| `app` | the `Repack` / `Repack app` step failed, or Metro, Babel or Expo named the app's code or files: `Unable to resolve`, `SyntaxError`, `TransformError`, `Cannot find module`, `could not read ./…` | 0 |
| `timeout` | `EAS_BUILD_TIMEOUT_ERROR`: a rerun would run out its time again (a bundling hang, say) | 0 |
| `infrastructure` | a failed `Spin up…`, `Set up…`, `Download…` or `Upload…` step; EAS's or eas-cli's exact words (`Turtle build is not in progress`, `Failed to upload application archive`, `SERVER_ERROR`, `Failed to upload project sources`, `Failed to start the workflow with the API`, `Unexpected response from server (5xx)`, `ECONNRESET`, `ETIMEDOUT`, `ENOTFOUND`, `EAI_AGAIN`, `socket hang up`, `fetch failed`) | 2 |
| `unknown` | nothing above | 1 |
Each new start logs the failed run's link and reason, for example `the hosted EAS repack https://expo.dev/…/workflows/<id> failed (…); starting another repack (1/2)`. The session is started once, beside the first repack, and installs whichever repack succeeds.
**One repack per commit for a run attempt's matrix jobs.** Captures that run in one process already share a build in a directory (`src/capture/shared-build.ts`): the appearance passes of one side, which run in worker threads, and the targets of a local review. The first to claim a key builds and repacks, and the others wait for its `result.json` ("using the build another capture of this run is preparing for this commit"). A GitHub matrix job is a runner of its own, with no directory in common, so in #176 the four iOS jobs each logged `reusing EAS build a81d5d80 … repacking it with this commit's JavaScript` and made four identical repacks. The jobs of one run attempt meet on EAS instead:
1. `hostedRepackPeers` lists, in config order, the targets that repack the same app on EAS: one platform, EAS profile and bundle id. `capture:review`, `capture:review --compare` and `capture:baseline` pass them on.
- A capture gets a `RepackShare` when it has two or more peers and `GITHUB_RUN_ID`, `GITHUB_RUN_ATTEMPT` and `GITHUB_REPOSITORY` are set. Its scope is the run attempt, its leader is the first peer, and its rank is its place among the peers.
- A single target, a local review, `capture.repack: local`, and a native build this job made itself behave as before. Nobody else's key can match a native build this job made.
2. The build to repack is the newest finished build with the fingerprint that Explore did not repack itself (`existingBuild`, from 20 builds). If repacks carry their source's fingerprint, every job still picks the same build.
3. Once the copy is staged, the capture takes a key (`hostedRepackShareKey`) from:
- the run attempt, the platform and the native build id;
- the EAS profile and environment name;
- a snapshot of the environment's variables: `eas env:list --format long` for the project and the account, which names each variable with its value (masked when sensitive or secret) and when it was last updated;
- the Explore release;
- a digest of every staged file with its path, executable bit and content (`stagedProjectDigest`), which covers the commit's JavaScript and the generated observer.
If either variable list fails, the capture repacks without sharing. Its own repack's message ends with `[explore share <key>]`.
4. It reads the platform's 50 most recent builds (`eas build:list --limit 50 --json`, a single page) for one whose message ends with that marker, and joins any it finds, whichever job started it. A finished one is downloaded instead of repacking. A queued or running one is then read by its id (`eas build:view <id> --json`) every 30 s, so newer builds on a busy project cannot push it off the list.
5. How long a capture looks is staggered (`sharedRepackWaitMs`). The leader looks once and repacks if it finds nothing. The next target looks every 15 s for 3 minutes, and each target after it a minute longer. When the leader makes no repack (its baseline covered every route, or `max-parallel` kept it out of the first jobs), the next target's repack appears before the one after it would start another, so they all join one. In a re-run attempt, which runs only the jobs it re-runs, the stagger starts at once (1, 2, 3 minutes).
6. When the repack it follows fails, is cancelled or loses the key, the capture looks for another with the same key for 60 s, staggered the same way, before it repacks itself. That catches the leader's retry, which starts at once (#176).
7. It also repacks itself, with the retries above, when nothing appears in its wait, or when EAS fails to answer three times in a row. Looking stops while the budget still has 5 minutes for a repack of its own. A followed repack that does not finish within the budget fails the target. No capture waits on a job that cannot answer, and none waits past its budget.
Only a build whose marker is this capture's own key is used, and every input to the key is what this capture would itself have uploaded and built with. So two jobs share a repack only when their uploads, native build, profile, environment variables and Explore release are the same, in the same run attempt. The run attempt is in the scope, and a review's or a first attempt's repacks are never used by another run or attempt. That covers the baseline workflow, whose artifacts later reviews reuse, and "Re-run failed jobs" after an `EXPO_PUBLIC_*` value was fixed, which rebundles with the new value.
What the key does not protect against: every input to it is public or predictable, and the pull request's head code runs in the capture job with `EXPO_TOKEN` (`app.config.*` under `eas config` and `eas fingerprint:generate`, and the app's build code on EAS). Head code could therefore make a build whose message carries the base side's marker, and base captures of the same run attempt in other matrix jobs would install it. That is within the trust the capture already gives EAS: the same credentials could publish a simulator build with a reused fingerprint, which every capture already reuses by fingerprint (the `build` stage above), and head code in one job can already tamper with that job's own base side on the runner. The run attempt in the scope keeps such a build out of every other run, including the baseline. Requiring the build's `initiatingActor` to be this job's own credentials would not exclude head code, which runs with the same token. Explore cannot tell offline which actor EAS records for a workflow's repack build, so it does not check it.
Cost, per side and job:
- The leader makes 1 `build:list` and 2 `env:list`.
- A follower makes up to 13 `build:list` while the next target waits, plus 4 more per target after it.
- About 2 `build:view` a minute while it follows a repack: about 10 for a 5-minute repack.
- Each repack this job runs reads its run every 20 s: about 15 `workflow:view` for 5 minutes, and one `workflow:logs` if it fails.
In #176's run that is about 20 calls per side for each of the three iOS targets that would otherwise have repacked, in place of three repacks. A follower whose leader and earlier peers make no repack of this side loses its wait, 3 to 5 minutes, before it repacks itself. Native builds are not shared across jobs: a fingerprint EAS has no build for is built by every job that needs it.
## 5. Deterministic and agent modes
| | Deterministic | Agent |
+184 -110
View File
@@ -1,35 +1,39 @@
# Expo Explore
Explore maps your app: a screenshot of every screen, connected by the taps between them. On each pull request it captures the screens the PR changes, before and after, on EAS Simulator, and posts a visual review.
Explore maps your app: a screenshot of every screen, connected by the taps between them. On each pull request it captures the screens the PR changes, before and after, on EAS Simulator, and posts a visual review. On your Mac it can also run on a local simulator.
<picture>
<source media="(prefers-color-scheme: dark)" srcset="docs/screenshots/readme-hero-dark.png">
<img alt="Explore's Map of a euxy pull request, zoomed in on the new About screen, marked Added among the modified Patterns screens, with its transitions highlighted and the Inspector open on it, shown as a window on a soft stage" src="docs/screenshots/readme-hero-light.png">
<source media="(prefers-color-scheme: dark)" srcset="docs/screenshots/readme-hero-dark.svg">
<img alt="Explore's Map of a euxy pull request, animated: the screen cards appear row by row, the new /lane-editor/0 marked Added beside the removed /lane-editor and the modified /new-pattern, the transitions between them wipe in, then a click selects /lane-editor/0: a blue ring circles it, its transitions turn blue while the rest fade to hairlines, and the Inspector slides in on it, shown as a window on a soft stage" src="docs/screenshots/readme-hero-light.svg">
</picture>
**See a real report:** [euxy](https://explore-example--euxy.expo.app/p/rfcoic27ccgnrmzgijubxkmuzmptcrva?platform=ios), a MIDI sequencer, reviewed by Explore. It is redeployed with the current Viewer on every merge to main.
<picture>
<source media="(prefers-color-scheme: dark)" srcset="docs/screenshots/readme-compare-dark.svg">
<img alt="The euxy New Pattern sheet wiping from its capture before the pull request to its capture after it: the suggested pattern name and the row of pattern icons differ, and those regions are outlined on the after capture" src="docs/screenshots/readme-compare-light.svg">
</picture>
**Compare**: each changed screen before and after, with the changed regions outlined on the after.
<table>
<tr>
<td width="33%" valign="top">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="docs/screenshots/readme-compare-dark.png">
<img width="100%" alt="The Compare view with the euxy Patterns screen's before and after captures side by side: each pattern row's tempo moved out of the subtitle into its own column on the right, the changed subtitles outlined in both" src="docs/screenshots/readme-compare-light.png">
</picture>
<strong>Compare</strong>: each changed screen before and after, with the changed regions outlined.
</td>
<td width="33%" valign="top">
<td width="50%" valign="top">
<a href="docs/screenshots/readme-screens-light.png">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="docs/screenshots/readme-screens-dark.png">
<img width="100%" alt="The Screens view of a euxy review: the app/ route tree on the left under Changed 9 and All 19, and a grid of screen cards beside it in the same order, each with its route below the screenshot and a dot before it: green for the new /about, orange for modified screens such as /activity-log, /settings and /patterns, red for the removed /channel-surf" src="docs/screenshots/readme-screens-light.png">
<img width="100%" alt="The Screens view of a euxy review showing Changed 7 of All 19: the app/ route tree on the left, and beside it a grid of the changed screens in the same order, each with its route below the screenshot: a red dot for the removed /lane-editor, orange for the modified /new-pattern and /channel-surf, green for the new /lane-editor/0, and Changed in dark under /p and /patterns" src="docs/screenshots/readme-screens-light.png">
</picture>
<strong>Screens</strong>: every capture in one grid, beside the route tree, with a dot for what changed.
</a>
<strong>Screens</strong>: every capture in one grid, or only the changed ones, beside the route tree, with a dot for what changed.
</td>
<td width="33%" valign="top">
<td width="50%" valign="top">
<a href="docs/screenshots/readme-pr-comment-light.png">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="docs/screenshots/readme-pr-comment-dark.png">
<img width="100%" alt="Explore's GitHub PR comment on a euxy PR: the heading links to the iOS and Android reports, a table with one row per platform giving device, result and an Open link, the Added, Removed, Modified, Needs review and Source-affected screens as collapsed lists, fading out below them" src="docs/screenshots/readme-pr-comment-light.png">
<img width="100%" alt="Explore's GitHub PR comment on a euxy PR: the heading 2 screens changed with a Review in Explore link; a line of links to each device's report (iPhone 17, three iPhone Duo poses and Android) and to the two commits compared; then the changed screens as a list, /tempo with the text it added and the devices it is on, fading out below" src="docs/screenshots/readme-pr-comment-light.png">
</picture>
</a>
<strong>A comment on every PR</strong>: with <a href="#set-up-ci">CI</a>, the report links and the added, removed and modified screens.
</td>
</tr>
@@ -41,7 +45,9 @@ Explore maps your app: a screenshot of every screen, connected by the taps betwe
Paste this into Claude Code, Codex, Cursor, or any tool that can fetch a URL. In the Viewer, **Hand off to AI** fills it in for you:
```text
Review the visual changes in this PR. Expo Explore report: <report link>
Review the visual changes in PR #<number> of <owner>/<repo>.
Expo Explore report:
<report link>
Fetch the link first. It explains the report's JSON API. Go through the changed screens,
compare the before and after screenshots, and tell me which changes look intended and
@@ -53,52 +59,82 @@ which look like regressions. For each regression, name the source file to fix.
<details>
<summary><strong>Requirements</strong></summary>
- An iOS app, an Android app, or both. Explore is in v0; Android capture is newer than iOS.
- Expo SDK 57 or newer, with Expo Router (57 or newer) or React Navigation (6 or newer).
- An EAS build profile per platform: `ios.simulator: true` for iOS, `android.buildType: apk` for Android. Its EAS environment cannot have secret-visibility variables.
- EAS Simulator access (in preview, [join the waitlist](https://expo.dev/services/simulators)).
- Node.js 24.12 or newer, and admin access to the GitHub repository.
- To review a change: Expo SDK 57 or newer, with Expo Router (57 or newer) or React Navigation (6 or newer), and an Expo account, since Explore builds the app on EAS. An EAS build profile per platform: `ios.simulator: true` for iOS, `android.buildType: apk` (not a development build) for Android. Its EAS environment cannot have secret-visibility or file variables.
- To explore a build: a simulator build of any app, and a Claude Code or Codex login (or an API key). On a local simulator it needs no Expo account.
- Somewhere to run the app: EAS Simulator (in preview, [join the waitlist](https://expo.dev/services/simulators)), which CI uses; or a [local simulator](#simulator) on your Mac: Xcode 26 or newer for iOS (the default devices are iPhone 17 models) and, for Android, the Android SDK with `ANDROID_HOME` set (emulator, platform-tools, cmdline-tools and a system image; to review a change, also build-tools and a JDK 17).
- Node.js 24.12 or newer. For CI, admin access to the GitHub repository.
</details>
## Quick start
Review a change to your own app, on your machine. You need an Expo SDK 57+ app with Expo Router or React Navigation, Node.js 24.12+, and EAS Simulator access ([join the waitlist](https://expo.dev/services/simulators)). No API keys, no CI.
Paste this into Claude Code or Codex from your app's directory. It reviews a change to your app, before and after, or, if you point it to a build, maps every screen of that build:
From your app's directory:
```text
Show me Expo Explore working on this app, with npx @expo/explore. If I point you to a
build (.app, .tar.gz, simulator .ipa, or .apk), explore it: follow
https://unpkg.com/@expo/explore/skills/explore-binary/SKILL.md. If this is not an Expo
project, ask me for a simulator build and explore that. Otherwise review a change to
this app: follow the "Review a change on this machine" section of
https://unpkg.com/@expo/explore/skills/explore-setup/SKILL.md. If the app cannot build
for the iOS Simulator on EAS yet, set that up (an EAS project, ios.bundleIdentifier, an
"explore" profile with ios.simulator: true) and ask me before you commit it. If I have
no EAS Simulator access, pass --simulator local (it needs Xcode; exploring an .apk needs
the Android SDK instead). If I have no uncommitted changes, change some text near the
top of a screen. If the app needs a sign-in, ask me for a test account the way the skill
describes, never for a password in chat. Open the report, and if I need to log in to
something first, tell me the command.
```
1. **Make the app buildable on EAS, if it is not yet.**
You need Node.js 24.12+. Reviewing a change needs an Expo account and an Expo SDK 57+ app with Expo Router or React Navigation: it is built on EAS and captured on EAS Simulator or, with Xcode, on a simulator on your Mac. Exploring a build needs a Claude Code or Codex login, and EAS Simulator or a simulator on your Mac; on your Mac, no Expo account. No API keys, no CI.
> [!NOTE]
> EAS Simulator is in preview. [Join the waitlist](https://expo.dev/services/simulators), or pass `--simulator local` to capture on a simulator on your Mac.
<details>
<summary><strong>Run it yourself</strong></summary>
There are two ways to use Explore: review a change to your Expo app, which needs an Expo account, or explore an iOS or Android build of any app directly, which needs no Expo account when it runs on your Mac.
**Review a change**, from your app's directory:
1. Make the app buildable on EAS, if it is not yet:
```sh
npx eas-cli@latest login
npx eas-cli@latest init
```
Then set `ios.bundleIdentifier` in your app config, add a simulator profile to `eas.json`, and commit. The review builds your last commit as the "before".
Then set `ios.bundleIdentifier` in your app config, add a simulator profile to `eas.json`, and commit. The "before" is where your branch left the default branch: on the default branch, your last commit.
```json
{ "build": { "explore": { "ios": { "simulator": true } } } }
```
2. **Change a screen, then review the change.** Edit some text near the top of a screen (Explore captures the first screenful) and run:
2. Edit some text near the top of a screen (Explore captures the first screenful) and run:
```sh
npx @expo/explore capture
npx @expo/explore capture # on EAS Simulator
npx @expo/explore capture --simulator local # or on a simulator on your Mac
```
Explore builds your last commit and your uncommitted changes on EAS, captures the screens the changes can affect, before and after, on EAS Simulator, and opens the report in your browser. The first run waits for an EAS build; later runs reuse it while your native code is unchanged.
Explore builds your last commit and your uncommitted changes on EAS, captures the screens the changes can affect, before and after, and opens the report in your browser. The first run waits for an EAS build; later runs reuse it while your native code is unchanged. A [local simulator](#simulator) is one Explore creates on your Mac and removes afterwards.
Or paste this into Claude Code or Codex from your app's directory:
**Explore a build:**
```text
Show me an Expo Explore review of a change to this app. Follow the "Review a change on
this machine" section of https://unpkg.com/@expo/explore/skills/explore-setup/SKILL.md.
If the app cannot build for the iOS Simulator on EAS yet, set that up (an EAS project,
ios.bundleIdentifier, an "explore" profile with ios.simulator: true) and ask me before
you commit it. If I have no uncommitted changes, change some text near the top of a
screen. Then run npx @expo/explore capture and open the report.
```sh
npx @expo/explore explore:binary ./my-app.apk # Android APK
npx @expo/explore explore:binary ./my-app.app # iOS simulator build (.app, .tar.gz, or simulator .ipa)
npx @expo/explore explore:binary ./my-app.app --simulator local # on a simulator on your Mac
```
Explore starts an EAS Simulator, installs the build, taps through the app with your Claude Code or Codex login, and prints a link to the hosted report. With `--simulator local` it runs on a simulator or emulator it creates on your Mac, needs no Expo account, and opens the report in your browser. Its [options](#options) are below.
</details>
> [!WARNING]
> A review runs both commits' code on your machine, as you: it installs their dependencies, evaluates the app config and bundles the JavaScript. Review only commits you would run yourself, and someone else's branch only if you trust it.
Every run is kept on your machine for 30 days. `npx @expo/explore reports` lists your local reviews, explorations and uploaded reports with their commits, and reopens or hosts one without capturing again:
<picture>
@@ -115,12 +151,11 @@ npx @expo/explore capture --committed # HEAD alone
npx @expo/explore capture --base v1.3.0 --head v1.4.0 --scope all --upload
```
- It checks out both commits as temporary worktrees, installs their dependencies, captures every configured platform on EAS Simulator (`--platform` picks one), and opens the report in a local Viewer. `--upload` hosts it and prints the link; `report:upload <report.zip>` hosts one you already have.
- It checks out both commits as temporary worktrees, installs their dependencies, captures every configured platform or [target](#targets) on EAS Simulator or a [local simulator](#simulator), and opens the report in a local Viewer. `--target` or `--platform` picks some; they run at once, at most `--parallel N` at a time. `--upload` hosts it and prints the link; `report:upload <report.zip>` hosts one you already have.
- Your checkout, index and branches are never touched. Your uncommitted changes, untracked files included, are reviewed as `eas build` would upload them; `--committed`, or `cli.requireCommit: true` in `eas.json`, leaves them out.
- It runs both commits' code on your machine, as you: installing their dependencies, evaluating the app config, and bundling the JavaScript. That is what checking them out and running them would do, so review only commits you would run, and someone else's branch only if you trust it.
- Installs run without credential environment variables, which keeps tokens out of reach of a registry a commit's `.npmrc` names; that is not a sandbox. A registry that needs `${NPM_TOKEN}` in the project's `.npmrc` must be authenticated in your own `~/.npmrc`.
- Captures are cached in `~/.cache/expo-explore`, so the same base is not captured twice (`--no-cache` skips it); a capture unused for 30 days goes, and the cache stays under 2 GB. Ctrl+C stops the run's simulator sessions.
- Like a CI review, it builds on EAS (reusing a matching build) and runs simulator sessions: expect several minutes and EAS usage.
- Captures are cached in `~/.cache/expo-explore`, so the same base is not captured twice (`--no-cache` skips it); a capture unused for 30 days goes, and the cache stays under 2 GB. Ctrl+C stops the run's simulator sessions and local devices; when a run is stopped harder than that, the next local run stops what it left running.
- Like a CI review, it builds on EAS (reusing a matching build) and, unless it captures on a local simulator, runs EAS Simulator sessions: expect several minutes and EAS usage.
- Saved runs take up to 2 GB in all, the oldest going first. `reports add <report.zip>` keeps a report you downloaded, such as a PR's `explore-review` artifact.
</details>
@@ -129,44 +164,22 @@ Explore uses its defaults until you run `npx @expo/explore setup`, which writes
## Map every screen of any app
To map every screen of an app instead, with no config, paste this into Claude Code or Codex from your app's directory. It explores a build you point to, or an Expo project straight from its source:
```text
Show me Expo Explore working on this app. Follow
https://unpkg.com/@expo/explore/skills/explore-binary/SKILL.md and run Explore with
npx @expo/explore. If I point you to a build, explore that. Otherwise, if this is an
Expo project, run it from source without making a new build; if it is not, get a
simulator build of it. If the app needs a sign-in, ask me for a test account the way
the skill describes, never for a password in chat. Give me the report link, and if I
need to log in to something first, tell me the command.
```
For apps behind a sign-in, see [Sign-in and other app state](#sign-in-and-other-app-state).
<details>
<summary><strong>Run it yourself</strong></summary>
```sh
npx @expo/explore explore:binary ./my-app.apk # Android APK
npx @expo/explore explore:binary ./my-app.app # iOS simulator build (.app, .tar.gz, or simulator .ipa)
```
Explore starts an EAS Simulator, installs the build, taps through the app with your Claude Code or Codex login, and prints a link to the hosted report.
</details>
Pointed to a build, the [quick start](#quick-start) maps every screen of it, with no source and no config: one card per screen it reached, connected by the taps between them. For apps behind a sign-in, see [Sign-in and other app state](#sign-in-and-other-app-state).
<details>
<summary id="options"><strong>What it needs, and its options</strong></summary>
- An EAS login (`eas whoami`) and a `claude` or `codex` login, or `ANTHROPIC_API_KEY` / `OPENAI_API_KEY`.
- A simulator build. A device `.ipa` or an `.aab` is rejected before a session starts.
- A `claude` or `codex` login, or `ANTHROPIC_API_KEY` / `OPENAI_API_KEY`.
- EAS Simulator access and an EAS login (`eas whoami`), or `--simulator local` to run on a [local simulator](#simulator) with no EAS login.
- A simulator build. A device `.ipa`, an `.aab`, an unsigned APK, or an APK without native libraries for the emulator (x86_64 on EAS Simulator, `arm64-v8a` locally on Apple silicon; a universal APK has both) is rejected before a device starts.
- Platform, bundle id, name, and version are read from the build; you pass nothing else.
- A run stops after 15 minutes, 40 screens, or 150 actions (`--max-duration-minutes`, `--max-screens`, `--max-actions`).
- Buttons that delete, pay, sign out, send, or leave the app are never tapped. `--allow-destructive "Clear cache"` allows one by label.
- Buttons that delete, pay, sign out, send, or leave the app are never tapped. `--allow-destructive "Reset filters"` allows one by its exact label.
- `--hints "…"` tells the model what to focus on or how to sign in ([details](#sign-in-and-other-app-state)).
- The hosted link lasts 7 days (`--expires-in 30d|never`). `--no-upload` keeps the report local, and `--json` prints the result for scripts.
- For a link that stays the same from run to run, pass `--alias NAME` and set `EXPLORE_PREVIEW_TOKEN` to a fixed 32-character token.
- Read a local result with `expo-explore view --dir <folder>`, `screen:list`, or `gap:list`.
- On EAS Simulator the report is hosted, and the link lasts as long as its EAS Hosting deployment (see [report access](docs/private-review.md#hosted-previews)); `--no-upload` keeps it local. On a local simulator it opens in the local Viewer (`--no-open` or `--json` skips that; `view --dir <folder>` opens it later), and `--upload` hosts it, which needs an EAS login. `--json` prints the result for scripts.
- For a link that stays the same from run to run, pass `--alias NAME` and set `EXPLORE_PREVIEW_TOKEN` to a fixed token of 32 characters from a–z and 2–7 (`LC_ALL=C tr -dc a-z2-7 </dev/urandom | head -c 32`).
- EAS Simulator sessions and hosted reports run under an EAS project named `expo-explore` in your account, created on first use (`--owner` and `--project` pick another).
- Read a local result with `npx @expo/explore view --dir <folder>`, `screen:list`, or `gap:list`.
</details>
@@ -185,10 +198,45 @@ npx @expo/explore explore:session <session-id> --env-file path/.env.eas-simulato
</details>
<details>
<summary><strong>Take the same steps on every run and device</strong></summary>
A model picks its own path, so two runs (or an iPhone and an Android phone) can reach different screens. To get the same screens every time, write the steps down once, for example after a model run showed you the app, and pass them with `--replay`. No model login or API key is needed.
```sh
npx @expo/explore explore:binary ./my-app.app --replay steps.json
npx @expo/explore explore:session --replay steps.json # an iPhone Duo pose, or any session you set up by hand
```
```json
{
"steps": [
{ "capture": "Home" },
{ "action": "tap", "label": "Settings" },
{ "capture": "Settings" },
{ "action": "tap", "label": "Dark" },
{ "capture": "Settings", "state": "dark" },
{ "action": "tap", "label": "Learn", "platforms": ["ios"] },
{ "capture": "Learn", "platforms": ["ios"] },
{ "action": "tap", "label": "Got it", "optional": true }
]
}
```
- An action is one of `tap` or `fill` (with `label` or `identifier`), `type`, `scroll` (`direction`), `wait` (`milliseconds`), `open` (`url`), `back`, and `dismiss-keyboard`, as in the model's own tools. The same safety rules apply: destructive or app-leaving taps are refused. A step takes only the fields listed here; any other field (a typo such as `platform`) is an error.
- `{ "capture": "Title" }` photographs the current screen under that title; add `"state"` for another state of a screen already captured. A capture of a screen the plan already captured under another title is reported as looking the same, and the screen keeps its first title.
- `"platforms": ["ios"]` limits a step to those platforms, for screens one platform does not have. `"optional": true` skips an action that fails (a one-time dialog) instead of stopping; the summary lists the steps it skipped.
- A `fill` or `type` text can be a `{{secret:NAME}}`, as in `--hints`: Explore types the value of the environment variable `NAME`, and a variable that is not set fails before a device starts.
- If a required action fails, the run stops there and the report still shows what was captured, with a gap naming the step and the screens it did not reach.
- The screen and action budgets still apply. A plan with more captures or actions than they allow is refused before a device starts, as is one with no steps for the platform; raise `--max-screens` and `--max-actions` for a long plan. If a budget still runs out (each screen the plan reaches also counts), the gap names the planned screens not reached.
- `--provider`, `--model` and `--hints` only steer a model, so they cannot be combined with `--replay`.
</details>
<details>
<summary><strong>What a run without source shows</strong></summary>
One card per screen the run reached, laid out by how many taps it took, with the tapped control on each edge. There is no before and after; that needs [CI](#set-up-ci).
One card per screen the run reached, laid out by how many taps it took, with the tapped control on each edge. There is no before and after; that needs the app's source: [`capture`](#quick-start) on your machine, or [CI](#set-up-ci).
![The Map of an exploration report for the euxy app: one card per screen the run reached, laid out by taps from the launch screen, each edge labelled with the control that was tapped](docs/screenshots/readme-exploration-light.png)
@@ -200,7 +248,7 @@ Explore runs in your repository's GitHub Actions in two ways, both from the same
### Review on every PR
Each pull request gets a comment linking its report: the screens it added, removed or changed, before and after, on each platform you capture. Paste this into Claude Code or Codex from your app's repository:
Each pull request gets a comment linking its report: the screens it added, removed or changed, before and after, on each platform or device ([target](#targets)) you capture. Paste this into Claude Code or Codex from your app's repository:
```text
Set up this app for Expo Explore CI reviews. Follow
@@ -220,15 +268,14 @@ By default reviewers download the report from GitHub. With hosted previews, the
report:
preview:
hosting: eas # default: off
expiresAfter: 30d # default 30d, max 365d, or never
# project: acme-explore # default <repository name>-explore, created on first run
```
- Run `expo-explore setup --workflows` after enabling it to add the `deploy` job.
- The generated workflows already have the `deploy` job; it runs once `hosting: eas` is set.
- A [comparison of two commits](#compare-any-two-commits) links its hosted report from the run's summary instead of a PR comment. `capture --upload` on your machine deploys to the same project with your `eas login`.
- `expo-explore report:upload <report.zip>...` hosts reports already on your machine the same way: a review's reports (one per platform, from `capture --output` or a downloaded `explore-review` artifact) or an `explore:binary` report.
- `expo-explore report:upload <report.zip>...` hosts reports already on your machine the same way: a review's reports (one per platform or target, from `capture --output` or a downloaded `explore-review` artifact) or an `explore:binary` report.
- `EXPO_TOKEN` must be able to create the EAS project, so use a [robot user's](https://docs.expo.dev/accounts/programmatic-access/) token.
- The report is encrypted, and only a request carrying the link's token can open it. Treat the link like a shared file link; see [report access](docs/private-review.md#hosted-previews).
- Every file of the report sits under the link's 160-bit token, so only someone with the link can open it. Treat the link like a shared file link. It works as long as its EAS Hosting deployment does: EAS deletes one without an alias after 30 days on the Free plan and 90 on paid plans, and you can delete it sooner (`eas deploy:delete`) or on your own schedule. Explore sets no expiry of its own; see [report access](docs/private-review.md#hosted-previews).
</details>
@@ -238,11 +285,11 @@ report:
The prompt above walks through these steps with you. To do them yourself:
1. Run `npx @expo/explore setup` in your app's directory (in a monorepo, add `--project apps/mobile`). It writes `.expo-explore/config.yaml` and the workflows.
2. Run `npx @expo/explore doctor` and fix each problem it reports, until it exits 0. `doctor --capture` captures the committed app on EAS once, to check every screen before your first PR.
2. Run `npx @expo/explore doctor` and fix each problem it reports, until it exits 0. `doctor --capture` captures the committed app once, to check every screen before your first PR (with several targets, pass `--target`).
3. Set the repository secrets `doctor` lists:
- `EXPO_TOKEN`: create it in [Expo access tokens](https://expo.dev/settings/access-tokens). For [hosted previews](#hosted-previews), use a [robot user](https://docs.expo.dev/accounts/programmatic-access/) that can create projects.
- A secret for each `{{secret:NAME}}` in your setup steps. Run `npx @expo/explore setup --workflows` after you add one.
4. Commit, merge to the default branch, and run **Expo Explore baseline** from GitHub Actions.
4. Commit and merge to the default branch. **Expo Explore baseline** runs on that push (or run it from GitHub Actions).
5. Open a PR. **Expo Explore review** runs and posts a comment with the report. To see a whole review of your branch first, run `npx @expo/explore capture` on your machine.
A failed run shows an error code and a next action in the PR comment; see [CI troubleshooting](docs/quickstart-ci.md#diagnose-a-failed-run).
@@ -264,7 +311,7 @@ Both modes use the same comparison rules and report. You can change modes later.
**To use agent mode:**
1. Create an [Anthropic](https://platform.claude.com/settings/keys) or [OpenAI](https://platform.openai.com/api-keys) API key.
2. Set `mode: agent` and an `agent.model` your account can use.
2. Set `mode: agent`, `agent.provider` (`anthropic` or `openai`), an `agent.model` your account can use, and `agent.instructions`: the path of a file in the repository that tells the agent how to use the app.
3. Store the key as the repository secret `ANTHROPIC_API_KEY` or `OPENAI_API_KEY`. A key that is not workspace-scoped also needs the `ANTHROPIC_WORKSPACE_ID` variable.
4. Export the key and run `expo-explore doctor --check-model`.
@@ -292,7 +339,7 @@ platforms:
easProfile: explore-android # android.buildType: apk in eas.json
```
- The workflows run one capture job per platform, and the PR gets one comment with a section per platform.
- The workflows run one capture job per platform, and the PR gets one comment that links each platform's report and lists the changed screens of both.
- A failed platform fails the check; the other platform's report still publishes.
- An Android-only app sets `app.platform: android` with no `platforms` map.
@@ -311,17 +358,20 @@ targets:
android: { platform: android, device: pixel_7, bundleId: com.example.myapp, easProfile: explore-android }
```
- A target takes the flat `app` and `device` fields it does not set; a target of another platform than `app.platform` sets its own `bundleId` and `easProfile`. `label` names it in reports (by default its device and pose).
- Ids are lowercase letters, digits and dashes. A config with `platforms` has one target per platform, named `ios` and `android`.
- The workflows run one capture job per target, the PR comment has a section per target, and each report links its sibling targets' reports.
- `capture --target duo-open` reviews one target on this machine; `--platform ios` the iOS ones.
- A target takes the `app` fields it does not set, and `device.name` when it is of `app.platform`; a target of another platform sets its own `bundleId` and `easProfile` (an Android one's device defaults to `pixel_7`). `pose` is `closed`, `half-open` or `open`. `label` names it in reports (by default its device and pose).
- Ids are up to 32 lowercase letters, digits and dashes, start with a letter, and are not `base`, `head` or `report`. A config names at most 8 targets. A config with `platforms` has one target per platform, named `ios` and `android`.
- The workflows run one capture job per target. The PR comment links each target's report and says which targets a change is on, and each report links its sibling targets' reports.
- A review runs at most `capture.maxSessions` EAS Simulator sessions at once across its targets (default 5, a typical plan's limit); see the [config reference](docs/quickstart-ci.md#config-reference).
- A target can also set its own `controller` and `simulator` over `capture`'s, such as Android on agent-device beside iOS targets on Argent. `--controller` and `--simulator` override every target's for one run.
- `capture --target duo-open` reviews one target on this machine; `--platform ios` the iOS ones. On your machine a review's targets run at once (at most `--parallel N`), and targets of one platform and EAS profile share each commit's build; in CI each target is its own job. A target that fails outright costs only its own report.
- A device controller that stops answering on one screen costs that route, up to three times per capture: Explore restarts the app, goes on with the next route, and records what happened in `capture-diagnostics/controller-recoveries.jsonl`.
</details>
<details>
<summary id="controller"><strong>Device controller: agent-device or Argent</strong></summary>
Explore drives EAS Simulator with [agent-device](https://github.com/callstackincubator/agent-device) by default. To use [Argent](https://github.com/software-mansion/argent) instead, set it in the config or pass it for one run:
Explore drives the simulator with [agent-device](https://github.com/callstackincubator/agent-device) by default. To use [Argent](https://github.com/software-mansion/argent) instead, set it in the config or pass it for one run:
```yaml
capture:
@@ -335,16 +385,16 @@ npx @expo/explore explore:binary ./app.apk --controller argent
Explore pins each controller's version and starts the EAS session with it. Reports and baselines record the controller, and a review compares captures made with the same one.
Neither controller is installed with Explore. The first run that needs one downloads its pinned npm release (agent-device 1.4 MB, Argent 27 MB), checks it against the integrity hash Explore pins, and keeps it in `~/.cache/expo-explore/controllers` (`EXPLORE_CONTROLLER_CACHE` moves it). Behind a private registry, set `EXPLORE_NPM_REGISTRY` to a mirror of it.
Neither controller is installed with Explore. The first run that needs one downloads its pinned npm release (agent-device 1.6 MB, Argent 27 MB), checks it against the integrity hash Explore pins, and keeps it in `~/.cache/expo-explore/controllers` (`EXPLORE_CONTROLLER_CACHE` moves it). Behind a private registry, set `EXPLORE_NPM_REGISTRY` to a mirror of it.
Argent cannot switch the simulator's appearance yet ([software-mansion/argent#1509](https://github.com/software-mansion/argent/pull/1509)), so an Argent capture on EAS Simulator records one appearance: the one the app forces, else light. Explore warns when it skips one. On a [local simulator](#simulator) the host switches it, so both are captured.
Argent cannot switch the simulator's appearance yet ([software-mansion/argent#1509](https://github.com/software-mansion/argent/pull/1509)), so the host switches it: on an iOS EAS Simulator session (`eas simulator:control appearance.set`) and on a [local simulator](#simulator), every appearance the config asks for is captured. An Argent capture on an Android EAS Simulator session records one appearance: the one the app forces, else light. Explore warns when it skips one.
</details>
<details>
<summary id="simulator"><strong>Local simulator instead of EAS Simulator</strong></summary>
To capture on your Mac instead of EAS Simulator, set it in the config or pass it for one run:
To capture on your Mac instead of EAS Simulator, pass it for one run, or set it in a config you use only on your Mac:
```yaml
capture:
@@ -355,16 +405,22 @@ capture:
npx @expo/explore capture --simulator local
```
CI runs on GitHub's Ubuntu runners and captures on EAS Simulator, so keep `simulator: local` out of the config you commit.
For each commit, Explore starts a device of its own and removes it afterwards; head and base each get one:
- **iOS:** a simulator of the configured device with Xcode's newest runtime that runs it. `DEVELOPER_DIR` picks another Xcode, such as one that adds iPhone Duo. Explore approves the app's URL scheme on it, so links open without iOS asking "Open in …?", and checks a foldable's pose by the shape of its lit display.
- **Android:** an emulator of the configured device profile (such as `pixel_7`), from the newest Google APIs system image installed for this machine (`arm64-v8a` on Apple silicon), in a private AVD folder. The APK must ship native libraries for that ABI; a universal build does.
- **Android:** an emulator of the configured device profile (such as `pixel_7`), from the newest Google APIs system image installed for this machine (`arm64-v8a` on Apple silicon), in a private AVD folder. The APK must ship native libraries for that ABI; a universal build does. It needs the SDK's emulator, platform-tools and cmdline-tools, with `ANDROID_HOME` set.
Builds still run on EAS. The app is repacked with the route observer on your machine as well, so it is never uploaded; on Android that needs the SDK's build-tools and a JDK. `capture.repack: eas` (or `--repack eas`) keeps the repack in a hosted EAS job. `doctor --simulator local` checks that Xcode and the Android SDK can start each device, and that Android can be repacked locally.
`explore:binary --simulator local` starts one the same way for the build it explores, needs no EAS login, and opens the report in the local Viewer (`--upload` hosts it).
With `--controller argent`, Explore installs Argent's tool-server on first use (with its dependencies, about 165 MB, in the controller cache) and starts one per device. The host switches the appearance, so a local Argent capture records light and dark. Reports record where they were captured, and a baseline is reused only by a review captured the same way.
For a review, builds still run on EAS. The app is repacked with the route observer on your machine as well, so it is never uploaded; on Android that needs the SDK's build-tools and a JDK. `capture.repack: eas` (or `--repack eas`) keeps the repack in a hosted EAS job. `doctor --simulator local` checks that Xcode and the Android SDK can start each device, and that Android can be repacked locally.
A review's targets run at once, the targets of a platform sharing each commit's build; `--parallel N` runs at most N at a time. A device controller that stops answering on one screen costs that route: Explore restarts the app, goes on with the next route, and records what happened in `capture-diagnostics/controller-recoveries.jsonl`. A target that fails outright costs its own report; the others still open.
With `--controller argent`, Explore installs Argent's tool-server on first use with `npm ci` from a lockfile Explore ships, so every dependency is pinned (about 160 MB, in the controller cache), and starts one per device. The host switches the appearance, so a local Argent capture records every appearance the config asks for. Reports record where they were captured, and a baseline is reused only by a review captured the same way.
A foldable folds before the app first opens (`pose` on a [target](#targets)); with agent-device on a local simulator it folds just after, since agent-device folds only inside a session. With Argent, an opened iPhone Duo is captured upright, in landscape. After a fold, Argent's reader can stay on the panel that went dark, so Explore checks that the lit panel is read. If it isn't, Explore restarts the tool-server on a local simulator. On EAS Simulator, where the tool-server is the session's, it resets Argent's device helpers, then restarts SpringBoard and folds again. On EAS Simulator, an opened iPhone Duo usually needs that SpringBoard restart, which adds about a minute.
Each capture records what it starts on your Mac (simulators, emulators, controller daemons, tool-servers) and removes it when it ends. If a review or a local `explore:binary` run is stopped hard, the next local one stops what it left running; nothing Explore did not record is touched.
</details>
@@ -391,11 +447,11 @@ give me the report link.
There are a few ways to snapshot a release:
- **On demand:** in your repository's Actions tab, open **Expo Explore compare**, choose **Run workflow**, and set `base` to the last release's tag (`v1.3.0`), `head` to the new tag or release candidate (`v1.4.0-rc.1`), `scope` to `all` and `expires-in` to `never`. The run's summary links the report.
- **On demand:** in your repository's Actions tab, open **Expo Explore compare**, choose **Run workflow**, and set `base` to the last release's tag (`v1.3.0`), `head` to the new tag or release candidate (`v1.4.0-rc.1`), and `scope` to `all`. The run's summary links the report.
- **Automatically, on every tag:** add the `release-review.yml` in [The comparison workflow](#compare) below. It compares each tag, release candidates included, with the last stable tag, so a release candidate is reviewed against what people have now.
- **On your machine, without CI:** `npx @expo/explore capture --base v1.3.0 --head v1.4.0 --scope all --upload`, with your `eas login`. `npx @expo/explore reports` lists the runs you have kept.
`scope: all` captures every screen, not only those whose source changed, so the report shows the whole release with what changed since the last one marked; `expires-in: never` keeps its link.
`scope: all` captures every screen, not only those whose source changed, so the report shows the whole release with what changed since the last one marked.
<details>
<summary id="compare"><strong>The comparison workflow</strong></summary>
@@ -428,16 +484,15 @@ jobs:
base: '${{ needs.previous.outputs.tag }}'
head: '${{ github.ref_name }}'
scope: all # every screen, not only those whose source changed
expires-in: never # keep the hosted link (with report.preview.hosting: eas)
secrets: inherit
```
- Pick whatever `base` the event calls for: `main` for a release branch, yesterday's tag for a nightly, the last stable tag for a release candidate.
- `scope` and `expires-in` are optional and override `exploration.scope` and `report.preview.expiresAfter` for that run. Without them a comparison captures what a PR review would (`scope: changes` captures only screens whose source changed), and the hosted link expires like a PR's.
- The run summary links each platform's report (**Open** and **zip**, like a PR comment) and, with hosting, the hosted report.
- `scope` is optional and overrides `exploration.scope` for that run. Without it a comparison captures what a PR review would (`scope: changes` captures only screens whose source changed). The hosted link lasts as long as its deployment, like a PR's; `expires-in` from earlier versions of this workflow is accepted and ignored.
- The run summary links each target's report (**Open** and **zip**, like a PR comment) and, with hosting, the hosted report.
- `setup --workflows` installs it from 0.19.0; `npx @expo/explore@latest setup --workflows` adds it to an existing setup. The caller grants `contents: read` and `actions: read`; `secrets: inherit` passes `EXPO_TOKEN` and any app-test secrets.
- Call it once per workflow run. A called workflow runs inside its caller's run, so a second call in the same run would reuse the first one's artifact names; use separate workflows instead.
- When the baseline workflow captured `base` in the last 30 days, that capture is reused and only `head` is captured. The report artifact is kept 30 days; only a hosted link with `expires-in: never` lasts longer.
- When the baseline workflow captured `base` in the last 30 days, with the same config and Explore version, that capture is reused and only `head` is captured. When it is still capturing `base` (right after a merge), the review waits for it while `head` captures, up to `capture.baselineWaitMinutes` (15 by default). The report artifact is kept 30 days; a hosted link lasts as long as its EAS Hosting deployment (an aliased one until the alias moves).
</details>
@@ -465,7 +520,7 @@ If the sign-in can't be typed (one-time codes, email links, Sign in with Apple o
| Needs data to show a screen | A test account that has it, and a hint saying where ("open the first project") |
| Asks for a permission | Explore denies it. Allow it by hand on a session, then run `explore:session` |
| Has a paywall | An account that already has the entitlement; purchase buttons are never tapped |
| Has a safe button that looks destructive | `--allow-destructive "Clear cache"` |
| Has a safe button that looks destructive | `--allow-destructive "Reset filters"` |
| Is large | [Raise the budgets](#options) (15 minutes, 40 screens, 150 actions by default) and use `--hints` to set priorities |
When a run stops early, `gap:list --dir <output>` says what blocked it.
@@ -491,7 +546,7 @@ The same link answers differently depending on who asks. It already carries the
- Anything that does not ask for HTML (`curl`, `fetch`, agents' URL tools) gets a Markdown index of the report, also at `/llms.txt`.
- The index links to a read-only JSON API under `/api/v1`: `/screens?status=modified,accessibility-only,added,removed`, `/screens/<id>`, `/transitions`, and `/raw` for the report file.
- Reports without source also expose `/journeys`, `/gaps`, and `/frontier`.
- Multi-platform reports take `?platform=ios|android`. Expired links answer `410`.
- A link to several targets takes `?target=<id>` and serves the first without it; `/reports.json` lists them (`?platform=ios|android` from older links still works).
```sh
curl -s "$LINK/api/v1/screens?status=modified" | jq '.items[] | {route, status, review}'
@@ -502,11 +557,28 @@ curl -s "$LINK/api/v1/screens?status=modified" | jq '.items[] | {route, status,
<details>
<summary><strong>How it works</strong></summary>
1. The review workflow runs on open, non-draft PRs from the repository whose author has write access. Bots such as Devin go under `ci.trustedAuthors` in `.expo-explore/config.yaml`.
2. Explore reads the diff and your router source to pick the affected screens.
3. It builds both revisions and captures each screen on EAS Simulator, reusing a baseline of the base commit when one exists.
4. It compares the captures. A screen with a source change is **Modified** if its captures differ.
5. It uploads the report to GitHub Actions and posts a PR comment.
```mermaid
sequenceDiagram
autonumber
participant PR as Pull request
participant GA as GitHub Actions
participant B as EAS Build
participant S as EAS Simulator
PR->>GA: opened or updated
GA->>GA: read the diff and router source, pick the affected screens
GA->>B: build base and head
Note over B: reuses a build with the same native fingerprint
B-->>GA: simulator builds
GA->>S: capture each affected screen, in each appearance
Note over S: the base reuses its baseline when one exists
S-->>GA: screenshots
GA->>GA: compare the captures, outline the changed regions
GA->>PR: upload the report, comment with its links
```
- The review workflow runs on open, non-draft PRs from the repository whose author has write access. Bots such as Devin go under `ci.trustedAuthors` in `.expo-explore/config.yaml`.
- A screen with a source change is **Modified** if its captures differ.
- If EAS fails to start a session, Explore starts a new one, up to five times within 30 minutes. On your machine, a local simulator can stand in for EAS Simulator.
The same review runs for any two commits without a PR ([`explore-compare.yml`](#compare-any-two-commits), linking the report from the run summary) and on your machine (`capture`, [above](#quick-start)). The workflows' logic lives in `expo-explore ci:*` commands, so a fix reaches CI with the `@expo/explore` version the workflows pin.
@@ -517,13 +589,15 @@ Reports and baselines are kept for 30 days and diagnostics for 7. Hosted deploym
<details>
<summary><strong>Known limitations</strong></summary>
- Android capture has had less use than iOS. The first Android device is `pixel_7`.
- Android capture has had less use than iOS. The default Android device is `pixel_7`.
- PRs from forks are not reviewed.
- Each route gets one capture per appearance (light and dark with `capture.appearance: auto`). Other states, such as modals, are planned.
- Each route gets one capture per appearance (`capture.appearance: auto`, the default, captures light and dark when the app's `userInterfaceStyle` is `automatic`). Other states, such as modals, are planned.
- The launch is captured once per appearance and compared like any screen. The comment says how long the splash stayed as it first appeared, when it first changed (a native launch screen handing over to an app-drawn one, an animation starting on it), and when the first screen settled. `capture.launch: video` adds a recording; a launch too fast to observe is reported as "no splash observed".
- On Android the splash usually shows only in the recording, which needs `ffmpeg`. The generated workflows install it.
- Hosted acceptance of agent mode with an API key is incomplete.
- With `capture.controller: argent` on EAS Simulator, each capture records one appearance (see [device controller](#controller)).
- With `capture.controller: argent` on an Android EAS Simulator session, each capture records one appearance (see [device controller](#controller)).
- A local simulator runs only on your machine; the CI workflows capture on EAS Simulator.
- `explore:session` explores EAS Simulator sessions only.
</details>
@@ -548,7 +622,7 @@ npm test
- `npm test` rebuilds the CLI and the Viewer first, because the node tests run the compiled `dist/`. After one `npm run build`, `npm run test:node` skips the rebuild.
- The agent runtimes and `eas-cli` are optional dependencies that only capture needs. Do not install with `--omit=optional`, which also drops esbuild's platform package.
- To release, bump `version` in `package.json` and merge to main. The [publish workflow](.github/workflows/publish.yml) publishes any version npm does not have yet.
- To release, bump `version` in `package.json` and merge to main. The [publish workflow](.github/workflows/publish.yml) publishes any version npm does not have yet. A prerelease version such as `0.20.0-next.0` goes to the `next` tag, so `latest` stays on the last release.
```sh
npx playwright install chromium
+7 -3
View File
@@ -157,6 +157,10 @@ runs:
EXPO_TOKEN: ${{ inputs.expo_token }}
run: |
set -euo pipefail
if [[ ! -e "$OUTPUT/head-session" && ! -L "$OUTPUT/head-session" && ! -e "$OUTPUT/base-session" && ! -L "$OUTPUT/base-session" ]]; then exit 0; fi
node "$TOOLS/scripts/cleanup-ci-sessions.mjs" --eas "$TOOLS/node_modules/.bin/eas" \
--session-file "$OUTPUT/head-session" --session-file "$OUTPUT/base-session"
# A side that captures an appearance in a session of its own records it as <side>-session-<scheme>.
files=()
for name in head-session base-session head-session-light head-session-dark base-session-light base-session-dark; do
if [[ -e "$OUTPUT/$name" || -L "$OUTPUT/$name" ]]; then files+=(--session-file "$OUTPUT/$name"); fi
done
if (( ${#files[@]} == 0 )); then exit 0; fi
node "$TOOLS/scripts/cleanup-ci-sessions.mjs" --eas "$TOOLS/node_modules/.bin/eas" "${files[@]}"
+8 -8
View File
@@ -12,22 +12,22 @@ The Viewer reads the ZIP supplied by the reviewer; it has no hosted artifact dow
## Hosted previews
Hosted previews are off by default. When `report.preview.hosting: eas` is set, the review workflow's `deploy` job uploads each run's reports (one per captured platform) to an [EAS Hosting](https://docs.expo.dev/eas/hosting/introduction/) deployment in your Expo account, and the comment shows an **Explore report** link that opens it directly, in place of the **Open** drop page. The GitHub artifact and **zip** stay as they are.
Hosted previews are off by default. When `report.preview.hosting: eas` is set, the review workflow's `deploy` job uploads each run's reports (one per captured platform or target) to an [EAS Hosting](https://docs.expo.dev/eas/hosting/introduction/) deployment in your Expo account, and the comment shows an **Explore report** link that opens it directly, in place of the **Open** drop page. The GitHub artifact and **zip** stay as they are.
**What a reviewer needs.** Only the link. It looks like `https://acme-explore--k7f3z0d1q8.expo.app/p/f6r2…` and carries a 160-bit token generated for that run. It opens the Viewer and the report for one PR revision; it does not show the repository, other runs, or anything else in the Expo account.
**Several platforms.** One deployment holds every platform's report. The link opens the primary platform's report (`?platform=ios` or `?platform=android` selects one); the Viewer's platform menu lists the others from `/p/<token>/reports.json`, which names each platform, its device, and its change count and nothing else. Every report and the listing sit behind the same token and expiry.
**Several platforms or targets.** One deployment holds every target's report (a config without `targets` has one per platform). The link opens the first one (`?target=<id>` selects another, and `?platform=ios` or `?platform=android` from older links still works); the Viewer's menu lists the others from `/p/<token>/reports.json`, which names each target, its platform, label, device, and change count and nothing else. Every report and the listing sit behind the same token.
**What is protected, and how.**
- *The report bytes.* `ci:deploy` zips the report and encrypts it (AES-256-GCM) before upload. The key is stored only in that deployment's server environment; it never appears in a URL, a log line, or a comment. The files on the CDN are ciphertext.
- *Access to the decrypted report.* A server route in the deployment checks the token and the expiry, fetches the ciphertext from its own deployment, decrypts it, and streams it to the browser. Without the exact token every path under the deployment answers `404`, including the Viewer itself; `ci:deploy` verifies this before it publishes the link.
- *Expiry.* The expiry (`report.preview.expiresAfter`, 30 days by default, `never` to disable) is sealed into the deployment when it is created and enforced by the route, not by the browser. After it passes, the exact token still opens the Viewer for a browser, but the report routes, the Markdown index, and the agent API answer `410` and the Viewer says the link expired on that date; every other path still answers `404`. To revoke a link earlier, delete the deployment in the EAS Hosting dashboard; the CDN stops serving it within seconds and the route within a few minutes.
- *Deletion.* Nothing is deleted automatically. Each review of a pull request deploys a new preview and the comment links the newest one (recorded in a hidden `<!-- expo-explore-deployments:… -->` line under the comment marker); earlier deployments and the deployments of closed pull requests stay until someone deletes them in the EAS Hosting dashboard or with `eas deploy:delete`, so a link keeps opening for as long as its expiry allows.
- *The token.* Every path of a preview carries the link's 160-bit token: the page at `/p/<token>`, each target's manifest at `/p/<token>/targets/<target>.json`, and every screenshot at `/p/<token>/assets/<sha256>.<ext>`. Nobody can find a path without the token. With any other token, every path under the deployment answers `404`, including the Viewer itself; `ci:deploy` verifies this before it publishes the link.
- *The report's files.* `ci:deploy` uploads the report as plain static files of the deployment, under the token. They are not encrypted: the token in their paths is what keeps them private, the same way it keeps the link private. EAS Hosting serves them directly, with its own default headers (the browser may cache a screenshot for an hour), and it cannot add `noindex` to them, so the deployment's `robots.txt` keeps crawlers away from them instead. The deployment's server route serves the page, the manifest the Viewer asks for, the target listing and the agent API, with `private, no-store` and `noindex`.
- *How long a link works.* A link works as long as its EAS Hosting deployment exists. Explore sets no expiry of its own. EAS Hosting deletes a deployment after 30 days on the Free plan and 90 days on paid plans, **unless an alias points at it**: an aliased deployment (such as `--alias` uploads of `explore:binary`) is kept until the alias moves or is deleted. To end a link sooner, delete the deployment in the EAS Hosting dashboard or with `eas deploy:delete` (delete its alias first if it has one); the CDN stops serving its files within about five minutes. To expire links on your own schedule, delete deployments from a scheduled job. Deleting a deployment does not reach into a browser that already opened the link: what that browser cached stays until it ages out. `report.preview.expiresAfter` and `--expires-in` from earlier versions are accepted and ignored, with a warning.
- *Deletion.* Explore deletes nothing. Each review of a pull request deploys a new preview and the comment links the newest one (recorded in a hidden `<!-- expo-explore-deployments:… -->` line under the comment marker); earlier deployments and the deployments of closed pull requests stay until EAS Hosting deletes them as above, or someone deletes them in the dashboard or with `eas deploy:delete`.
- *The Viewer itself.* The Viewer's JavaScript is public and the same for every deployment; it holds no report data.
- *Agent access.* The same link also answers as Markdown and as a JSON API (`/llms.txt`, `/api/v1/...`, see [the agent guide](../README.md#agents)) so a coding agent can read the report without a download. Those routes are served by the same server route from the same decrypted bytes, behind the same token, expiry, `private, no-store`, and `noindex`; they add no second credential and no new place where the report is stored.
- *Agent access.* The same link also answers as Markdown and as a JSON API (`/llms.txt`, `/api/v1/...`, see [the agent guide](../README.md#agents)) so a coding agent can read the report without a download. Those routes read the same static files, behind the same token, with `private, no-store` and `noindex`; they add no second credential and no new place where the report is stored.
**What is not protected.** The link is a bearer credential. Anyone who receives it, by forwarding the comment, a notification email, or a chat message, can open the report until it expires. Members of the Expo account who can inspect the preview project's deployments can read the key. Expo's request logs record URL paths, so they record the token; they do not record the report. Keep the default drop flow when a leaked link would be a problem.
**What is not protected.** The link is a bearer credential. Anyone who receives it, by forwarding the comment, a notification email, or a chat message, can open the report until the deployment is deleted. A screenshot's URL contains the token, so it is the link too: anyone who copies a screenshot's address can open the whole report. Members of the Expo account who can inspect the preview project's deployments can read its files. Expo's request logs record URL paths, so they record the token. Keep the default drop flow when a leaked link would be a problem.
**Comparison with the default.** The default flow uses GitHub's own permissions: a reviewer must be signed in to GitHub with read access to the repository to download the report. Hosted previews trade that check for one-click opening. Choose them when the review audience is the people who already see the PR.
+4 -1
View File
@@ -8,7 +8,7 @@ The same workflow supports Expo Router and React Navigation applications on iOS.
Expo Router inventory comes from route files and layouts. React Navigation inventory comes from statically readable JSX or object screen registrations and navigator hierarchies. Linking determines entry paths independently of route discovery. Unsupported computed declarations and unresolved paths remain diagnostics; analysis does not execute the app to guess their values.
`changes` starts with the exact two-dot Git diff and dependency impact, then adds, independently on each revision, the routes one navigation step before or after an affected route. Layout or screen guards do not count as steps. Forwarding routes have no screenshot of their own, so their redirect chains are followed to the first screen. `all` selects the full inventory. Source context does not imply a visible modification.
`changes` starts with the exact two-dot Git diff and dependency impact (a JavaScript or TypeScript file whose two revisions parse to the same syntax tree, differing only in comments or in whitespace the compiler and JSX transform drop, reaches no screen; a comment a compiler reads, such as `@jsx` or `#__PURE__`, still counts), then adds, independently on each revision, the routes one navigation step before or after an affected route. Layout or screen guards do not count as steps. Forwarding routes have no screenshot of their own, so their redirect chains are followed to the first screen. `all` selects the full inventory. Source context does not imply a visible modification.
Before app build or simulator allocation, CI regenerates the canonical plan in fresh processes and compares bytes. A mismatch fails with diagnostic plans and their diff. Run identities, timestamps, absolute checkout paths, screenshots, and cache hits are excluded from planning data. Tests also vary enumeration order and checkout location.
@@ -103,6 +103,9 @@ exploration:
capture:
maxMinutes: 30
maxActions: 500
# maxSessions: 5 # the most EAS Simulator sessions one review runs at once, across all its targets and sides (default: 5, a typical plan's concurrent simulator limit). Each target's job gets floor(maxSessions / targets) sessions, at least one, and sessions left over go to foldable targets (pose): one target gets 5 (head and base at once, light and dark in their own sessions when that pays), two get 2 each, five get 1 each (head, then base). Below the number of targets, only that many target jobs run at once (the workflow's max-parallel, and a local review's targets), one session each. A side spreads its light and dark passes over separate sessions only when that saves it at least a minute, so the ceiling is a limit, not a target: a 3-route target stays on two sessions. euxy's four targets: 8 is two per target, 16 captures each appearance in its own session (measured: an iPhone side 6:12 → 4:26, one repack shared by its sessions, ~3.5 session-minutes more per side).
# On a starter plan (one simulator per platform), set maxSessions to the plan's limit, e.g. 2 for an iOS and an Android target.
# baselineWaitMinutes: 15 # a review that starts while the default branch's baseline run of its base commit is still capturing (right after a merge) waits up to this long for that target's baseline instead of capturing the base too (default 15; 0 never waits). Head captures meanwhile; base waits only where it would start, and captures itself if the baseline job fails or is cancelled, or the time runs out.
repack: eas # eas (hosted workflow job; the default on EAS Simulator) | local (repack-app on the runner; the default with simulator: local)
appearance: auto # auto (follow the app's userInterfaceStyle: automatic captures light and dark) | dark | light | [dark, light]
launch: screenshot # screenshot (relaunch the app per appearance and keep its splash and first settled frame, default) | video (also keep a launch recording in the report) | off
+21 -8
View File
@@ -105,10 +105,13 @@ device:
exploration:
scope: changes # changes | all
capture:
# controller: agent-device # agent-device (default) | argent (one appearance per capture on EAS Simulator until Argent can switch it; --controller overrides it for one run)
# controller: agent-device # agent-device (default) | argent (one appearance per capture on Android EAS Simulator until Argent can switch it; --controller overrides it for one run)
# simulator: eas # eas (EAS Simulator, default) | local (a simulator or emulator Explore starts on this machine with Xcode or the Android SDK; --simulator overrides it for one run)
maxMinutes: 30
maxActions: 500
# maxSessions: 5 # the most EAS Simulator sessions one review runs at once, across all its targets and sides (default: 5, a typical plan's concurrent simulator limit). Each target's job gets floor(maxSessions / targets) sessions, at least one, and sessions left over go to foldable targets (pose): one target gets 5 (head and base at once, light and dark in their own sessions when that pays), two get 2 each, five get 1 each (head, then base). Below the number of targets, only that many target jobs run at once (the workflow's max-parallel, and a local review's targets), one session each. A side spreads its light and dark passes over separate sessions only when that saves it at least a minute, so the ceiling is a limit, not a target: a 3-route target stays on two sessions. euxy's four targets: 8 is two per target, 16 captures each appearance in its own session (measured: an iPhone side 6:12 → 4:26, one repack shared by its sessions, ~3.5 session-minutes more per side).
# On a starter plan (one simulator per platform), set maxSessions to the plan's limit, e.g. 2 for an iOS and an Android target.
# baselineWaitMinutes: 15 # a review that starts while the default branch's baseline run of its base commit is still capturing (right after a merge) waits up to this long for that target's baseline instead of capturing the base too (default 15; 0 never waits). Head captures meanwhile; base waits only where it would start, and captures itself if the baseline job fails or is cancelled, or the time runs out.
repack: eas # eas (hosted EAS workflow job, the default on EAS Simulator) | local (repack-app on the runner, the default with simulator: local; --repack overrides it for one run)
appearance: auto # auto (follow the app's userInterfaceStyle: automatic captures light and dark) | dark | light | [dark, light]
launch: screenshot # screenshot (relaunch the app per appearance and keep its splash and first settled frame, default) | video (also keep a launch recording in the report) | off
@@ -130,7 +133,7 @@ platforms:
easProfile: explore-android
```
Commands that capture or publish (`source:verify`, `capture:review`, `capture:baseline`, `ci:report`, `ci:publish`) take `--platform ios|android`; with one configured platform it is optional, with two it is required (`PLATFORM_REQUIRED`). `expo-explore config:platforms` prints the configured list as JSON, which the workflows use as their matrix. Artifacts carry the platform (`explore-capture-<run>-<attempt>-android`, `explore-baseline-<sha>-ios`); an iOS job still reuses an older unsuffixed baseline. The PR comment has one section per platform, and a hosted preview serves both reports from one link.
Commands that capture or publish (`source:verify`, `capture:review`, `capture:baseline`, `ci:report`, `ci:publish`) take `--platform ios|android`; with one configured platform it is optional, with two it is required (`PLATFORM_REQUIRED`). `expo-explore config:platforms` prints the configured list as JSON, which the workflows use as their matrix. Artifacts carry the platform (`explore-capture-<run>-<attempt>-android`, `explore-baseline-<sha>-ios`); an iOS job still reuses an older unsuffixed baseline. The PR comment is one list of changed screens that says which platforms each is on, and a hosted preview serves both reports from one link.
### Several devices (targets)
@@ -144,7 +147,7 @@ targets:
android: { platform: android, device: pixel_7, bundleId: com.example.myapp, easProfile: explore-android }
```
A target's id (lowercase letters, digits and dashes; not `base`, `head` or `report`) takes the place of the platform everywhere a capture is named: `--target duo-open`, `explore-capture-<run>-<attempt>-duo-open`, `explore-baseline-<sha>-duo-open`. `expo-explore config:targets` prints the list the workflows use as their matrix. Fields a target leaves out come from `app` and `device`; a target of another platform than `app.platform` sets its own `bundleId` and `easProfile`. A config with `platforms` has one target per platform, named `ios` and `android`, so its names do not change. `--platform` still selects a target when only one has that platform.
A target's id (up to 32 lowercase letters, digits and dashes, starting with a letter; not `base`, `head`, `report` or `install-…`; at most 8 per config) takes the place of the platform everywhere a capture is named: `--target duo-open`, `explore-capture-<run>-<attempt>-duo-open`, `explore-baseline-<sha>-duo-open`. `expo-explore config:targets` prints the list the workflows use as their matrix. Fields a target leaves out come from `app` and `device`; a target of another platform than `app.platform` sets its own `bundleId` and `easProfile`. A config with `platforms` has one target per platform, named `ios` and `android`, so its names do not change. `--platform` still selects a target when only one has that platform.
Use the actual bundle identifier, scheme, profile, and available simulator device name. `scheme` is optional when static app configuration supplies it. `navigation.entry` resolves ambiguous entry modules; it does not list screens.
@@ -215,13 +218,13 @@ exploration:
- /c/[channel]
```
An excluded route stays in the map and in its neighbors' navigation context, but neither revision captures it, and a change to it gets no before/after comparison. A name that matches no route at either compared revision fails the plan with `EXCLUDE_ROUTE_UNKNOWN`; `doctor` warns about it first. Prefer `parameters` or `setup` when a route does show a screen.
An excluded route stays in its neighbors' navigation context (a screen that links to it is still captured when it changes), but neither revision captures it, it is left off the report (the Map, Screens, search and the screen counts), and a change to it gets no before/after comparison. A name that matches no route at either compared revision fails the plan with `EXCLUDE_ROUTE_UNKNOWN`; `doctor` warns about it first. Prefer `parameters` or `setup` when a route does show a screen.
Capture handles iOS's confirmation to open a link in the expected app. After a form submits a password, iOS may offer to save or update it depending on the simulator's AutoFill state; capture declines that prompt with Not Now in both modes, never stores the credential, and records each decision in `capture-diagnostics/system-prompts.jsonl`. Do not script that tap in setup. Include required system-alert handling in setup or app instructions; Explore does not automatically grant application permissions.
### Stable captures (optional)
Screenshots are compared pixel by pixel, so anything that changes between runs shows up as a Screen variant or noise: the current time and relative dates, animations and loading indicators that can be caught mid-frame, screens that move on by themselves, and permission prompts that depend on earlier runs. If your app has these, consider making the capture profile's build fix them, for example with an environment variable in that EAS profile that the app reads. What to fix depends on your app; your coding agent can find candidates in the source, and each change is an app change for you to approve. The [example app's `explore` profile](../test/ci/apps/wallet/README.md#explore-build-profile) is one example.
Screenshots are compared pixel by pixel, so anything that changes between runs shows up as a Screen variant or noise: the current time and relative dates, animations and loading indicators that can be caught mid-frame, screens that move on by themselves, and permission prompts that depend on earlier runs. When both builds capture light and dark, text that differs between each build's own two captures (a clock, a relative date, a random name) is reported as runtime state, never as the change that makes a screen Modified; text that only follows the appearance, or a word beside it that changed, is still a change. If your app has these, consider making the capture profile's build fix them, for example with an environment variable in that EAS profile that the app reads. What to fix depends on your app; your coding agent can find candidates in the source, and each change is an app change for you to approve. The [example app's `explore` profile](../test/ci/apps/wallet/README.md#explore-build-profile) is one example.
### Agent usage and reuse
@@ -286,7 +289,7 @@ The provider selects the runtime that owns the navigation loop:
| `anthropic` | Claude Code through `@anthropic-ai/claude-agent-sdk` 0.3.285 and its platform-native package | `ANTHROPIC_API_KEY` |
| `openai` | Codex app-server from `@openai/codex` 0.159.2 and its platform-native package | `OPENAI_API_KEY` |
Both runtimes use Explore's device tools to operate the existing EAS Simulator through official `agent-device` 0.21.28. The runtime and native binaries come from the pinned tools installation. CI does not use the development-only `node_modules/.bin/claude` executable or an interactive subscription login.
Both runtimes use Explore's device tools to operate the existing EAS Simulator through official `agent-device` 0.21.29-nightly.20261018.74. The runtime and native binaries come from the pinned tools installation. CI does not use the development-only `node_modules/.bin/claude` executable or an interactive subscription login.
Create and commit the nonempty English prose file named by `agent.instructions`. The shared setup skill authors it from source and asks for missing test data or environment inputs. Describe how to reach the app's screens, necessary prerequisites, the representative state to use on both revisions, readiness signals, and actions to avoid. This file guides the runtime agent; it is not a list of mandatory YAML steps or a second screen inventory.
@@ -360,7 +363,17 @@ Create an Expo access token for an account with access to the app's EAS project.
Capture uses the automatic GitHub token with `pull-requests: write`, so it can keep the sticky comment current while the run works; publication rewrites that comment from a separate job. `GITHUB_TOKEN_MISSING` means the workflow environment is missing the automatic token mapping.
The sticky comment is one comment for the whole run: a header with the base and head commits, one row per captured platform (device, result, **Open** / **zip** / **run** links), every changed screen listed once with a link per platform, and a collapsed validation table with one column per platform. Each platform's job updates only its own hidden record in the comment, so iOS and Android jobs never overwrite each other. When a new run starts (a push, a rerun), the earlier run moves to a collapsed **Previous runs** list that keeps its head commit, result, report and run links; the last five runs are kept. A run that finishes after the PR moved on is still published, under previous runs, instead of failing. The failure job posts through the same renderer (`ci:failure`), so a crash before the pipeline's own comment still lands in that platform's row.
The sticky comment is one comment for the whole run: a heading with what changed (**17 screens changed · Review in Explore**), a line naming every target with a link to its report plus the compared commits and when the comment was last updated, one line for any target that is still working, failed, or finished with captures missing, then the changed screens as a short list (the first ten, each linking its Before/After in the report, with a link to the rest). Capture problems, when there are any, sit in a collapsed **Capture issues** block; the footer links the workflow run and the config. Each target's job updates only its own hidden record in the comment, so iOS and Android jobs never overwrite each other. When a new run starts (a push, or a rerun's new attempt), the comment keeps the last review under a **New run in progress** line until the new run has a ready target of its own; a target of the new run that fails is named on that line, which reads **New run failed** once nothing is still working. A push also moves the earlier run to a collapsed **Previous runs** list that keeps its head commit, result, report and run links; the last five runs are kept. A run that finishes after the PR moved on is still published, under previous runs, instead of failing. The failure job posts through the same renderer (`ci:failure`), so a crash before the pipeline's own comment still lands in that target's line.
The comment can also draw a map of the change: the added, removed and visually changed screens, the screens one step before them, and the transitions among them, as a collapsed Mermaid flowchart that GitHub draws in place. It is off by default; turn it on with:
```yaml
report:
comment:
map: true # default false
```
The map leaves out screens whose source changed but whose captures match, and it is skipped when more than 30 routes changed or when the comment would be over GitHub's size limit.
PR review links always use [explore-preview.expo.app](https://explore-preview.expo.app/); no viewer-origin variable is needed or accepted by the CI publisher. Reports stay in GitHub Actions: the comment's **zip** link fetches the `explore-review-<run>-<attempt>.zip` artifact (GitHub asks for sign-in and repository read access), and the page that **Open** opens takes that ZIP by drag-and-drop or **Choose file**. Nothing is enabled per repository on Expo's side. See [report access](private-review.md).
@@ -379,7 +392,7 @@ The review workflow resolves the current PR through GitHub, checks out its exact
| Review packaging | The bundle contains metadata and assets, not a standalone HTML page |
| Publication | The artifact is uploaded before current-PR identity is checked again and the **Open** and **zip** links are posted |
The baseline workflow runs on default-branch pushes or manual default-branch dispatch. It captures one complete commit without needing a PR. Reports and baselines are retained for 30 days; intermediate evidence and diagnostics for 7 days.
The baseline workflow runs on default-branch pushes or manual default-branch dispatch. It captures one complete commit without needing a PR. A review that starts while the baseline of its base commit is still being captured, as the next pull request's review does right after a merge, waits for it (`capture.baselineWaitMinutes`, 15 minutes by default) while head captures, instead of capturing the base too. Reports and baselines are retained for 30 days; intermediate evidence and diagnostics for 7 days.
`exploration.scope: changes` includes source-affected routes and the routes one navigation step before or after them, so a reviewer sees how each change is reached and where it leads. The context is not transitive: in most apps tabs, guards and Home connect every screen, so all ancestors and descendants would be the whole app. A layout or screen guard that sends every signed-out screen to sign-in is not a step. Forwarding routes have no screenshot of their own, so their redirect chains are followed to the first screen. Unresolved navigation still selects every route. `all` captures the full inventory.
Binary file not shown.

Before

Width:  |  Height:  |  Size: 451 KiB

+1
View File
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1000 600" width="1000" height="600" role="img" aria-label="The /new-pattern screen of euxy wiping from its capture before the pull request to its capture after it, with the changed regions outlined"><style>.after{animation:wipe 7s cubic-bezier(.65,0,.35,1) infinite}@keyframes wipe{0.00%{clip-path:inset(0 100% 0 0)}14.29%{clip-path:inset(0 100% 0 0)}37.14%{clip-path:inset(0 0 0 0)}82.86%{clip-path:inset(0 0 0 0)}97.14%{clip-path:inset(0 100% 0 0)}100.00%{clip-path:inset(0 100% 0 0)}}.divider{opacity:0;animation:divider 7s cubic-bezier(.65,0,.35,1) infinite}@keyframes divider{0.00%{transform:translateX(0);opacity:0}11.43%{transform:translateX(0);opacity:1}14.29%{transform:translateX(0);opacity:1}37.14%{transform:translateX(248px);opacity:1}41.43%{transform:translateX(248px);opacity:0}82.86%{transform:translateX(248px);opacity:0}85.71%{transform:translateX(248px);opacity:1}97.14%{transform:translateX(0);opacity:1}100.00%{transform:translateX(0);opacity:0}}.regions{animation:regions 7s ease-in-out infinite}@keyframes regions{0.00%{opacity:0}38.57%{opacity:0}44.29%{opacity:1}51.43%{opacity:.3}58.57%{opacity:1}81.43%{opacity:1}85.71%{opacity:0}100.00%{opacity:0}}.before-label{opacity:.3;animation:before-label 7s linear infinite}.after-label{animation:after-label 7s linear infinite}@keyframes before-label{0.00%{opacity:1}25.71%{opacity:1}28.57%{opacity:.3}88.57%{opacity:.3}91.43%{opacity:1}100.00%{opacity:1}}@keyframes after-label{0.00%{opacity:.3}25.71%{opacity:.3}28.57%{opacity:1}88.57%{opacity:1}91.43%{opacity:.3}100.00%{opacity:.3}}@media (prefers-reduced-motion: reduce){*{animation:none!important}}</style><defs><clipPath id="stage"><rect width="1000" height="600" rx="16"/></clipPath><clipPath id="phone"><rect x="376" y="30" width="248" height="540" rx="40"/></clipPath><filter id="shadow" x="-20%" y="-20%" width="140%" height="140%"><feDropShadow dx="0" dy="8" stdDeviation="14" flood-color="rgba(1,4,9,0.6)"/></filter></defs><g clip-path="url(#stage)"><rect width="1000" height="600" fill="#151B23"/><rect x="376" y="30" width="248" height="540" rx="40" fill="#151B23" filter="url(#shadow)"/><g clip-path="url(#phone)"><image href="data:image/jpeg;base64,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 truncated

After

Width:  |  Height:  |  Size: 103 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 452 KiB

@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1000 600" width="1000" height="600" role="img" aria-label="The /new-pattern screen of euxy wiping from its capture before the pull request to its capture after it, with the changed regions outlined"><style>.after{animation:wipe 7s cubic-bezier(.65,0,.35,1) infinite}@keyframes wipe{0.00%{clip-path:inset(0 100% 0 0)}14.29%{clip-path:inset(0 100% 0 0)}37.14%{clip-path:inset(0 0 0 0)}82.86%{clip-path:inset(0 0 0 0)}97.14%{clip-path:inset(0 100% 0 0)}100.00%{clip-path:inset(0 100% 0 0)}}.divider{opacity:0;animation:divider 7s cubic-bezier(.65,0,.35,1) infinite}@keyframes divider{0.00%{transform:translateX(0);opacity:0}11.43%{transform:translateX(0);opacity:1}14.29%{transform:translateX(0);opacity:1}37.14%{transform:translateX(248px);opacity:1}41.43%{transform:translateX(248px);opacity:0}82.86%{transform:translateX(248px);opacity:0}85.71%{transform:translateX(248px);opacity:1}97.14%{transform:translateX(0);opacity:1}100.00%{transform:translateX(0);opacity:0}}.regions{animation:regions 7s ease-in-out infinite}@keyframes regions{0.00%{opacity:0}38.57%{opacity:0}44.29%{opacity:1}51.43%{opacity:.3}58.57%{opacity:1}81.43%{opacity:1}85.71%{opacity:0}100.00%{opacity:0}}.before-label{opacity:.3;animation:before-label 7s linear infinite}.after-label{animation:after-label 7s linear infinite}@keyframes before-label{0.00%{opacity:1}25.71%{opacity:1}28.57%{opacity:.3}88.57%{opacity:.3}91.43%{opacity:1}100.00%{opacity:1}}@keyframes after-label{0.00%{opacity:.3}25.71%{opacity:.3}28.57%{opacity:1}88.57%{opacity:1}91.43%{opacity:.3}100.00%{opacity:.3}}@media (prefers-reduced-motion: reduce){*{animation:none!important}}</style><defs><clipPath id="stage"><rect width="1000" height="600" rx="16"/></clipPath><clipPath id="phone"><rect x="376" y="30" width="248" height="540" rx="40"/></clipPath><filter id="shadow" x="-20%" y="-20%" width="140%" height="140%"><feDropShadow dx="0" dy="8" stdDeviation="14" flood-color="rgba(17,17,17,0.14)"/></filter></defs><g clip-path="url(#stage)"><rect width="1000" height="600" fill="#F6F8FA"/><rect x="376" y="30" width="248" height="540" rx="40" fill="#F6F8FA" filter="url(#shadow)"/><g clip-path="url(#phone)"><image href="data:image/jpeg;base64,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 truncated

After

Width:  |  Height:  |  Size: 113 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 847 KiB

+1
View File
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1600 900" width="1600" height="900" role="img" aria-label="Explore's Map of a euxy pull request: the screen cards appear row by row, the transitions between them wipe in, then a click selects /lane-editor/0 and the Inspector opens on it"><style>.c0{transform-origin:0 0;animation:c0 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c0{0.00%{opacity:0;transform:scale(.92)}4.44%{opacity:0;transform:scale(.92)}8.19%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c1{transform-origin:0 0;animation:c1 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c1{0.00%{opacity:0;transform:scale(.92)}7.86%{opacity:0;transform:scale(.92)}11.61%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c2{transform-origin:0 0;animation:c2 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c2{0.00%{opacity:0;transform:scale(.92)}7.19%{opacity:0;transform:scale(.92)}10.94%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c3{transform-origin:0 0;animation:c3 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c3{0.00%{opacity:0;transform:scale(.92)}9.27%{opacity:0;transform:scale(.92)}13.02%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c4{transform-origin:0 0;animation:c4 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c4{0.00%{opacity:0;transform:scale(.92)}4.77%{opacity:0;transform:scale(.92)}8.52%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c5{transform-origin:0 0;animation:c5 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c5{0.00%{opacity:0;transform:scale(.92)}8.19%{opacity:0;transform:scale(.92)}11.94%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c6{transform-origin:0 0;animation:c6 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c6{0.00%{opacity:0;transform:scale(.92)}9.60%{opacity:0;transform:scale(.92)}13.35%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c7{transform-origin:0 0;animation:c7 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c7{0.00%{opacity:0;transform:scale(.92)}9.94%{opacity:0;transform:scale(.92)}13.69%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c8{transform-origin:0 0;animation:c8 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c8{0.00%{opacity:0;transform:scale(.92)}10.27%{opacity:0;transform:scale(.92)}14.02%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c9{transform-origin:0 0;animation:c9 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c9{0.00%{opacity:0;transform:scale(.92)}10.61%{opacity:0;transform:scale(.92)}14.36%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c10{transform-origin:0 0;animation:c10 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c10{0.00%{opacity:0;transform:scale(.92)}6.85%{opacity:0;transform:scale(.92)}10.60%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c11{transform-origin:0 0;animation:c11 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c11{0.00%{opacity:0;transform:scale(.92)}10.94%{opacity:0;transform:scale(.92)}14.69%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c12{transform-origin:0 0;animation:c12 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c12{0.00%{opacity:0;transform:scale(.92)}11.27%{opacity:0;transform:scale(.92)}15.02%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c13{transform-origin:0 0;animation:c13 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c13{0.00%{opacity:0;transform:scale(.92)}11.61%{opacity:0;transform:scale(.92)}15.36%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c14{transform-origin:0 0;animation:c14 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c14{0.00%{opacity:0;transform:scale(.92)}7.52%{opacity:0;transform:scale(.92)}11.27%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.full{animation:full 12s cubic-bezier(.45,0,.25,1) infinite}@keyframes full{0.00%{clip-path:inset(0 0 100% 0)}13.75%{clip-path:inset(0 0 100% 0)}24.58%{clip-path:inset(0 0 0 0)}100.00%{clip-path:inset(0 0 0 0)}}.ptr{opacity:0;animation:ptr 12s cubic-bezier(.4,0,.2,1) infinite}@keyframes ptr{0.00%{opacity:0;transform:translate(853.5px,575.2px)}28.75%{opacity:0;transform:translate(853.5px,575.2px)}30.42%{opacity:1;transform:translate(853.5px,575.2px)}38.75%{opacity:1;transform:translate(593.5px,355.2px)}39.58%{opacity:1;transform:translate(593.5px,355.2px) scale(.82)}40.83%{opacity:1;transform:translate(593.5px,355.2px)}48.75%{opacity:0;transform:translate(593.5px,355.2px)}100.00%{opacity:0;transform:translate(593.5px,355.2px)}}.sel{animation:sel 12s linear infinite}@keyframes sel{0.00%{opacity:0}39.58%{opacity:0}40.42%{opacity:1}100.00%{opacity:1}}.insp{animation:insp 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes insp{0.00%{opacity:0;transform:translateX(48px)}40.00%{opacity:0;transform:translateX(48px)}45.00%{opacity:1;transform:translateX(0)}100.00%{opacity:1;transform:translateX(0)}}.cover{oLine truncated

After

Width:  |  Height:  |  Size: 1.2 MiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 830 KiB

+1
View File
@@ -0,0 +1 @@
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1600 900" width="1600" height="900" role="img" aria-label="Explore's Map of a euxy pull request: the screen cards appear row by row, the transitions between them wipe in, then a click selects /lane-editor/0 and the Inspector opens on it"><style>.c0{transform-origin:0 0;animation:c0 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c0{0.00%{opacity:0;transform:scale(.92)}4.44%{opacity:0;transform:scale(.92)}8.19%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c1{transform-origin:0 0;animation:c1 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c1{0.00%{opacity:0;transform:scale(.92)}7.86%{opacity:0;transform:scale(.92)}11.61%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c2{transform-origin:0 0;animation:c2 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c2{0.00%{opacity:0;transform:scale(.92)}7.19%{opacity:0;transform:scale(.92)}10.94%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c3{transform-origin:0 0;animation:c3 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c3{0.00%{opacity:0;transform:scale(.92)}9.27%{opacity:0;transform:scale(.92)}13.02%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c4{transform-origin:0 0;animation:c4 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c4{0.00%{opacity:0;transform:scale(.92)}4.77%{opacity:0;transform:scale(.92)}8.52%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c5{transform-origin:0 0;animation:c5 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c5{0.00%{opacity:0;transform:scale(.92)}8.19%{opacity:0;transform:scale(.92)}11.94%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c6{transform-origin:0 0;animation:c6 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c6{0.00%{opacity:0;transform:scale(.92)}9.60%{opacity:0;transform:scale(.92)}13.35%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c7{transform-origin:0 0;animation:c7 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c7{0.00%{opacity:0;transform:scale(.92)}9.94%{opacity:0;transform:scale(.92)}13.69%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c8{transform-origin:0 0;animation:c8 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c8{0.00%{opacity:0;transform:scale(.92)}10.27%{opacity:0;transform:scale(.92)}14.02%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c9{transform-origin:0 0;animation:c9 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c9{0.00%{opacity:0;transform:scale(.92)}10.61%{opacity:0;transform:scale(.92)}14.36%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c10{transform-origin:0 0;animation:c10 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c10{0.00%{opacity:0;transform:scale(.92)}6.85%{opacity:0;transform:scale(.92)}10.60%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c11{transform-origin:0 0;animation:c11 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c11{0.00%{opacity:0;transform:scale(.92)}10.94%{opacity:0;transform:scale(.92)}14.69%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c12{transform-origin:0 0;animation:c12 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c12{0.00%{opacity:0;transform:scale(.92)}11.27%{opacity:0;transform:scale(.92)}15.02%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c13{transform-origin:0 0;animation:c13 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c13{0.00%{opacity:0;transform:scale(.92)}11.61%{opacity:0;transform:scale(.92)}15.36%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.c14{transform-origin:0 0;animation:c14 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes c14{0.00%{opacity:0;transform:scale(.92)}7.52%{opacity:0;transform:scale(.92)}11.27%{opacity:1;transform:scale(1)}100.00%{opacity:1;transform:scale(1)}}.full{animation:full 12s cubic-bezier(.45,0,.25,1) infinite}@keyframes full{0.00%{clip-path:inset(0 0 100% 0)}13.75%{clip-path:inset(0 0 100% 0)}24.58%{clip-path:inset(0 0 0 0)}100.00%{clip-path:inset(0 0 0 0)}}.ptr{opacity:0;animation:ptr 12s cubic-bezier(.4,0,.2,1) infinite}@keyframes ptr{0.00%{opacity:0;transform:translate(853.5px,575.2px)}28.75%{opacity:0;transform:translate(853.5px,575.2px)}30.42%{opacity:1;transform:translate(853.5px,575.2px)}38.75%{opacity:1;transform:translate(593.5px,355.2px)}39.58%{opacity:1;transform:translate(593.5px,355.2px) scale(.82)}40.83%{opacity:1;transform:translate(593.5px,355.2px)}48.75%{opacity:0;transform:translate(593.5px,355.2px)}100.00%{opacity:0;transform:translate(593.5px,355.2px)}}.sel{animation:sel 12s linear infinite}@keyframes sel{0.00%{opacity:0}39.58%{opacity:0}40.42%{opacity:1}100.00%{opacity:1}}.insp{animation:insp 12s cubic-bezier(.2,.8,.2,1) infinite}@keyframes insp{0.00%{opacity:0;transform:translateX(48px)}40.00%{opacity:0;transform:translateX(48px)}45.00%{opacity:1;transform:translateX(0)}100.00%{opacity:1;transform:translateX(0)}}.cover{oLine truncated

After

Width:  |  Height:  |  Size: 1.1 MiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 151 KiB

After

Width:  |  Height:  |  Size: 149 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 147 KiB

After

Width:  |  Height:  |  Size: 147 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 169 KiB

After

Width:  |  Height:  |  Size: 157 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 179 KiB

After

Width:  |  Height:  |  Size: 167 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 642 KiB

After

Width:  |  Height:  |  Size: 315 KiB

Binary file not shown.

Before

Width:  |  Height:  |  Size: 643 KiB

After

Width:  |  Height:  |  Size: 316 KiB

+9
View File
@@ -0,0 +1,9 @@
# Ideas
Plans for work we may build later. An idea records the problem, the evidence
for it, and a plan someone can pick up. Nothing here is scheduled. Defects that
CI hits today belong in [`issues/`](../issues/README.md), not here.
Start each file with its status: `Not started`, `In progress` (with a link to
the branch or PR), or `Built` (with a link to what shipped). Line references
are to the commit named at the top of each file and will drift.
+245
View File
@@ -0,0 +1,245 @@
# Capture base and head under the same conditions
Status: Not started. Line references are to `4ed584c9` (Release 0.20.7).
A review compares a route's base frame with its head frame and reports the
difference as the PR's change. That only holds when both frames were captured
from the same app state and the same keyboard. Today neither is guaranteed,
and a screen the PR did not touch can lead the comment with a large change.
## What we saw
In euxy#176's 0.20.7 review (iPhone, runs 39521260432 head and 39521251171
baseline), two rows were wrong:
1. **App state.** Dark `/rename-pattern` showed "Untitled" on base and
"Lo-Fi Bounce" on head. The row led with "Added button 'Pattern Lo-Fi
Bounce — menu'".
2. **Keyboard.** The software keyboard was drawn in some captures of the
autofocused rename sheet and not others, a 46–48% visual mismatch.
`keyboard.jsonl` said `shown` for every frame, drawn or not.
## Why
**App data carries from one appearance pass to the next.** Head ran light
routes 1–16 and then dark routes 1–16 in one session. The light pass's `/p`
import (routes 10–11) appends a pattern and makes it active, persisted in
sqlite:
```ts
// euxy src/state/store.ts
patterns: [...s.patterns, pattern],
activePatternId: pattern.id,
```
euxy captures with `launch: video`, so the dark pass relaunches through
`captureLaunch` ([`run.ts:1030`](../src/capture/run.ts)), which keeps app data
on purpose. Launch errors there are caught and logged, and routes are captured
anyway. The plain relaunch at [`run.ts:1041`](../src/capture/run.ts) only runs
when launch observation is off. The baseline split light and dark into fresh
sessions, so its dark pass started clean. Session layout is excluded from the
capture recipe on the grounds that it does not change the evidence
([`artifact-cache.ts:34`](../src/ci/artifact-cache.ts)), which is not true
today.
**Partial baseline reuse changes the order.** When only some routes are cached,
the base side captures just the rest
([`run.ts:357`](../src/capture/run.ts)):
```ts
const missingRoutes = routes.filter(route => !cachedImages[route.id]);
```
Those frames see a different history within the pass from head's.
**"Shown" is only the accessibility tree's word.** `keyboardSettle`
([`keyboard.ts`](../src/capture/keyboard.ts)) reports `shown` after two
matching tree reads, including frames with no keyboard drawn.
`keyboard.jsonl` is not in the exported diagnostics
([`run.ts:384`](../src/capture/run.ts)).
**A caveat cannot win the row.** `decideReview`
([`evidence-assessment.ts:471`](../src/review/evidence-assessment.ts)) makes
any difference `modified` and uses the first fact as the description:
```ts
const differs = visual.verdict === "modified" || changes.length > 0;
```
Review issues are appended only when there are no facts. Added and removed
elements rank ahead of text changes (priority 20 vs 30,
[`evidence-comparator.ts:143`](../src/review/evidence-comparator.ts)). With the
keyboard up, euxy's raised sheet covers Hat, Clap and Bass, so those become the
leading "Added/Removed button" facts.
**Reinstalling is not a clean slate.** Argent's `reinstall-app`
([`argent-driver.ts:712`](../src/capture/argent-driver.ts)) uninstalls and
installs, but ignores uninstall errors. iOS keychain and SecureStore data
survive an uninstall. EAS `simulator:control` has no clear-data, keychain or
keyboard action.
## Goal
For each route, base and head frames in the same appearance come from the same
app history and an equally verified keyboard state, whatever the session layout
and including reused baselines. Where the conditions still differ, the row says
so instead of reporting a PR change. Real text changes are still reported.
Done means:
- Rerun euxy#176 against a fresh baseline. Dark `/rename-pattern` and
`/patterns` show the same pattern on both sides.
- Each keyboard screen either draws the keyboard on both sides, or reads
"Added text 'Pattern name' · keyboard shown on one side only". It no longer
leads with a 46% visual change or with buttons the sheet covers.
- An unchanged screen with an autofocused field is never a PR change.
## Plan
### Phase 0: probes and diagnostics
These results decide whether phase 2 goes ahead and which reset each backend
uses.
1. **Fuller keyboard log.** Add the scheme, the pass, the tree's keyboard
bounds, Argent's `keyboardVisible`, the screen size and the app's keyboard
metrics to `keyboard.jsonl`, and export it.
2. **App-side keyboard probe on EAS**, with Argent and with agent-device. Read
`Keyboard.metrics()` and the show and hide events, and work out how much of
the window the keyboard covers. Compare that with a screenshot, a tree read,
and a second screenshot about 2 s later. Capture at least one session where
the keyboard is not drawn. The question is whether any signal separates
drawn from not drawn, and whether a missing keyboard is a late draw or a
suppressed one.
3. **Reset probe.** Write a test value into app storage through a fixture
app's deep link, reinstall through each controller on EAS and locally, and
check that it is gone. Check the keychain too, and time each reset.
### Phase 1: identical app history
4. **`resetAppData` per backend**, using what phase 0 verified:
- Argent: `reinstall-app`, checked with the test value.
- agent-device: uninstall, then install.
- Local iOS: `simctl uninstall` and `install`, plus
`simctl keychain <udid> reset`.
- Local Android: `pm clear`.
Reuse the built `localApp`. Any other backend reports `unsupported`.
5. **Reset at the start of every later captured pass** (`capturePass` in
`run.ts`):
- Run it before appearance and launch observation, so it covers
`launch: video` and `launch: off` alike.
- Count the passes captured in this session, not the loop index. A
dark-only split session starts at index 1.
- A failed reset escapes into the per-pass `try` and lands in
`sessionFailures`, not in the launch catch.
- Clear `deepLinkPromptAccepted` in the bridge's install case and approve
URL schemes again.
6. **Fall back to separate sessions.** When no verified reset exists, or the
app keeps state in the keychain on EAS, run each appearance in its own
session ([`session-plan.ts`](../src/ci/session-plan.ts)).
7. **Same history on both sides.**
- Reuse a baseline pass only if every selected route in it is cached;
otherwise capture the whole pass.
- Capture routes that exist on only one side last in that side's pass.
- Record `passStart: "fresh" | "continued"` and `reused` on each image
entry.
8. **Cache recipe.** Add `passStart` to `captureRecipeConfig`, so a capture
without the reset is never reused against one with it, even without a
version bump. Reword the `maxSessions` comment.
### Phase 2: keyboard state and an honest row
Only if phase 0 finds a signal that separates a drawn keyboard from an
undrawn one.
9. **Keyboard signal from the app.** The observer
([`observer.ts`](../src/capture/observer.ts)) reads `Keyboard.metrics()`
when it mounts, then listens for show and hide. Add the key to the
`parseCaptureDestination` allowlist so it never becomes a text node.
10. **Three-state wait:**
- `shown` when the tree and the app agree;
- `suppressed` after two reads with the tree stable and the app reporting
no coverage, without waiting out the 4 s cap;
- `unknown` when the app gives no signal.
11. **Tie the state to the screenshot.** Read it just before and just after
the screenshot ([`route-capture.ts`](../src/capture/route-capture.ts)). If
it changed, retake once, or record `unstable`. Store it on
`images[id].keyboard` so it survives the scheme merge, `export-merge`,
restore and the schema.
12. **Review rule** (`evidence-assessment.ts`, after the facts are computed).
It applies when the two sides' states differ, or when either is `unknown`
or `unstable` on a screen with a focused text field. Then:
- mark the visual verdict not comparable;
- drop added or removed elements that sit where the other side's raised
sheet or keyboard would cover them;
- keep paired text edits, and text inside the focused container;
- if text facts remain, the row is `modified` with the keyboard caveat
appended to the description after `decideReview`;
- otherwise a `scope === "none"` screen stays a variant, and anything else
is `needs-review` with the caveat;
- show the caveat in the PR comment and the Viewer.
13. **Keep the keyboard on screen where possible.** Local Android:
`settings put secure show_ime_with_hard_keyboard 1`. EAS: a serve-sim
action, if one is added. Local iOS boots headless, so phase 0 confirms
whether the Simulator.app preference applies.
### Rejected
- Dismissing the keyboard: Return submits euxy's sheet.
- Masking the keyboard band: the sheet shift is most of the mismatch.
- Detecting the keyboard from pixels: brittle across keyboards and themes.
- A fresh context for every route: cost, and
[`route-reset.ts`](../src/capture/route-reset.ts) records the settle
failures that led away from it.
- Replaying the other side's history: fragile; whole-pass reuse covers the
case.
## Risks
- **Cost:** about 25 s of reset per extra pass, measured in phase 0.
Whole-pass reuse costs more only when a pass is partly cached.
- **No usable keyboard signal:** phase 2 ships only the "keyboard couldn't be
verified" path.
- **A real change hidden behind the sheet gets dropped:** the visual verdict is
already not comparable there, and the caveat flags the screen for a person.
- **New fields lost in merge or restore:** one test runs them through every
step.
## Tests
- **`ci-map`:** reset before later passes with `launch: video` and
`launch: off`; no reset for a dark-only first pass; a failed reset lands in
`sessionFailures`; the session split when no reset exists; whole-pass reuse;
one-side-only routes last.
- **`keyboard`:** each state, Android `keyboardVisible`, and an `unstable`
change around the screenshot.
- **Observer:** the keyboard key never becomes a text node.
- **Merge, cache and schema:** `images[id].keyboard` and `passStart` survive.
- **Assessment, on the real rename trees from the runs above:** covered Hat,
Clap and Bass are dropped; "Pattern name" carries the caveat; an unchanged
autofocus screen is not a PR change; the comment row wording.
- **Local end to end:** force no keyboard on Android
(`show_ime_with_hard_keyboard 0` with a hardware keyboard) and check the
recorded state and the row.
- **Real run:** a fresh euxy baseline, then #176. Check the reset log and
timing, dark rename and `/patterns`, `keyboard.jsonl`, and the row wording.
## Open questions
1. Can serve-sim add clear-app-data, keychain-reset and hardware-keyboard
actions to EAS sessions? Without them, EAS relies on Argent's reinstall and
on separate sessions for apps that keep state in the keychain.
2. Is about 25 s per extra appearance pass acceptable by default, or should
the reset be a setting?
3. If phase 0 finds no signal that separates a drawn keyboard from an undrawn
one, is labelling every focused-field screen with a differing or unverified
keyboard as "keyboard couldn't be verified" acceptable?
## How this plan was made
Codex `gpt-6-astra`, Grok `grok-4.7` and Claude each wrote a plan
independently. Each plan was reviewed by a different model, merged, and the
merged plan reviewed again by all three, at high effort. The backbone is
Claude's; the per-frame condition record came from Codex's; keeping the
keyboard on screen and rejecting dismissal came from Grok's.
+49 -35
View File
@@ -1,12 +1,12 @@
{
"name": "@expo/explore",
"version": "0.19.5",
"version": "0.21.0",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "@expo/explore",
"version": "0.19.5",
"version": "0.21.0",
"license": "MIT",
"dependencies": {
"@expo/repack-app": "0.10.1",
@@ -29,7 +29,7 @@
"@types/react-dom": "^19.2.4",
"@vitejs/plugin-react": "^4.7.0",
"@xyflow/react": "^12.11.3",
"agent-device": "0.21.28",
"agent-device": "0.21.29-nightly.20261018.74",
"babel-plugin-react-compiler": "^1.0.0",
"eslint": "^10.12.0",
"eslint-plugin-react-hooks": "^7.1.1",
@@ -48,7 +48,7 @@
"optionalDependencies": {
"@anthropic-ai/claude-agent-sdk": "0.3.285",
"@openai/codex": "0.159.2",
"eas-cli": "25.0.0"
"eas-cli": "25.2.0"
}
},
"node_modules/@0no-co/graphql.web": {
@@ -1600,15 +1600,15 @@
}
},
"node_modules/@expo/eas-build-job": {
"version": "25.0.0",
"resolved": "https://registry.npmjs.org/@expo/eas-build-job/-/eas-build-job-25.0.0.tgz",
"integrity": "sha512-Vuai42tVopKuK2KFyTWUsW9cL1YMY1dWhFAKz+evUV5AKaMCk4dYxOYZ6tBWNhIJ15bGK3p4N0YUomlAqubFWg==",
"version": "25.2.0",
"resolved": "https://registry.npmjs.org/@expo/eas-build-job/-/eas-build-job-25.2.0.tgz",
"integrity": "sha512-RUp8sZniYWCiLQLGJrnMeAx8094BoejS6lm7EdSUs4nKkm6TtxyydRutGtoHlvBpvXOw3+OgimemqXUpmU5R8Q==",
"license": "MIT",
"optional": true,
"dependencies": {
"@expo/logger": "25.0.0",
"@expo/logger": "25.2.0",
"@expo/results": "1.0.0",
"@expo/turtle-spawn": "25.0.0",
"@expo/turtle-spawn": "25.2.0",
"joi": "^17.13.1",
"semver": "^7.6.2",
"zod": "^4.3.5"
@@ -1642,14 +1642,14 @@
}
},
"node_modules/@expo/eas-json": {
"version": "25.0.0",
"resolved": "https://registry.npmjs.org/@expo/eas-json/-/eas-json-25.0.0.tgz",
"integrity": "sha512-ilPw4O0Lnce/6lbu1GPwega8KF2GqTio9vuNFjNaGL78pfxz7GdktBsJ9JRaGtiM0LYdPa09W2FYb7DcMY8xWA==",
"version": "25.2.0",
"resolved": "https://registry.npmjs.org/@expo/eas-json/-/eas-json-25.2.0.tgz",
"integrity": "sha512-ZikfZLZHadl5gzq5UZ7qWMTWFZNAO4K6NJC6iF8JmVh3kR1Blv+Zo2AolC0CYzSYOCyaRCZx+v2ZlJZeiVCf8w==",
"license": "MIT",
"optional": true,
"dependencies": {
"@babel/code-frame": "7.23.5",
"@expo/eas-build-job": "25.0.0",
"@expo/eas-build-job": "25.2.0",
"chalk": "4.1.2",
"env-string": "1.0.1",
"fs-extra": "11.2.0",
@@ -1836,6 +1836,19 @@
"node": ">=6"
}
},
"node_modules/@expo/experimentation": {
"version": "0.0.5",
"resolved": "https://registry.npmjs.org/@expo/experimentation/-/experimentation-0.0.5.tgz",
"integrity": "sha512-eQtx5f7tUDZLeq2EXYOkBX+KivYa73CjUMkrEFh5sNMOpCWynisbdSiaSdSLHALWzYKz3uc0QlgWTz3V/6JESQ==",
"license": "MIT",
"optional": true,
"dependencies": {
"nullthrows": "^1.1.1"
},
"engines": {
"node": ">=16"
}
},
"node_modules/@expo/image-utils": {
"version": "0.7.6",
"resolved": "https://registry.npmjs.org/@expo/image-utils/-/image-utils-0.7.6.tgz",
@@ -1900,9 +1913,9 @@
}
},
"node_modules/@expo/logger": {
"version": "25.0.0",
"resolved": "https://registry.npmjs.org/@expo/logger/-/logger-25.0.0.tgz",
"integrity": "sha512-KcNOYmeJH8GESYEjbiewg+LU98XH/yBs368XlPqmEHVZekkbOONpIzPd5AFcFRWO1BW0c2nl/58BUI876UFT8A==",
"version": "25.2.0",
"resolved": "https://registry.npmjs.org/@expo/logger/-/logger-25.2.0.tgz",
"integrity": "sha512-gtf/Lk9zE3k2wfEalAwioxDwvse69A2e/OhoDT7IRjlorRgcNJghr6V4FvDAZckQUkFR1IEHz2+FXvLKf9kjgg==",
"license": "BUSL-1.1",
"optional": true,
"dependencies": {
@@ -2592,14 +2605,14 @@
}
},
"node_modules/@expo/steps": {
"version": "25.0.0",
"resolved": "https://registry.npmjs.org/@expo/steps/-/steps-25.0.0.tgz",
"integrity": "sha512-+o04mm5akQJRX0j9TwwkdZiuNM9swLxw9DsDPtFhbWceg95gky/vc/0qF200kjZL4DEMfd/WhzngBW9aqJ0JQg==",
"version": "25.2.0",
"resolved": "https://registry.npmjs.org/@expo/steps/-/steps-25.2.0.tgz",
"integrity": "sha512-KKM8STsb0dAdI6lyN7WHfczvDdQlCx+vF4unq/qFk6PZgZMvE9+IT8mh25fqgwxNcpR7Ti+xFCip2hhvpK2yng==",
"license": "BUSL-1.1",
"optional": true,
"dependencies": {
"@expo/eas-build-job": "25.0.0",
"@expo/logger": "25.0.0",
"@expo/eas-build-job": "25.2.0",
"@expo/logger": "25.2.0",
"@expo/spawn-async": "^1.7.2",
"arg": "^5.0.2",
"fast-glob": "^3.3.2",
@@ -2637,13 +2650,13 @@
"optional": true
},
"node_modules/@expo/turtle-spawn": {
"version": "25.0.0",
"resolved": "https://registry.npmjs.org/@expo/turtle-spawn/-/turtle-spawn-25.0.0.tgz",
"integrity": "sha512-XBQZ2azxh0IWozO5hCxIJt150LjIGbk6VZakeyA6UXPHp1ESenm97o0JIHOvzQNOgBV481muV88KmIWHGt0yIA==",
"version": "25.2.0",
"resolved": "https://registry.npmjs.org/@expo/turtle-spawn/-/turtle-spawn-25.2.0.tgz",
"integrity": "sha512-1Lor7/+4T1JEb6kqu0dHstD+xOWTGbwA7QqWTBKYx4qtgtJ4Wbor17bzlp87KENAYDeWuX+7VHKxT66170m9wg==",
"license": "BUSL-1.1",
"optional": true,
"dependencies": {
"@expo/logger": "25.0.0",
"@expo/logger": "25.2.0",
"@expo/spawn-async": "^1.7.2"
}
},
@@ -4580,9 +4593,9 @@
}
},
"node_modules/agent-device": {
"version": "0.21.28",
"resolved": "https://registry.npmjs.org/agent-device/-/agent-device-0.21.28.tgz",
"integrity": "sha512-uGEENVhPzSpIACrVlt16O1mloODWJyAemTCT0boDb4L006JXN2rG747u1TRoQJudOpdP/Pm2urxNRHeECLMLfg==",
"version": "0.21.29-nightly.20261018.74",
"resolved": "https://registry.npmjs.org/agent-device/-/agent-device-0.21.29-nightly.20261018.74.tgz",
"integrity": "sha512-7+ajWLr0J+8iLZ9q11ERbj6U14mC20umBK5BQQdw5aERA79yeXc3Ze6KmtvnILnWjEtoVkYQsXQasL8RL6Uatw==",
"dev": true,
"license": "MIT",
"bin": {
@@ -5722,9 +5735,9 @@
}
},
"node_modules/eas-cli": {
"version": "25.0.0",
"resolved": "https://registry.npmjs.org/eas-cli/-/eas-cli-25.0.0.tgz",
"integrity": "sha512-FBY3pIo/3L0s+8KK5ZpMKzlLrA5jOmr5kOpoOwkAXsxMrI4vo+LNoEsciRcOxjce9efc08GqEEPc/CxLRAWIvA==",
"version": "25.2.0",
"resolved": "https://registry.npmjs.org/eas-cli/-/eas-cli-25.2.0.tgz",
"integrity": "sha512-FFA6fSl2bPet9D3Eslupig3icXjyBzAbrsLE5r67wbEbh/CPDZQ+Q4W5TrpRui0uqfF16w58EGsFUQoQ0te0KQ==",
"license": "MIT",
"optional": true,
"dependencies": {
@@ -5732,11 +5745,12 @@
"@expo/code-signing-certificates": "0.0.5",
"@expo/config": "55.0.10",
"@expo/config-plugins": "55.0.7",
"@expo/eas-build-job": "25.0.0",
"@expo/eas-json": "25.0.0",
"@expo/eas-build-job": "25.2.0",
"@expo/eas-json": "25.2.0",
"@expo/env": "^1.0.0",
"@expo/experimentation": "0.0.5",
"@expo/json-file": "8.3.3",
"@expo/logger": "25.0.0",
"@expo/logger": "25.2.0",
"@expo/multipart-body-parser": "2.0.0",
"@expo/osascript": "2.1.4",
"@expo/package-manager": "1.9.10",
@@ -5748,7 +5762,7 @@
"@expo/results": "1.0.0",
"@expo/rudder-sdk-node": "1.1.1",
"@expo/spawn-async": "1.7.2",
"@expo/steps": "25.0.0",
"@expo/steps": "25.2.0",
"@expo/timeago.js": "1.0.0",
"@oclif/core": "^4.8.3",
"@oclif/plugin-autocomplete": "^3.2.40",
+4 -4
View File
@@ -1,6 +1,6 @@
{
"name": "@expo/explore",
"version": "0.19.5",
"version": "0.21.0",
"description": "Source-driven mobile app maps, deterministic CI capture, and visual reviews",
"license": "MIT",
"type": "module",
@@ -50,7 +50,7 @@
"test:node": "node --test test/node/*.test.mjs",
"test:hosting": "npm --prefix viewer-host run typecheck && npm --prefix viewer-host test",
"test:viewer:artifacts": "node --test test/browser/artifact-report.browser.mjs test/browser/artifact-cache.browser.mjs",
"test:viewer:browser": "node --test test/browser/screenshot-image.browser.mjs test/browser/inspector-notes.browser.mjs test/browser/inspector-variants.browser.mjs test/browser/inspector-source-navigation.browser.mjs test/browser/map-edge-presentation.browser.mjs test/browser/local-ci-view.browser.mjs test/browser/report-viewer.browser.mjs",
"test:viewer:browser": "node --test test/browser/screenshot-image.browser.mjs test/browser/swipe-outlines.browser.mjs test/browser/split-outlines.browser.mjs test/browser/inspector-notes.browser.mjs test/browser/inspector-variants.browser.mjs test/browser/inspector-source-navigation.browser.mjs test/browser/map-edge-presentation.browser.mjs test/browser/local-ci-view.browser.mjs test/browser/report-viewer.browser.mjs",
"test:viewer:e2e": "npm run test:viewer:browser && playwright test",
"typecheck": "tsc -p tsconfig.json --noEmit && tsc -p viewer/tsconfig.json --noEmit",
"lint": "eslint .",
@@ -73,7 +73,7 @@
"optionalDependencies": {
"@anthropic-ai/claude-agent-sdk": "0.3.285",
"@openai/codex": "0.159.2",
"eas-cli": "25.0.0"
"eas-cli": "25.2.0"
},
"devDependencies": {
"@anthropic-ai/claude-code": "2.1.285",
@@ -83,7 +83,7 @@
"@types/react-dom": "^19.2.4",
"@vitejs/plugin-react": "^4.7.0",
"@xyflow/react": "^12.11.3",
"agent-device": "0.21.28",
"agent-device": "0.21.29-nightly.20261018.74",
"babel-plugin-react-compiler": "^1.0.0",
"eslint": "^10.12.0",
"eslint-plugin-react-hooks": "^7.1.1",
+9
View File
@@ -202,6 +202,15 @@
"mimeType": { "type": "string", "minLength": 1 },
"width": { "type": "integer", "minimum": 0 },
"height": { "type": "integer", "minimum": 0 },
"turned": {
"type": "object",
"additionalProperties": false,
"required": ["width", "height"],
"properties": {
"width": { "type": "number", "exclusiveMinimum": 0 },
"height": { "type": "number", "exclusiveMinimum": 0 }
}
},
"createdAt": { "$ref": "#/$defs/timestamp" }
}
},
+2 -1
View File
@@ -10,7 +10,8 @@ const SESSION_ID = /^[A-Za-z0-9][A-Za-z0-9._-]{0,199}$/;
/** Stop only exact sessions whose ownership the capture coordinator already verified. */
export async function cleanupCiSessions({ sessionFiles, easCommand, environment = process.env, runCommand = spawnSync }) {
if (!Array.isArray(sessionFiles) || sessionFiles.length > 3) throw new Error('Provide at most three explicit session handoff files.');
// head, base and baseline, each with a session per appearance pass (light, dark) beside its own.
if (!Array.isArray(sessionFiles) || sessionFiles.length > 9) throw new Error('Provide at most nine explicit session handoff files.');
const temporary = await realpath(environment.RUNNER_TEMP || tmpdir());
const result = { stopped: [], failures: [] };
let project;
+6 -5
View File
@@ -3,13 +3,14 @@
//
// node scripts/deploy-viewer-preview.mjs --fixtures DIR --alias pr-123 --project explore-viewer-pr [--owner expo-public]
//
// DIR holds reports.json ([{ platform, changes?, device? }], as a hosted preview's /reports.json lists them) and one
// <platform>.zip per entry (scripts/refresh-viewer-preview-fixtures.sh writes both). EXPLORE_PREVIEW_TOKEN keeps an
// earlier link's /p/<token>; without it the link gets a new one. EXPO_TOKEN must be able to deploy to the project.
// DIR holds reports.json ([{ target?, platform, label?, changes?, device? }], as a hosted preview's /reports.json lists
// them) and one <target>.zip per entry, or <platform>.zip for an entry without a target
// (scripts/refresh-viewer-preview-fixtures.sh writes both). EXPLORE_PREVIEW_TOKEN keeps an earlier link's /p/<token>;
// without it the link gets a new one. EXPO_TOKEN must be able to deploy to the project.
import { readFile } from "node:fs/promises";
import { join } from "node:path";
import { parseArgs } from "node:util";
import { deployFixturePreview } from "../dist/report/preview-deploy.js";
import { deployFixturePreview, fixtureReports } from "../dist/report/preview-deploy.js";
const { values } = parseArgs({ options: { fixtures: { type: "string" }, alias: { type: "string" }, project: { type: "string" }, owner: { type: "string" } } });
if (!values.fixtures || !values.alias || !values.project) {
@@ -19,7 +20,7 @@ if (!values.fixtures || !values.alias || !values.project) {
const listing = JSON.parse(await readFile(join(values.fixtures, "reports.json"), "utf8"));
const token = process.env.EXPLORE_PREVIEW_TOKEN?.trim() || undefined;
const result = await deployFixturePreview({
reports: listing.map(({ platform, changes, device }) => ({ platform, zipPath: join(values.fixtures, `${platform}.zip`), changes, device })),
reports: fixtureReports(listing, values.fixtures),
project: values.project, alias: values.alias, ...(values.owner ? { owner: values.owner } : {}), ...(token ? { token } : {}),
});
console.log(JSON.stringify({ ...result, reports: listing }));
+265
View File
@@ -0,0 +1,265 @@
// The README's two animated SVGs, written by readme-screenshots.mjs. Each is a plain image: CSS keyframes on one loop,
// every picture inlined as a JPEG data URI (an SVG shown with <img> loads nothing), and with prefers-reduced-motion
// it shows its last frame, which is also what a viewer that does not animate SVG shows.
//
// readme-hero the Viewer's Map, built from real Viewer frames: the screens appear row by row, the transitions
// wipe in from the top, then a click selects a screen and the Inspector slides in
// readme-compare one modified screen wiping from its base capture to its head capture, the changed regions pulsing
import { ZipReader, Uint8ArrayReader, Uint8ArrayWriter } from "@zip.js/zip.js";
/** The soft stage the PR comment picture sits on too (stagedComment in readme-screenshots.mjs), and its accents. */
const STAGE = {
light: { stage: "#F6F8FA", border: "#D1D9E0", shadow: "rgba(17,17,17,0.14)", text: "#1F2328", muted: "#59636E", accent: "#0969DA", region: "#D4520F" },
dark: { stage: "#151B23", border: "#3D444D", shadow: "rgba(1,4,9,0.6)", text: "#F0F6FC", muted: "#9198A1", accent: "#4493F8", region: "#F0883E" },
};
const FONT = "-apple-system, BlinkMacSystemFont, &apos;Segoe UI&apos;, &apos;Noto Sans&apos;, Helvetica, Arial, sans-serif";
const MONO = "ui-monospace, SFMono-Regular, &apos;SF Mono&apos;, Menlo, Consolas, monospace";
const JPEG_QUALITY = 82;
const dataUri = jpeg => `data:image/jpeg;base64,${jpeg.toString("base64")}`;
const escapeXml = text => text.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;");
const box = ({ x, y, width, height }) => `x="${x}" y="${y}" width="${width}" height="${height}"`;
/** Keyframes given in seconds on a shared loop, so every element stays in step. */
function keyframes(name, loop, frames) {
return `@keyframes ${name}{${frames.map(([at, css]) => `${((at / loop) * 100).toFixed(2)}%{${css}}`).join("")}}`;
}
/**
* The Map card of `route`: the screenshot button of the node whose id is that route's surface in `review`. The button
* is named after the screen's title, which is not the route when the report has plain titles.
*/
export function mapCard(page, review, route) {
const surface = review.surfaces.find(s => (s.head ?? s.base)?.route === route && s.colorScheme !== "dark");
if (!surface) throw new Error(`The report has no screen at ${route}.`);
return page.locator(`.react-flow__node-screen[data-id="${surface.id}"] .screen-screenshot-target`);
}
/**
* Photographs the hero's states from a Viewer page whose Map has its camera set and nothing selected: the Map without
* its screens, without its edges, and whole; then it clicks `select`, which opens the Inspector without moving the
* camera, and photographs that (`card` is its screenshot button, from mapCard). Returns the frames and where each screen card, the Inspector and the selected card are.
*/
export async function captureHeroFrames(page, { select, card: target, viewport }) {
const shoot = () => page.screenshot({ type: "jpeg", quality: JPEG_QUALITY });
// Park the pointer on bare canvas, found once the camera is set: over a card it takes a grey ring and darkens the
// card's transitions, over a transition it darkens that one, and over a control it can raise a tooltip.
const rest = await page.evaluate(({ width, height }) => {
for (let y = height - 120; y > 120; y -= 20) for (let x = 40; x < width - 40; x += 20) {
if (document.elementFromPoint(x, y)?.classList.contains("react-flow__pane")) return { x, y };
}
return null;
}, viewport);
if (!rest) throw new Error("The hero's Map has no bare canvas to park the pointer on.");
const neutral = async () => {
await page.mouse.move(rest.x, rest.y);
await page.waitForTimeout(200);
if (await page.locator(".canvas-hovering, .graph-node-hover-subject").count()) throw new Error("The pointer at rest still hovers the Map.");
};
const rect = locator => locator.boundingBox().then(b => b && { x: Math.round(b.x), y: Math.round(b.y), width: Math.round(b.width), height: Math.round(b.height) });
const camera = () => page.locator(".react-flow__viewport").evaluate(el => el.style.transform);
await neutral();
await page.waitForTimeout(1000);
const at = await camera();
const selected = await rect(target);
if (!selected) throw new Error(`The hero needs ${select} on the Map.`);
const cards = [];
for (const node of await page.locator(".react-flow__node").all()) {
const b = await rect(node);
if (b && b.x < viewport.width && b.y < viewport.height && b.x + b.width > 0 && b.y + b.height > 0) cards.push(b);
}
const full = await shoot();
const hide = async css => {
const style = await page.addStyleTag({ content: `${css} { visibility: hidden !important; }` });
await page.waitForTimeout(300);
return style;
};
const edges = await hide(".react-flow__edges, .react-flow__edge, .react-flow__edgelabel-renderer");
const screensOnly = await shoot();
const nodes = await hide(".react-flow__node");
const empty = await shoot();
for (const style of [edges, nodes]) await style.evaluate(el => el.remove());
await target.click();
await neutral();
await page.waitForTimeout(1500);
const inspector = await rect(page.locator("aside.right-panel"));
if (!inspector) throw new Error(`Selecting ${select} did not open the Inspector.`);
if ((await camera()) !== at) throw new Error(`The Map's camera moved when ${select} was selected: ${at} → ${await camera()}.`);
const inspecting = await shoot();
return { viewport, empty, screensOnly, full, inspecting, cards, selected, inspector };
}
/**
* The hero: the frames from captureHeroFrames as a window on a rounded 1600×900 stage. Each screen
* card is a clip of the screens-only frame, so what appears is exactly what the Viewer draws.
*/
export function heroSvg(frames, theme, alt) {
const t = STAGE[theme];
const LOOP = 12;
const W = 1600, H = 900;
const { viewport: v } = frames;
const wx = (W - v.width) / 2, wy = (H - v.height) / 2;
const css = [];
const defs = [];
const layers = [];
const image = (id, jpeg) => defs.push(`<image id="${id}" width="${v.width}" height="${v.height}" href="${dataUri(jpeg)}"/>`);
image("empty", frames.empty);
image("screens", frames.screensOnly);
image("full", frames.full);
image("inspecting", frames.inspecting);
// Cards appear a row at a time, top to bottom, as the Map lays out taps from launch; a row's cards left to right.
const PAD = 18;
const rows = [...new Set(frames.cards.map(c => c.y))].sort((a, b) => a - b);
frames.cards.forEach((card, i) => {
const appear = 0.4 + rows.indexOf(card.y) * 0.35 + (card.x / v.width) * 0.3;
const clip = { x: card.x - PAD, y: card.y - PAD, width: card.width + PAD * 2, height: card.height + PAD * 2 };
defs.push(`<clipPath id="c${i}"><rect ${box(clip)}/></clipPath>`);
css.push(`.c${i}{transform-origin:0 0;animation:c${i} ${LOOP}s cubic-bezier(.2,.8,.2,1) infinite}`);
css.push(keyframes(`c${i}`, LOOP, [[0, "opacity:0;transform:scale(.92)"], [appear, "opacity:0;transform:scale(.92)"], [appear + 0.45, "opacity:1;transform:scale(1)"], [LOOP, "opacity:1;transform:scale(1)"]]));
// Scaled about the card's centre: the class sits on a group whose origin is there.
const cx = card.x + card.width / 2, cy = card.y + card.height / 2;
layers.push(`<g transform="translate(${cx} ${cy})"><g class="c${i}"><g transform="translate(${-cx} ${-cy})" clip-path="url(#c${i})"><use href="#screens"/></g></g></g>`);
});
const settled = 0.4 + (rows.length - 1) * 0.35 + 0.75;
// The transitions wipe in from the top: the full frame over the cards, which it draws in the same places.
const EDGES = settled - 0.2, EDGES_END = EDGES + 1.3;
css.push(`.full{animation:full ${LOOP}s cubic-bezier(.45,0,.25,1) infinite}`);
css.push(keyframes("full", LOOP, [[0, "clip-path:inset(0 0 100% 0)"], [EDGES, "clip-path:inset(0 0 100% 0)"], [EDGES_END, "clip-path:inset(0 0 0 0)"], [LOOP, "clip-path:inset(0 0 0 0)"]]));
layers.push(`<use class="full" href="#full"/>`);
// A pointer glides to the selected card and clicks; the selection lands and the Inspector slides in.
const { selected: s, inspector: p } = frames;
const target = { x: s.x + s.width * 0.55, y: s.y + s.height * 0.45 };
const start = { x: Math.min(v.width - 120, target.x + 260), y: Math.min(v.height - 60, target.y + 220) };
const MOVE = EDGES_END + 0.5, CLICK = MOVE + 1.2;
const pointer = (x, y, extra = "") => `transform:translate(${x.toFixed(1)}px,${y.toFixed(1)}px)${extra}`;
css.push(`.ptr{opacity:0;animation:ptr ${LOOP}s cubic-bezier(.4,0,.2,1) infinite}`);
css.push(keyframes("ptr", LOOP, [
[0, `opacity:0;${pointer(start.x, start.y)}`], [MOVE, `opacity:0;${pointer(start.x, start.y)}`], [MOVE + 0.2, `opacity:1;${pointer(start.x, start.y)}`],
[CLICK, `opacity:1;${pointer(target.x, target.y)}`], [CLICK + 0.1, `opacity:1;${pointer(target.x, target.y, " scale(.82)")}`],
[CLICK + 0.25, `opacity:1;${pointer(target.x, target.y)}`], [CLICK + 1.2, `opacity:0;${pointer(target.x, target.y)}`], [LOOP, `opacity:0;${pointer(target.x, target.y)}`],
]));
defs.push(`<clipPath id="canvas"><rect x="0" y="0" width="${p.x}" height="${v.height}"/></clipPath>`);
defs.push(`<clipPath id="panel"><rect ${box(p)}/></clipPath>`);
// The selection lands as the Viewer draws it, at once: a ring and its transitions in the accent, nothing fading.
css.push(`.sel{animation:sel ${LOOP}s linear infinite}`);
css.push(keyframes("sel", LOOP, [[0, "opacity:0"], [CLICK + 0.1, "opacity:0"], [CLICK + 0.2, "opacity:1"], [LOOP, "opacity:1"]]));
css.push(`.insp{animation:insp ${LOOP}s cubic-bezier(.2,.8,.2,1) infinite}`);
css.push(keyframes("insp", LOOP, [[0, "opacity:0;transform:translateX(48px)"], [CLICK + 0.15, "opacity:0;transform:translateX(48px)"], [CLICK + 0.75, "opacity:1;transform:translateX(0)"], [LOOP, "opacity:1;transform:translateX(0)"]]));
layers.push(`<use class="sel" href="#inspecting" clip-path="url(#canvas)"/>`);
layers.push(`<g class="insp"><use href="#inspecting" clip-path="url(#panel)"/></g>`);
layers.push(`<g class="ptr"><path d="M0 0L0 22L6 16.5L10.5 26L14 24.5L9.8 15.2L18 15.2Z" fill="#111" stroke="#fff" stroke-width="1.6" stroke-linejoin="round"/></g>`);
// The empty Map covers the reset: it fades in over the last frame, and the loop starts again beneath it.
css.push(`.cover{opacity:0;animation:cover ${LOOP}s linear infinite}`);
css.push(keyframes("cover", LOOP, [[0, "opacity:1"], [0.2, "opacity:0"], [LOOP - 0.7, "opacity:0"], [LOOP, "opacity:1"]]));
layers.push(`<use class="cover" href="#empty"/>`);
css.push("@media (prefers-reduced-motion: reduce){*{animation:none!important}}");
defs.push(`<clipPath id="window"><rect width="${v.width}" height="${v.height}" rx="14"/></clipPath>`);
defs.push(`<filter id="shadow" x="-10%" y="-10%" width="120%" height="125%"><feDropShadow dx="0" dy="2" stdDeviation="2" flood-color="${t.shadow}" flood-opacity=".3"/><feDropShadow dx="0" dy="30" stdDeviation="35" flood-color="${t.shadow}"/></filter>`);
defs.push(`<clipPath id="stage"><rect width="${W}" height="${H}" rx="22"/></clipPath>`);
return `<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 ${W} ${H}" width="${W}" height="${H}" role="img" aria-label="${escapeXml(alt)}">`
+ `<style>${css.join("")}</style><defs>${defs.join("")}</defs>`
+ `<g clip-path="url(#stage)"><rect width="${W}" height="${H}" fill="${t.stage}"/>`
+ `<rect x="${wx}" y="${wy}" width="${v.width}" height="${v.height}" rx="14" fill="${t.stage}" filter="url(#shadow)"/>`
+ `<g transform="translate(${wx} ${wy})"><g clip-path="url(#window)"><use href="#empty"/>${layers.join("")}</g>`
+ `<rect x=".5" y=".5" width="${v.width - 1}" height="${v.height - 1}" rx="14" fill="none" stroke="${t.border}"/></g></g></svg>`;
}
/** review.json and an asset reader from a sealed review report's bytes. */
export async function openReport(bytes) {
const reader = new ZipReader(new Uint8ArrayReader(new Uint8Array(bytes)));
const entries = new Map((await reader.getEntries()).map(entry => [entry.filename, entry]));
const read = async name => {
const entry = entries.get(name);
if (!entry) throw new Error(`The report has no ${name}.`);
return Buffer.from(await entry.getData(new Uint8ArrayWriter()));
};
const review = JSON.parse((await read("review.json")).toString("utf8")).review;
return { review, asset: sha => read(`assets/${sha}.png`) };
}
/** A PNG as a JPEG at most `maxHeight` tall, drawn by the browser so the script needs no image library. */
async function toJpeg(browser, png, maxHeight) {
const page = await browser.newPage();
try {
const base64 = await page.evaluate(async ({ png, maxHeight, quality }) => {
const bitmap = await createImageBitmap(await (await fetch(`data:image/png;base64,${png}`)).blob());
const scale = Math.min(1, maxHeight / bitmap.height);
const canvas = document.createElement("canvas");
canvas.width = Math.round(bitmap.width * scale);
canvas.height = Math.round(bitmap.height * scale);
const context = canvas.getContext("2d");
context.imageSmoothingQuality = "high";
context.drawImage(bitmap, 0, 0, canvas.width, canvas.height);
return canvas.toDataURL("image/jpeg", quality).split(",")[1];
}, { png: png.toString("base64"), maxHeight, quality: JPEG_QUALITY / 100 });
return Buffer.from(base64, "base64");
} finally {
await page.close();
}
}
/**
* The surface of the Compare key `key` (the slug of the Viewer's #/compare/<key>, which is the route without its
* slash) in `theme`, among the report's visually modified screens.
*/
export function compareSurface(review, key, theme) {
const route = `/${key}`;
const surface = review.surfaces.find(s => (s.head ?? s.base)?.route === route && s.colorScheme === theme && s.visual?.verdict === "modified");
if (!surface) throw new Error(`No visually modified ${route} in ${theme} in this report; pass --compare <key> from a #/compare/<key> URL.`);
return surface;
}
/** The Compare picture: `surface`'s base capture wiping to its head capture, its changed regions outlined. */
export async function compareSvg(browser, report, surface, theme, alt) {
const t = STAGE[theme];
const LOOP = 7;
const W = 1000, H = 600;
const { baseImage, headImage } = surface;
const PH = 540, PW = Math.round(PH * headImage.width / headImage.height);
const x = (W - PW) / 2, y = (H - PH) / 2;
const k = PW / headImage.width;
const before = dataUri(await toJpeg(browser, await report.asset(baseImage.sha256), PH * 2));
const after = dataUri(await toJpeg(browser, await report.asset(headImage.sha256), PH * 2));
const { pullRequest } = report.review;
const css = [];
// The After capture is revealed left to right, and the divider rides the edge of the reveal.
css.push(`.after{animation:wipe ${LOOP}s cubic-bezier(.65,0,.35,1) infinite}`);
css.push(keyframes("wipe", LOOP, [[0, "clip-path:inset(0 100% 0 0)"], [1.0, "clip-path:inset(0 100% 0 0)"], [2.6, "clip-path:inset(0 0 0 0)"], [5.8, "clip-path:inset(0 0 0 0)"], [6.8, "clip-path:inset(0 100% 0 0)"], [LOOP, "clip-path:inset(0 100% 0 0)"]]));
css.push(`.divider{opacity:0;animation:divider ${LOOP}s cubic-bezier(.65,0,.35,1) infinite}`);
css.push(keyframes("divider", LOOP, [[0, "transform:translateX(0);opacity:0"], [0.8, "transform:translateX(0);opacity:1"], [1.0, "transform:translateX(0);opacity:1"], [2.6, `transform:translateX(${PW}px);opacity:1`], [2.9, `transform:translateX(${PW}px);opacity:0`], [5.8, `transform:translateX(${PW}px);opacity:0`], [6.0, `transform:translateX(${PW}px);opacity:1`], [6.8, "transform:translateX(0);opacity:1"], [LOOP, "transform:translateX(0);opacity:0"]]));
css.push(`.regions{animation:regions ${LOOP}s ease-in-out infinite}`);
css.push(keyframes("regions", LOOP, [[0, "opacity:0"], [2.7, "opacity:0"], [3.1, "opacity:1"], [3.6, "opacity:.3"], [4.1, "opacity:1"], [5.7, "opacity:1"], [6.0, "opacity:0"], [LOOP, "opacity:0"]]));
css.push(`.before-label{opacity:.3;animation:before-label ${LOOP}s linear infinite}.after-label{animation:after-label ${LOOP}s linear infinite}`);
css.push(keyframes("before-label", LOOP, [[0, "opacity:1"], [1.8, "opacity:1"], [2.0, "opacity:.3"], [6.2, "opacity:.3"], [6.4, "opacity:1"], [LOOP, "opacity:1"]]));
css.push(keyframes("after-label", LOOP, [[0, "opacity:.3"], [1.8, "opacity:.3"], [2.0, "opacity:1"], [6.2, "opacity:1"], [6.4, "opacity:.3"], [LOOP, "opacity:.3"]]));
css.push("@media (prefers-reduced-motion: reduce){*{animation:none!important}}");
const regions = surface.visual.regions.map(({ rect: r }) =>
`<rect x="${(x + r.x * k - 3).toFixed(1)}" y="${(y + r.y * k - 3).toFixed(1)}" width="${(r.width * k + 6).toFixed(1)}" height="${(r.height * k + 6).toFixed(1)}" rx="3" fill="none" stroke="${t.region}" stroke-width="2"/>`).join("");
const label = (cls, anchor, lx, title, sub, note) => `<g class="${cls}" font-family="${FONT}" text-anchor="${anchor}">`
+ `<text x="${lx}" y="${y + PH / 2 - 14}" font-size="26" font-weight="700" fill="${t.text}">${title}</text>`
+ `<text x="${lx}" y="${y + PH / 2 + 12}" font-family="${MONO}" font-size="13" fill="${t.muted}">${escapeXml(sub)}</text>`
+ (note ? `<text x="${lx}" y="${y + PH / 2 + 36}" font-size="13" font-weight="600" fill="${t.region}">${note}</text>` : "") + "</g>";
const count = surface.visual.regions.length;
return `<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 ${W} ${H}" width="${W}" height="${H}" role="img" aria-label="${escapeXml(alt)}">`
+ `<style>${css.join("")}</style>`
+ `<defs><clipPath id="stage"><rect width="${W}" height="${H}" rx="16"/></clipPath><clipPath id="phone"><rect x="${x}" y="${y}" width="${PW}" height="${PH}" rx="40"/></clipPath>`
+ `<filter id="shadow" x="-20%" y="-20%" width="140%" height="140%"><feDropShadow dx="0" dy="8" stdDeviation="14" flood-color="${t.shadow}"/></filter></defs>`
+ `<g clip-path="url(#stage)"><rect width="${W}" height="${H}" fill="${t.stage}"/>`
+ `<rect x="${x}" y="${y}" width="${PW}" height="${PH}" rx="40" fill="${t.stage}" filter="url(#shadow)"/>`
+ `<g clip-path="url(#phone)"><image href="${before}" x="${x}" y="${y}" width="${PW}" height="${PH}"/>`
+ `<image class="after" href="${after}" x="${x}" y="${y}" width="${PW}" height="${PH}"/></g>`
+ `<rect x="${x}" y="${y}" width="${PW}" height="${PH}" rx="40" fill="none" stroke="${t.border}"/>`
+ `<g class="divider"><rect x="${x - 1.5}" y="${y}" width="3" height="${PH}" fill="${t.accent}"/><circle cx="${x}" cy="${y + PH / 2}" r="13" fill="${t.accent}"/>`
+ `<path d="M${x - 5} ${y + PH / 2 - 4}l-3 4l3 4M${x + 5} ${y + PH / 2 - 4}l3 4l-3 4" stroke="#fff" stroke-width="1.8" fill="none" stroke-linecap="round" stroke-linejoin="round"/></g>`
+ `<g class="regions">${regions}</g>`
+ label("before-label", "end", x - 44, "Before", `${pullRequest.base.ref} · ${pullRequest.base.sha.slice(0, 7)}`)
+ label("after-label", "start", x + PW + 44, "After", `PR #${pullRequest.number} · ${pullRequest.head.sha.slice(0, 7)}`, `${count} changed region${count === 1 ? "" : "s"}`)
+ "</g></svg>";
}
+51 -47
View File
@@ -1,4 +1,4 @@
// Regenerates the Viewer screenshots in docs/screenshots from a review report, in light and dark
// Regenerates the README's pictures in docs/screenshots from a review report, in light and dark
// (the README shows them with <picture>; GitHub Mobile ignores the dark source and shows light). The reference app is euxy
// (github.com/brentvatne/euxy): skills/readme-refresh/SKILL.md says which report to pass and how.
//
@@ -7,15 +7,20 @@
//
// --report a sealed review report (the hosted link's report.zip?platform=ios, or the PR's zip artifact)
// --select the route selected on the hero Map (opens the Inspector); default /settings. The hero zooms in around it
// (--zoom, default 73%) and is framed as a window on a soft stage, like the PR comment
// --compare the Compare key of a modified screen: the slug in the Viewer's #/compare/<key> URL; default patterns
// (--zoom, default 73%) and is an animated SVG of a window on a soft stage (readme-animations.mjs): the
// Map's cards appear, its transitions wipe in, and a click on this screen opens the Inspector
// --compare the Compare key of a modified screen: the slug in the Viewer's #/compare/<key> URL; default patterns.
// It becomes readme-compare-*.svg, its before capture wiping to its after capture
// --pr a GitHub PR URL whose Explore comment becomes readme-pr-comment-light.png (public repository): the comment
// at its own size on a soft stage, cut off with a fade after the list of changed screens
// --only the pictures to write, comma-separated, of hero, compare, screens and pr-comment; default all of them (the
// PR comment only with --pr). --only pr-comment needs no --report
import { chromium } from "@playwright/test";
import { readFile, writeFile } from "node:fs/promises";
import { parseArgs } from "node:util";
import { fileURLToPath } from "node:url";
import { startReportViewer } from "../test/ci/lib/report-viewer.mjs";
import { captureHeroFrames, compareSurface, compareSvg, heroSvg, mapCard, openReport } from "./readme-animations.mjs";
const { values: args } = parseArgs({
options: {
@@ -24,24 +29,42 @@ const { values: args } = parseArgs({
compare: { type: "string", default: "patterns" },
pr: { type: "string" },
zoom: { type: "string", default: "73" },
only: { type: "string" },
},
});
if (!args.report) {
const PICTURES = ["hero", "compare", "screens", "pr-comment"];
const only = args.only?.split(",").map(name => name.trim()).filter(Boolean);
const unknown = only?.filter(name => !PICTURES.includes(name)) ?? [];
if (unknown.length) {
console.error(`Unknown --only ${unknown.join(", ")}; pass some of ${PICTURES.join(", ")}.`);
process.exit(2);
}
const wanted = name => !only || only.includes(name);
const fromReport = PICTURES.filter(name => name !== "pr-comment").some(wanted);
if (fromReport && !args.report) {
console.error("Pass --report <review report zip>. See skills/readme-refresh/SKILL.md for where the euxy report comes from.");
process.exit(2);
}
const out = fileURLToPath(new URL("../docs/screenshots/", import.meta.url));
const SHOTS = [
{ file: "readme-hero", hash: "#/map", viewport: { width: 1376, height: 774 }, click: "Fit canvas", select: args.select, zoom: Number(args.zoom), framed: true },
{ file: "readme-compare", hash: `#/compare/${args.compare}`, viewport: { width: 1760, height: 1000 } },
{ file: "readme-screens", hash: "#/screens", viewport: { width: 1760, height: 1000 } },
];
{ name: "hero", file: "readme-hero", hash: "#/map", viewport: { width: 1376, height: 774 }, click: page => page.getByRole("button", { name: "Fit canvas" }), select: args.select, zoom: Number(args.zoom), animated: true },
// Screens shows its Changed filter, so the added, removed and modified dots are all in view wherever the tree sorts them.
{ name: "screens", file: "readme-screens", hash: "#/screens", viewport: { width: 1760, height: 1000 }, click: page => page.locator(".screens-chip", { hasText: "Changed" }) },
].filter(shot => wanted(shot.name));
const bytes = await readFile(args.report);
const { server, url } = await startReportViewer({ bytes, watch: false });
const bytes = fromReport ? await readFile(args.report) : undefined;
const report = bytes && await openReport(bytes);
const { server, url } = bytes ? await startReportViewer({ bytes, watch: false }) : {};
const browser = await chromium.launch();
try {
for (const theme of wanted("compare") ? ["light", "dark"] : []) {
const surface = compareSurface(report.review, args.compare, theme);
const path = `${out}readme-compare-${theme}.svg`;
await writeFile(path, await compareSvg(browser, report, surface, theme,
`The ${surface.head.route} screen of euxy wiping from its capture before the pull request to its capture after it, with the changed regions outlined`));
console.log(path);
}
for (const theme of ["light", "dark"]) for (const shot of SHOTS) {
const context = await browser.newContext({ viewport: shot.viewport, deviceScaleFactor: 2, colorScheme: theme });
await context.addInitScript(value => localStorage.setItem("expo-explore:theme", value), theme);
@@ -50,20 +73,20 @@ try {
await page.waitForSelector(".react-flow__node, .compare-stage, .screens-cell", { timeout: 60_000 });
await page.waitForTimeout(4000);
if (shot.click) {
await page.getByRole("button", { name: shot.click }).click();
await shot.click(page).click();
await page.mouse.move(400, shot.viewport.height - 30);
await page.waitForTimeout(1500);
}
if (shot.select) {
// Selecting a screen opens the Inspector without moving the fitted camera.
await page.locator(`.screen-screenshot-target[title="${shot.select}"]`).click();
if (shot.select && !shot.animated) {
// Selecting a screen opens the Inspector without moving the fitted camera. The animated hero selects it last.
await mapCard(page, report.review, shot.select).click();
await page.mouse.move(400, shot.viewport.height - 30);
await page.waitForTimeout(1500);
}
if (shot.zoom) {
// Zoom around the selected screen with the wheel, which keeps it under the pointer, then pan it left and up, clear
// of the minimap and the bottom bar. The camera stops at the first wheel step past the zoom asked for.
const box = await page.locator(`.screen-screenshot-target[title="${shot.select}"]`).boundingBox();
const box = await mapCard(page, report.review, shot.select).boundingBox();
await page.mouse.move(box.x + box.width / 2, box.y + box.height / 2);
for (let step = 0; step < 60; step++) {
if (Number((await page.locator(".zoom-button span").first().textContent()).replace("%", "")) >= shot.zoom) break;
@@ -76,31 +99,32 @@ try {
await page.mouse.move(shot.viewport.width - 60, shot.viewport.height / 2);
await page.waitForTimeout(1200);
}
if (shot.hash.startsWith("#/compare/") && !page.url().includes(shot.hash)) {
throw new Error(`Compare key ${args.compare} is not a screen of this report (the Viewer went to ${page.url()}); pass --compare <key> from a #/compare/<key> URL.`);
}
const path = `${out}${shot.file}-${theme}.png`;
if (shot.framed) await writeFile(path, await stagedWindow(browser, await page.screenshot(), shot.viewport, theme));
else await page.screenshot({ path });
const path = `${out}${shot.file}-${theme}.${shot.animated ? "svg" : "png"}`;
if (shot.animated) {
const frames = await captureHeroFrames(page, { select: shot.select, card: mapCard(page, report.review, shot.select), viewport: shot.viewport });
await writeFile(path, heroSvg(frames, theme,
`Explore's Map of a euxy pull request: the screen cards appear row by row, the transitions between them wipe in, then a click selects ${shot.select} and the Inspector opens on it`));
} else await page.screenshot({ path });
console.log(path);
await context.close();
}
for (const theme of args.pr ? ["light", "dark"] : []) {
for (const theme of args.pr && wanted("pr-comment") ? ["light", "dark"] : []) {
// Signed-out github.com follows prefers-color-scheme.
const context = await browser.newContext({ viewport: { width: 1000, height: 1400 }, deviceScaleFactor: 2, colorScheme: theme });
const page = await context.newPage();
await page.goto(args.pr, { waitUntil: "domcontentloaded" });
const comment = page.locator(".timeline-comment").filter({ has: page.locator("h3", { hasText: /Explore review/ }) }).first();
const comment = page.locator(".timeline-comment").filter({ has: page.locator("h3", { hasText: /Review in Explore|Explore review/ }) }).first();
await comment.waitFor({ timeout: 30_000 });
await comment.scrollIntoViewIfNeeded();
// The comment's pointer toward the avatar would show as a stray tick above the card.
await page.addStyleTag({ content: ".timeline-comment::before, .timeline-comment::after { display: none !important; }" });
await page.waitForTimeout(1000);
// Cut after the list of changed screens: the launch timings, validation, and footer below it are detail.
// Cut after the list of changed screens: the collapsed blocks and footer below it are detail.
const clip = await comment.evaluate((card) => {
const box = card.getBoundingClientRect();
const after = [...card.querySelectorAll("p")].find(p => /Launch:/.test(p.textContent ?? ""));
const bottom = after ? after.getBoundingClientRect().top : box.bottom;
const list = [...card.querySelectorAll("ul")].find(ul => ul.querySelector(":scope > li code"));
const after = list?.nextElementSibling;
const bottom = after ? after.getBoundingClientRect().top : list ? list.getBoundingClientRect().bottom : box.bottom;
return { x: box.left + window.scrollX, y: box.top + window.scrollY, width: box.width, height: Math.round(bottom - box.top) };
});
const shot = await page.screenshot({ clip, fullPage: true });
@@ -111,27 +135,7 @@ try {
}
} finally {
await browser.close();
await server.close?.();
}
/**
* A Viewer screenshot as a window on a rounded 1600×900 stage (drawn at 2x), with a hairline border and a soft shadow,
* as the post about Explore frames it. The corners outside the stage stay transparent, so it sits on either GitHub theme.
*/
async function stagedWindow(browser, shot, viewport, theme) {
const dark = theme === "dark";
const stage = dark ? "#151B23" : "#F6F8FA";
const border = dark ? "#3D444D" : "#D1D9E0";
const shadow = dark ? "0 2px 4px rgba(1,4,9,0.4), 0 30px 70px rgba(1,4,9,0.6)" : "0 2px 4px rgba(17,17,17,0.04), 0 30px 70px rgba(17,17,17,0.14)";
const context = await browser.newContext({ viewport: { width: 1600, height: 900 }, deviceScaleFactor: 2 });
const page = await context.newPage();
await page.setContent(`<!doctype html><html><body style="margin:0;background:transparent">
<div id="stage" style="position:relative;width:1600px;height:900px;background:${stage};border-radius:22px;overflow:hidden">
<img src="data:image/png;base64,${shot.toString("base64")}" style="position:absolute;left:${(1600 - viewport.width) / 2}px;top:${(900 - viewport.height) / 2}px;width:${viewport.width}px;height:${viewport.height}px;box-sizing:border-box;border:1px solid ${border};border-radius:14px;box-shadow:${shadow}">
</div></body></html>`);
const png = await page.locator("#stage").screenshot({ omitBackground: true });
await context.close();
return png;
await server?.close?.();
}
/**
+28 -4
View File
@@ -18,11 +18,35 @@ trap 'rm -rf "$work"' EXIT
if ! curl -fsS "$base/reports.json" -o "$work/listing.json"; then
echo '[{"platform":"ios"}]' > "$work/listing.json"
fi
for platform in $(jq -r '.[].platform' "$work/listing.json"); do
curl -fsS "$base/report.zip?platform=$platform" -o "$work/$platform.zip"
unzip -tq "$work/$platform.zip" > /dev/null
# One ZIP per target, named for it: <target>.zip, or <platform>.zip for a report from before targets. A target named
# after its platform is selected with ?platform=, any other with ?target= (as viewer-links.ts targetParameter does).
# A preview deployed before Explore wrote static files serves each target's ZIP; a newer one serves its manifest and
# screenshots, so the ZIP is put back together here. Screenshot requests keep the target's query, which deployments
# that answered them from the ZIP need and static files ignore.
for key in $(jq -r '.[] | .target // .platform' "$work/listing.json"); do
# The listing comes from the hosted preview: a key names files here, so it must be a target id.
[[ "$key" =~ ^[a-z][a-z0-9-]{0,31}$ ]] || { echo "Not a target id in reports.json: $key" >&2; exit 2; }
case "$key" in ios|android) query="platform=$key" ;; *) query="target=$key" ;; esac
if curl -fsS "$base/report.zip?$query" -o "$work/$key.zip" 2> /dev/null; then
unzip -tq "$work/$key.zip" > /dev/null
continue
fi
report="$work/$key"
mkdir -p "$report/assets"
curl -fsS "$base/manifest.json?$query" -o "$work/$key.manifest.json"
file="$(jq -r .file "$work/$key.manifest.json")"
[[ "$file" == review.json || "$file" == exploration.json ]] || { echo "Not a report manifest: $key" >&2; exit 1; }
jq -c .report "$work/$key.manifest.json" > "$report/$file"
for asset in $(jq -r '.entries | keys[]' "$work/$key.manifest.json"); do
[[ "$asset" =~ ^assets/[a-f0-9]{64}\.(png|jpg|mp4)$ ]] || { echo "Unexpected asset name: $asset" >&2; exit 1; }
curl -fsS "$base/$asset?$query" -o "$report/$asset"
[[ "$(shasum -a 256 "$report/$asset" | cut -c1-64)" == "${asset:7:64}" ]] || { echo "Checksum mismatch: $asset" >&2; exit 1; }
done
rm -f "$work/$key.zip"
(cd "$report" && zip -qrX "$work/$key.zip" "$file" assets)
unzip -tq "$work/$key.zip" > /dev/null
done
jq --arg source "$source_name" '[.[] | {platform, changes, device, source: $source} | with_entries(select(.value != null))]' "$work/listing.json" > "$work/reports.json"
jq --arg source "$source_name" '[.[] | {target, platform, label, changes, device, source: $source} | with_entries(select(.value != null))]' "$work/listing.json" > "$work/reports.json"
gh release view "$tag" > /dev/null 2>&1 || gh release create "$tag" --prerelease --latest=false --title "Viewer preview fixtures" \
--notes "Reports the Viewer preview workflow serves on every pull request. Refresh with scripts/refresh-viewer-preview-fixtures.sh."
+12 -8
View File
@@ -1,6 +1,6 @@
---
name: explore-binary
description: Explore a prebuilt mobile app archive (Android .apk, iOS simulator .app, .tar.gz, or simulator .ipa) with Expo Explore. Runs `expo-explore explore:binary`, which installs the archive in a fresh EAS Simulator session, drives it with the local claude or codex login, writes an exploration map and report, hosts the report on EAS Hosting, and prints the link. Use when asked to "run @expo/explore for some.app/.apk", to map or smoke-test an app you only have as a build, or to get screenshots of every screen of a binary. `expo-explore explore:session` does the exploration half alone against an EAS Simulator session you already started (app already installed; result opens in the local Viewer). Also explores an Expo project from its source with no new build: its development build (or Expo Go) on an EAS Simulator, loading JavaScript from a tunneled local dev server. Not for device .ipa files.
description: Explore a prebuilt mobile app archive (Android .apk, iOS simulator .app, .tar.gz, or simulator .ipa) with Expo Explore. Runs `expo-explore explore:binary`, which installs the archive in a fresh EAS Simulator session (or, with `--simulator local`, on a simulator or emulator it starts on this Mac, with no EAS login), drives it with the local claude or codex login, writes an exploration map and report, and hosts the report on EAS Hosting or opens it in the local Viewer. Use when asked to "run @expo/explore for some.app/.apk", to map or smoke-test an app you only have as a build, or to get screenshots of every screen of a binary. `expo-explore explore:session` does the exploration half alone against an EAS Simulator session you already started (app already installed; result opens in the local Viewer). Also explores an Expo project from its source with no new build: its development build (or Expo Go) on an EAS Simulator, loading JavaScript from a tunneled local dev server. Not for device .ipa files.
---
# Explore a prebuilt app
@@ -11,28 +11,32 @@ One command, run from any directory:
npx @expo/explore explore:binary ./path/to/app.apk --json
```
Accepted inputs: Android `.apk`; iOS simulator builds as a `.app` directory, a `.tar.gz`/`.tgz` of one, or an `.ipa` that contains a simulator build. An iOS device `.ipa` and an Android `.aab` are rejected before any session starts (`ARCHIVE_NOT_SIMULATOR`, `ARCHIVE_AAB_UNSUPPORTED`); ask for a simulator build or a universal/x86_64 APK instead.
Accepted inputs: Android `.apk`; iOS simulator builds as a `.app` directory, a `.tar.gz`/`.tgz` of one, or an `.ipa` that contains a simulator build. An iOS device `.ipa` and an Android `.aab` are rejected before any device starts (`ARCHIVE_NOT_SIMULATOR`, `ARCHIVE_AAB_UNSUPPORTED`); ask for a simulator build or a universal APK instead.
Platform, bundle id, app name and version are read from the archive; do not pass them unless Explore asks.
## Before running
- EAS login: `eas whoami` succeeds, or `EXPO_TOKEN` is set. The session and the hosted report live under that account (`--owner` picks one when the login has several, `--project` renames the default `expo-explore` project).
- Where it runs. By default on EAS Simulator, which needs an EAS login: `eas whoami` succeeds, or `EXPO_TOKEN` is set. The session and the hosted report live under that account (`--owner` picks one when the login has several, `--project` renames the default `expo-explore` project). Without EAS Simulator access, or without an Expo account, pass `--simulator local`: Explore creates a simulator (iOS, needs Xcode) or an emulator (Android, needs the Android SDK with platform-tools, cmdline-tools, the emulator and a system image for this Mac, and `ANDROID_HOME`) and removes it afterwards. An APK must then ship native libraries for this Mac's ABI (`arm64-v8a` on Apple silicon); a universal one does.
- A model: Explore tries `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` when set, then the local `claude` login (Claude Code), then the local `codex` login, and uses the first that passes a preflight request. `--provider` or `--model` limits it to that one choice. If none passes, the run stops before a simulator is started.
- Node 24.12 or newer.
## What happens
1. Archive inspection and compatibility checks (size, simulator build, emulator ABI, sha256).
2. `eas simulator:start` of an empty agent-device session (an `argent` one with `--controller argent`), verified by exact session id before use.
3. `agent-device install` of the archive straight into the session (Argent: `reinstall-app`), then `open <bundle id>`.
2. `eas simulator:start` of an empty agent-device session (an `argent` one with `--controller argent`), verified by exact session id before use. With `--simulator local`: a new simulator or emulator on this Mac instead, with the app's URL schemes approved so its links open without a prompt.
3. `agent-device install` of the archive straight into the session (Argent: `reinstall-app`; locally `simctl install` or `adb install`), then `open <bundle id>`.
4. The model explores through a small tool set (observe, act, capture, finish); app text is treated as untrusted and destructive controls (delete, reset, sign out, purchase, send, share) are blocked outside the model.
5. The map and screenshots are written under `--output` (default `./explore-<archive>-<timestamp>`), plus `report/` for the Viewer.
6. The report is sealed and deployed to EAS Hosting (`--no-upload` skips this; `--expires-in 7d|30d|never`).
7. The exact session is stopped; a failure to stop reports `SESSION_CLEANUP_FAILED` with the id to stop by hand.
6. The report is deployed to EAS Hosting as files under the link's token (`--no-upload` skips this); the link lasts as long as the deployment. With `--simulator local` it opens in the local Viewer instead, which keeps running until stopped; `--json` and `--no-open` skip it, so with `--json` open it afterwards with `view --dir <output>` (in the background, since it serves until stopped). `--upload` hosts it, and that needs an EAS login.
7. The exact session is stopped, or the local device deleted; a failure to stop reports `SESSION_CLEANUP_FAILED` with the command to stop it by hand. A local device a hard-stopped run left behind is removed by the next local run.
Budgets default to 15 minutes, 40 screens, 150 actions (`--max-duration-minutes`, `--max-screens`, `--max-actions`). Steer the model with `--hints "focus on the onboarding and settings"`, skip onboarding with `--launch-url myapp://home`, and allow a specific risky control with `--allow-destructive "Clear cache"`.
## The same steps on every run: `--replay`
To compare runs, or an iPhone with an Android phone, screen for screen, pass `--replay steps.json` to `explore:binary` or `explore:session`: Explore takes the steps in the file instead of a model's, and no model access is needed. The format is in the README ("Take the same steps on every run and device"): `{ "steps": [{ "capture": "Home" }, { "action": "tap", "label": "Settings" }, { "capture": "Settings" }] }`, with `platforms` to limit a step and `optional` for an action that may fail. The plan is checked before a device starts: unknown fields, a plan with no steps for the platform, more captures or actions than `--max-screens` / `--max-actions`, and a `{{secret:NAME}}` in a `fill` or `type` whose variable is not set are all refused (`REPLAY_PLAN_INVALID`, `REPLAY_PLAN_EMPTY`, `REPLAY_PLAN_OVER_BUDGET`, `EXPLORATION_SECRET_MISSING`). `--provider`, `--model` and `--hints` cannot be combined with it (`REPLAY_OPTIONS_IGNORED`).
## Sign-in and other app state
Without help, the model treats a sign-in wall as the end of the road: it captures it and stops (`blocked`). Before running, read what you can about the app (its source if you have it, its first screens otherwise) and decide what it needs.
@@ -111,7 +115,7 @@ With `--json` the last stdout line is the result:
```json
{ "ok": true, "app": { "name": "euxy", "bundleId": "dev.brent.euxy", "platform": "ios", "version": "1.2.0" },
"exploration": { "stopReason": "finished", "screens": 9, "observations": 11, "actions": 23, "summary": "..." },
"map": ".../explore-euxy-.../map.json", "report": ".../explore-euxy-.../report", "url": "https://....expo.app/p/<token>", "expiresAt": "..." }
"map": ".../explore-euxy-.../map.json", "report": ".../explore-euxy-.../report", "url": "https://....expo.app/p/<token>", "expiresAt": null }
```
Share `url` (it is the only copy of the access token; it is not written to disk). Then query the map locally without a model or device:
+9 -8
View File
@@ -5,7 +5,7 @@ description: Set up Expo Explore CI reviews in deterministic or agent mode. Star
# Set up Expo Explore
Make the app ready for CI capture in the user's chosen mode, using its real configuration and source. This is the shared **local authoring skill** for both modes. Android capture is not available yet.
Make the app ready for CI capture in the user's chosen mode, using its real configuration and source. This is the shared **local authoring skill** for both modes.
Explore's tools are app-agnostic: `expo-explore setup` and `expo-explore doctor` know routes, parameters, links and source redirects, but not what the app means by them. **Your job is the app-specific part**: read this app's source to learn which flows (sign-in, onboarding, a subscription, a feature flag, a permission, a record that must exist, …) each screen needs, and write that into this app's `config.yaml` or instructions. Look for what the app actually has; do not assume the flows of another app, and do not impose an account or persona schema.
@@ -164,18 +164,19 @@ npx @expo/explore capture --base v1.3.0 --head v1.4.0 --scope all
- It runs both commits' code as the user (dependency installs, the app config, bundling). Before reviewing a branch the user did not write, say so and ask; credential-free installs are not a sandbox.
- Logins: `EXPO_TOKEN` and the provider's API key when they are set; otherwise the user's `eas login` and their local `claude` or `codex` login for the configured provider. If `eas whoami` fails, tell them to run `eas login`.
- Without `--head`, the head is HEAD plus the working tree's changes, untracked non-ignored files included, which is what `eas build` uploads. It lists them; check that list for secrets the user would not want built. `--committed` reviews HEAD alone, and an `eas.json` with `cli.requireCommit: true` leaves the changes out unless `--uncommitted` is passed.
- Every configured platform is reviewed unless `--platform` picks one. It opens the report in a local Viewer; `--upload` hosts it instead and prints the link, `--no-open` or `--json` only prints the report paths.
- Every configured platform or target is reviewed, at once, unless `--target` or `--platform` picks some (`--parallel N` caps how many run together). It opens the report in a local Viewer; `--upload` hosts it instead and prints the link, `--no-open` or `--json` only prints the report paths.
- Without EAS Simulator access, add `--simulator local` (or `capture.simulator: local`): each commit is captured on a simulator Explore creates on this Mac (Xcode) or an emulator (Android SDK), and the app is repacked there too. The native build still comes from EAS. `npx @expo/explore doctor --simulator local` checks that this machine can start each device.
- Every run is kept on this machine for 30 days: `npx @expo/explore reports list` (or `--json`) shows them, `reports open <id|latest>` reopens one in the local Viewer, and `reports upload <id>` hosts it without capturing again. Run bare in a terminal, `reports` is an arrow-key picker for the user.
- To host a report after the fact, run `npx @expo/explore report:upload <report.zip>...` with the paths it printed (or the `--output` directory's `<platform>/report.zip`), one per platform. It uses the same project as `--upload` and does not capture again.
- Captures are cached in `~/.cache/expo-explore`, so a second review of the same base does not capture it again (`--no-cache` skips the cache). Ctrl+C stops the EAS Simulator sessions the review started.
- It builds on EAS and runs simulator sessions, like a CI review: expect several minutes and EAS usage per run.
- To host a report after the fact, run `npx @expo/explore report:upload <report.zip>...` with the paths it printed (or the `--output` directory's `<target>/report.zip`), one per platform or target. It uses the same project as `--upload` and does not capture again.
- Captures are cached in `~/.cache/expo-explore`, so a second review of the same base does not capture it again (`--no-cache` skips the cache). Ctrl+C stops the simulator sessions and local devices the review started; after a harder stop, the next local review stops what was left running.
- It builds on EAS and, unless it captures locally, runs EAS Simulator sessions, like a CI review: expect several minutes and EAS usage per run.
## Compare any two commits
`explore-compare.yml` (installed by `setup --workflows` from 0.19.0; run `npx @expo/explore@latest setup --workflows` to add it to an existing setup) reviews a `head` against a `base`, each a branch, tag or commit SHA, with no pull request. It captures, compares and reports like a PR review and links the report from the run's summary. Its inputs are `base`, `head`, and the optional `scope` (`changes` or `all`) and `expires-in` (such as `30d`, or `never`).
`explore-compare.yml` (installed by `setup --workflows` from 0.19.0; run `npx @expo/explore@latest setup --workflows` to add it to an existing setup) reviews a `head` against a `base`, each a branch, tag or commit SHA, with no pull request. It captures, compares and reports like a PR review and links the report from the run's summary. Its inputs are `base`, `head`, and the optional `scope` (`changes` or `all`); `expires-in` from earlier versions is accepted and ignored, since a hosted link lasts as long as its EAS Hosting deployment.
- **Releases.** Unless the user asks otherwise, compare each release with the last stable release, with `scope: all` (every screen, so the report shows the whole release with what changed marked) and, when hosted previews are on, `expires-in: never` (the link outlives the 30-day artifact). For every tag as it is pushed, add the `release-review.yml` from the README's "Compare any two commits" section: it finds the newest earlier tag without a `-` (so a release candidate is compared with the last stable release) and calls `explore-compare.yml` with `secrets: inherit` and `permissions: { contents: read, actions: read }`. A nightly compares with yesterday's tag; a release branch with `main`.
- **Once, on demand.** The workflow must be on the default branch to be dispatched: `gh workflow run explore-compare.yml -f base=<ref> -f head=<ref> -f scope=all -f expires-in=never`, or **Run workflow** under Actions → Expo Explore compare.
- **Releases.** Unless the user asks otherwise, compare each release with the last stable release, with `scope: all` (every screen, so the report shows the whole release with what changed marked). A hosted link lasts as long as its EAS Hosting deployment (30 days on Free, 90 on paid plans, longer with an alias), which can outlive the 30-day artifact. For every tag as it is pushed, add the `release-review.yml` from the README's "Compare any two commits" section: it finds the newest earlier tag without a `-` (so a release candidate is compared with the last stable release) and calls `explore-compare.yml` with `secrets: inherit` and `permissions: { contents: read, actions: read }`. A nightly compares with yesterday's tag; a release branch with `main`.
- **Once, on demand.** The workflow must be on the default branch to be dispatched: `gh workflow run explore-compare.yml -f base=<ref> -f head=<ref> -f scope=all`, or **Run workflow** under Actions → Expo Explore compare.
- **Without CI.** `npx @expo/explore capture --base <ref> --head <ref> --scope all --upload` runs the same comparison on this machine (see the previous section).
- Call the workflow once per workflow run (a second call in the same run reuses the first one's artifact names). A `base` the baseline workflow captured in the last 30 days is reused, so only `head` is captured.
- To verify, run it once on the latest tag and open the run's summary: it lists each platform's result and links the report.
+23 -15
View File
@@ -1,6 +1,6 @@
---
name: readme-refresh
description: Refresh Expo Explore's README.md with euxy (github.com/brentvatne/euxy) as the reference app. Regenerate every screenshot in docs/screenshots from a current euxy review report and PR comment, then re-read the README as a first-time user: the top level only says what Explore does, links the live euxy example, gives numbered quick-start steps for reviewing a change to your own app, a prompt to copy for exploring any build, and one for setting up CI; everything else lives in collapsibles. Use after a Viewer or PR-comment change that the README shows, a new command or setting, or a release that made a README claim stale. Not for the docs/ guides or the shipped skills.
description: Refresh Expo Explore's README.md with euxy (github.com/brentvatne/euxy) as the reference app. Regenerate every screenshot and animated SVG in docs/screenshots from a current euxy review report and PR comment, then re-read the README as a first-time user: the top level only says what Explore does, links the live euxy example, gives one quick-start prompt to copy that reviews a change to your own app or explores a build you point it to, and one for setting up CI; everything else lives in collapsibles. Use after a Viewer or PR-comment change that the README shows, a new command or setting, or a release that made a README claim stale. Not for the docs/ guides or the shipped skills.
---
# Refresh the README
@@ -12,8 +12,8 @@ This is a repository-maintenance skill (it is not shipped in the npm package; `s
The README's screenshots and its live example (`https://explore-example--euxy.expo.app/p/<token>`, deployed by `.github/workflows/example-report.yml`) must show the same report. The example serves the `viewer-preview-fixtures` release, the same reports every pull request's Viewer preview uses, so start there:
```sh
gh release download viewer-preview-fixtures --pattern reports.json --output - | jq -r '.[0].source' # e.g. brentvatne/euxy#154
mkdir -p ~/euxy-report && gh release download viewer-preview-fixtures --pattern ios.zip --dir ~/euxy-report --clobber
gh release download viewer-preview-fixtures --pattern reports.json --output - | jq -r '.[0] | "\(.target) \(.source)"' # e.g. iphone brentvatne/euxy#163
mkdir -p ~/euxy-report && gh release download viewer-preview-fixtures --pattern iphone.zip --dir ~/euxy-report --clobber
```
Keep those fixtures when their euxy PR still shows a real change (an added or removed screen, modified screens with outlined regions). To move to a newer euxy PR, open its Explore comment and follow **iOS report** to the hosted link `https://euxy-explore--<id>.expo.app/p/<token>`. Then refresh the fixtures from that link and download them again:
@@ -24,17 +24,25 @@ scripts/refresh-viewer-preview-fixtures.sh '<hosted link>' brentvatne/euxy#<n>
Do not pick a same-tree demo PR, where every change is noise. The refreshed fixtures reach the live example on the next push to main, or when you run **Example report** from GitHub Actions. `curl -s "<hosted link>/api/v1/screens?status=added,removed,modified&platform=ios"` lists what changed, which tells you which screen to select and compare.
## 2. Regenerate every screenshot
## 2. Regenerate every screenshot and animation
```sh
npm ci
node scripts/readme-screenshots.mjs --report ~/euxy-report/ios.zip --select /settings --compare patterns --pr https://github.com/brentvatne/euxy/pull/<n>
node scripts/readme-screenshots.mjs --report ~/euxy-report/iphone.zip --select /lane-editor/0 --compare new-pattern --pr https://github.com/brentvatne/euxy/pull/<n>
```
- `--select` is the route selected on the hero Map (the Inspector opens on it): choose an added or clearly modified screen.
- `--compare` is the key in the Viewer's `#/compare/<key>` URL for a **modified** screen with outlined regions (open the report, click the screen in Compare, copy the key). The script fails if the key is not in the report.
- `--pr` screenshots the Explore comment on that PR: the fixtures' `source`. Headless Chromium needs a colour emoji font for the result glyphs (`sudo apt-get install fonts-noto-color-emoji` on Debian/Ubuntu); a comment with hollow boxes where 🟩 should be is not acceptable.
- The script serves the report through the Viewer's Vite dev server, so no build is needed. It writes a `-light` and a `-dark` PNG for every shot (the PR comment too: signed-out github.com follows `prefers-color-scheme`). The README shows each pair with `<picture>`: a `<source media="(prefers-color-scheme: dark)">` for the dark file and an `<img>` with the alt text and the light file, which is what GitHub Mobile shows (it ignores the dark source), so the light file is the one that must always be right.
- `--select` is the route the hero's pointer clicks on the Map (the Inspector opens on it): choose an added or clearly modified screen. The script zooms in around it, so it must be a screen the Map shows.
- `--compare` is the key in the Viewer's `#/compare/<key>` URL for a **modified** screen with outlined regions (open the report, click the screen in Compare, copy the key): the route without its slash. The script fails if the report has no visually modified screen at that route.
- `--pr` screenshots the Explore comment on that PR: the fixtures' `source`, or a newer euxy PR when that comment predates the current layout. `--only pr-comment --pr <url>` rewrites only the comment pictures (it needs no `--report`); `--only` takes any of `hero`, `compare`, `screens` and `pr-comment`. Headless Chromium needs a colour emoji font for the result glyphs (`sudo apt-get install fonts-noto-color-emoji` on Debian/Ubuntu); a comment with hollow boxes where 🟩 should be is not acceptable.
- The script serves the report through the Viewer's Vite dev server, so no build is needed. It writes a `-light` and a `-dark` file for every picture (the PR comment too: signed-out github.com follows `prefers-color-scheme`). The README shows each pair with `<picture>`: a `<source media="(prefers-color-scheme: dark)">` for the dark file and an `<img>` with the alt text and the light file, which is what GitHub Mobile shows (it ignores the dark source), so the light file is the one that must always be right. The Screens and PR comment pictures, shown at half the column, are each wrapped in an `<a>` to their light PNG so a reader can open them full size; GitHub links a plain Markdown image itself, but not a `<picture>`.
Two of the pictures are animated SVGs, built by `scripts/readme-animations.mjs` in the same run; there is nothing else to run when the screenshots change:
- `readme-hero-*.svg` is made of real Viewer frames at the zoomed Map camera: the Map without screens, the screens without edges, the whole Map, and the Map after the pointer clicks `--select`. The cards appear row by row, the edges wipe in from the top, then the pointer clicks and the Inspector slides in. Each is about 1.4 MB: four 2x JPEG frames inlined, since an SVG shown with `<img>` can load nothing. Anything the Viewer draws on the Map or in the Inspector shows up here on the next run, so regenerate it after a Viewer change the hero shows.
- `readme-compare-*.svg` is the `--compare` screen's base capture wiping to its head capture, with the report's changed regions pulsing. It is drawn from `review.json` and the report's assets, not from a Viewer screenshot, so a Viewer change does not stale it, but a new report does.
- Both loop on CSS keyframes, and with `prefers-reduced-motion` they hold the last frame: Inspector open, regions outlined. That frame is also what a viewer that does not animate SVG shows, so it must read on its own.
Open each SVG in a browser before committing, light and dark, and watch a full loop: every card should appear once and land where the whole Map has it, with no hover highlight in the frames before the click; the pointer should click the `--select` card; the Inspector should open on that screen; Compare's outlines should sit on the changed text, not beside it. To look at one moment, pause the animations in the console: `document.getAnimations().forEach(a => { a.pause(); a.currentTime = 5000; })`. Keep the README's `<img>` and `<source>` pointing at the `.svg` files.
`docs/screenshots/readme-exploration-light.png` is the black-box map and comes from a euxy `explore:binary` run (`npx @expo/explore explore:binary <euxy simulator build> --no-upload`, then `expo-explore view --dir <output>` at 1760×1000, Map, Fit canvas). It has no dark variant and is a plain Markdown image. Regenerate it when the Map's rendering of edges or labels changed; otherwise keep it, and say so in the PR.
@@ -45,7 +53,7 @@ npx @expo/explore reports list # your sa
export EXPLORE_CACHE_DIR=$(mktemp -d) # a scratch store for the image, not your own runs
cd ~/code/euxy # inside euxy, so the commit messages are found
npx @expo/explore reports add ~/.cache/expo-explore/reports/<that id>/ios.zip
npx @expo/explore reports add ~/euxy-report/ios.zip # the PR report from step 1
npx @expo/explore reports add ~/euxy-report/iphone.zip # the PR report from step 1
npx @expo/explore reports add <an explore:binary report directory> # an exploration, so all three kinds show
cd - && npm run build:cli && node scripts/readme-reports-rows.mjs > /tmp/reports-rows.html
```
@@ -53,18 +61,18 @@ cd - && npm run build:cli && node scripts/readme-reports-rows.mjs > /tmp/reports
- The script renders the picker as the CLI draws it (96 columns, the newest run selected) and writes the terminal body as HTML: each coloured piece is exactly its characters wide, so columns line up, and the wordmark is drawn as cell rectangles, since a design tool's font does not draw `█▀▄` cell for cell.
- In Paper, replace the frame named "Terminal" inside each artboard's window with that HTML, check both artboards in a screenshot (crisp wordmark, aligned columns, the selected row highlighted to the end, coloured counts), and export both at 2x as `readme-reports-light.png` and `readme-reports-dark.png`. The window's title bar reads `npx @expo/explore reports`, as people run it.
Look at every PNG you produced before committing, light and dark: the selected screen should be the one you asked for, the Compare view must show Before and After with outlines, the Screens grid must show the added, removed and modified status dots, the PR comment must show the heading links and result glyphs. Describe each in its Markdown alt text as what the picture shows, not what the feature is.
Look at every PNG you produced before committing, light and dark: the Screens grid must show the added, removed and modified status dots, the PR comment must show the heading links, the line of device links and the list of changed screens. Describe each in its Markdown alt text as what the picture shows, not what the feature is.
## 3. Pass over the content as a first-time user
Read the README top to bottom once without opening a collapsible. At that level it may only answer, in this order:
1. What Explore is (one short paragraph), the hero image, and the link to the live euxy example (`explore-example--euxy.expo.app`, redeployed on every merge to main by `.github/workflows/example-report.yml`). Under it, one table row of three pictures with a one-line caption each: Compare, Screens, and the PR comment (which says it needs CI), then the agent prompt collapsed as **Ask an agent about a report**. The hero is a zoomed Map framed as a window on a stage (`stagedWindow` in the script); the three below are plain.
2. **Quick start**: the smallest set of steps to see a review of the reader's own Expo app on their machine, numbered and with nothing optional in them: make the app buildable on EAS (login, init, bundle identifier, a simulator profile named `explore`) and commit (the before build is the last commit), then edit a screen near its top and run `capture`, which reviews uncommitted changes as `eas build` would upload them and uses setup's defaults when there is no config. After the steps, a prompt that has Claude Code or Codex do the same through the explore-setup skill's "Review a change on this machine" section. Then the `reports` picker image (`readme-reports-*.png`) with one sentence on saved runs, at the top level, and one collapsible for other commits (`--committed`, `--base`/`--head`) and what a review does on the machine. Run the steps on a fresh `create-expo-app` before changing them.
3. **Map every screen of any app**: one sentence and a prompt to paste into Claude Code or Codex that gets a simulator build of any app (Expo or not, or just a build) and runs `explore:binary` on it. Each section keeps its own "Run it yourself"; never put one section between another's prompt and its collapsibles.
1. What Explore is (one short paragraph), the hero image, and the link to the live euxy example (`explore-example--euxy.expo.app`, redeployed on every merge to main by `.github/workflows/example-report.yml`). Under it, Compare at full width with a one-line caption, then one table row of two pictures with a one-line caption each: Screens and the PR comment (which says it needs CI), then the agent prompt collapsed as **Ask an agent about a report**. The hero is the animated zoomed Map, framed as a window on a stage, and Compare is the animated wipe (both from `readme-animations.mjs`); the Screens and PR comment pictures are PNGs. Compare gets its own row because a phone wiping from before to after is unreadable at a third of the column.
2. **Quick start**: one path. A single prompt to paste into Claude Code or Codex that reviews a change to the reader's own Expo app through the explore-setup skill's "Review a change on this machine" section, or, when they point it to a build (or the app is not an Expo project), explores a simulator build through the explore-binary skill. Then one sentence on what each case needs (a review needs an Expo account, is built on EAS, and is captured on EAS Simulator or a local simulator; exploring a build needs a Claude Code or Codex login and EAS Simulator or a local simulator, and no Expo account locally), and under it one `> [!NOTE]` alert: EAS Simulator is in preview, with the waitlist link and `--simulator local` as the alternative. Then one **Run it yourself** collapsible with both by hand: the numbered steps for a review, with nothing optional in them (make the app buildable on EAS: login, init, bundle identifier, a simulator profile named `explore`, and commit, since the before build is the last commit; then edit a screen near its top and run `capture`, or `capture --simulator local`), and the `explore:binary` commands, including `--simulator local`. Right after **Run it yourself**, one `> [!WARNING]` alert: a review runs both commits' code on the reader's machine, so review only commits they would run. These are the README's only two alerts; do not add more, or they stop standing out. Then the `reports` picker image (`readme-reports-*.png`) with one sentence on saved runs, at the top level, and one collapsible for other commits (`--committed`, `--base`/`--head`) and what a review does on the machine. Run the steps on a fresh `create-expo-app` before changing them.
3. **Map every screen of any app**: one sentence on what exploring a build shows (pointing back to the quick start, which runs it), then its collapsibles: its options, exploring a session already started, and what a run without source shows.
4. **Set up CI**: one sentence, then two subsections, each a sentence on what it gives and a prompt to paste: **Review on every PR** (the setup prompt, then a few grouped collapsibles: Hosted previews first, then Manual setup (the setup prompt's steps done by hand, in order: setup, doctor, secrets, baseline, first PR), Capture modes: deterministic or agent, iOS, Android, or both, Keep your setup current) and **Compare any two commits** (a prompt that follows the explore-setup skill's "Compare any two commits" section, the ways to snapshot a release as bullets, then the comparison workflow collapsed).
Lead with prompts, not explanations: a reader should be able to copy something and see Explore work. Everything else, the manual commands, requirements, options, the manual setup, capture modes, hosted previews, iOS and Android, agent mode, repository settings, how agents read a report, how it works, limitations, contributing, goes into a `<details>` block under the section it belongs to, with a `<summary>` that says what the reader will find. Keep bullets to one or two sentences. Group related collapsibles rather than adding one per topic, so a section shows a handful, not nine. Requirements sit collapsed right under the example report link. Keep the collapsibles' `id` attributes that other docs link to (`#hosted-previews`, `#platforms`, `#compare`, `#agents`); the Terms and How agents read a report collapsibles live under Additional information, and keep the Inspiration section (screenmap and Revyl's Atlas).
Lead with prompts, not explanations: a reader should be able to copy something and see Explore work. Everything else, the manual commands, requirements, options, the manual setup, capture modes, hosted previews, iOS and Android, agent mode, repository settings, how agents read a report, how it works, limitations, contributing, goes into a `<details>` block under the section it belongs to, with a `<summary>` that says what the reader will find. Keep bullets to one or two sentences. Group related collapsibles rather than adding one per topic, so a section shows a handful, not nine. Requirements sit collapsed right under the example report link. The **How it works** collapsible opens with a Mermaid `sequenceDiagram` of a PR review (pull request, GitHub Actions, EAS Build, EAS Simulator), followed by the facts it cannot show as bullets: who triggers a review, what Modified means, session retries. Check its steps against `CI-FLOW.md` and the `ci:*` commands, and edit the diagram's text, not a picture: GitHub draws it. Keep the collapsibles' `id` attributes that other docs link to (`#hosted-previews`, `#platforms`, `#compare`, `#agents`); the Terms and How agents read a report collapsibles live under Additional information, and keep the Inspiration section (screenmap and Revyl's Atlas).
Check that each prompt still works as written:
- the skill links it names (`https://unpkg.com/@expo/explore/skills/...`) are files the package ships (`files` in `package.json`), and they open from the published release;
+6 -3
View File
@@ -164,10 +164,13 @@ export async function readArchiveIcon(archive: InspectedArchive): Promise<Buffer
}
}
/** Explore's EAS Android emulator is x86_64 with no 32-bit support; an APK that ships only ARM or x86 libraries fails to install there. */
/**
* Explore's EAS Android emulator is x86_64 with no 32-bit support, and a local one runs this Mac's ABI; an APK that
* ships native libraries only for other ABIs fails to install there.
*/
export function assertEmulatorCompatible(archive: InspectedArchive, emulatorAbis: readonly string[] = ["x86_64"]): void {
if (archive.platform === "ios") {
if (!archive.simulator) throw archiveError("ARCHIVE_NOT_SIMULATOR", `${basename(archive.path)} is an iOS device build; it cannot run on the EAS Simulator.`,
if (!archive.simulator) throw archiveError("ARCHIVE_NOT_SIMULATOR", `${basename(archive.path)} is an iOS device build; it cannot run on a simulator.`,
"Build for the simulator (an EAS profile with ios.simulator: true) and pass that .app, .tar.gz, or simulator .ipa.");
return;
}
@@ -177,7 +180,7 @@ export function assertEmulatorCompatible(archive: InspectedArchive, emulatorAbis
}
if (archive.abis.length && !archive.abis.some(abi => emulatorAbis.includes(abi))) {
throw archiveError("ARCHIVE_ABI_UNSUPPORTED", `The APK ships native libraries only for ${archive.abis.join(", ")}; the emulator needs ${emulatorAbis.join(" or ")}.`,
"Build a universal APK or one that includes x86_64 native libraries.", { abis: archive.abis, emulatorAbis: [...emulatorAbis] });
`Build a universal APK or one that includes ${emulatorAbis[0]} native libraries.`, { abis: archive.abis, emulatorAbis: [...emulatorAbis] });
}
}
+59 -27
View File
@@ -7,6 +7,7 @@ import type { CaptureSetupAction } from "../config/explore-config.js";
import { createExplorationInputs, explorationSecretNames } from "../config/exploration-inputs.js";
import type { ExploreStore, PreparedArtifact } from "../map/store.js";
import { CiError } from "../shared/ci-errors.js";
import { turnedClockwise } from "../shared/png-evidence.js";
import type { ActionKind, ActionRisk, Artifact, Controller, ElementRole, ExploreMap, ObservedElement, Observation, Screen } from "../types.js";
/** One look at the device: accessibility always, a PNG when the host could take a safe one. */
@@ -21,13 +22,15 @@ export type BlackBoxFrame = {
imageUnavailable?: { code: string; message: string };
/** The foreground app differed from the explored one; the host already brought it back. */
leftApp?: boolean;
/** The image is on a portrait panel's axes under a landscape UI (an opened iPhone Duo), as the tree is; the report stores it turned upright. */
landscape?: boolean;
};
export type BlackBoxAction = CaptureSetupAction | { action: "back" } | { action: "dismiss-keyboard" } | { action: "type"; text: string }
| { action: "tap"; frameId: string; x: number; y: number };
export type BlackBoxExplorerOptions = {
session: CiAgentSession;
session: Pick<CiAgentSession, "run">;
store: ExploreStore;
map: ExploreMap;
app: { name: string; bundleId: string; platform: "ios" | "android"; version?: string };
@@ -49,6 +52,8 @@ export type BlackBoxExplorerOptions = {
appIcon?: Buffer;
/** App-test inputs the operator hints reference as {{secret:NAME}}: the model and the map only ever hold the reference. */
inputs?: ReturnType<typeof createExplorationInputs>;
/** A sentence a budget stop's gap adds to its own, read once the session has stopped: a replay names the planned screens it did not reach. */
budgetNote?: () => string | undefined;
};
export type BlackBoxExplorerResult = {
@@ -67,6 +72,12 @@ const EXTERNAL = /\b(open in|safari|chrome|browser|app store|play store|share|ma
const VOLATILE = /\d|[:%$€£]|\b(am|pm)\b/gi;
const INTERACTIVE = new Set(["button", "link", "cell", "switch", "tab", "menuitem", "textfield", "securetextfield", "searchfield", "textview", "edittext", "slider", "checkbox", "radiobutton", "segmentedcontrol", "toggle"]);
/** What the tools accept: their schemas, the checks on each call, and a replay plan checked before any device starts share these. */
export const TOOL_LIMITS = { selector: 512, text: 2000, url: 2000, waitMs: 15_000, title: 80, state: 80, description: 500, summary: 4000, reason: 1000 } as const;
/** A screen seen after an action but never named keeps this title until the run ends and names it after the path to it. */
export const isPlaceholderTitle = (title: string): boolean => /^Screen \d+$/.test(title);
const selectorProperties = {
label: { type: "string", description: "Exact visible label of the target node, copied from the latest observation." },
identifier: { type: "string", description: "Exact accessibility identifier of the target node, when it has one." },
@@ -81,22 +92,22 @@ const tools: CiAgentTool[] = [
inputSchema: { type: "object", properties: {
action: { type: "string", enum: ["tap", "fill", "type", "scroll", "wait", "back", "dismiss-keyboard", "open"] },
...selectorProperties,
text: { type: "string", maxLength: 2000 }, url: { type: "string", maxLength: 2000 },
text: { type: "string", maxLength: TOOL_LIMITS.text }, url: { type: "string", maxLength: TOOL_LIMITS.url },
direction: { type: "string", enum: ["up", "down", "left", "right", "top", "bottom"] },
milliseconds: { type: "integer", minimum: 1, maximum: 15_000 },
milliseconds: { type: "integer", minimum: 1, maximum: TOOL_LIMITS.waitMs },
frameId: { type: "string", description: "For a coordinate tap only: the latest observed frame." },
x: { type: "number", minimum: 0, exclusiveMaximum: 1 }, y: { type: "number", minimum: 0, exclusiveMaximum: 1 },
target: { type: "string", minLength: 1, maxLength: 80, description: "For a coordinate tap: what is being tapped, as a user would name it (\"search icon in the header\", \"first manga cover\")." },
}, required: ["action"], additionalProperties: false } },
{ name: "capture", description: "Name the latest observed frame (observe with includeImage) as a screen of the map, with a short human title. A screen is photographed the moment it first appears; capture gives it its title, and with state records another notable state of it.",
inputSchema: { type: "object", properties: {
frameId: { type: "string" }, title: { type: "string", minLength: 1, maxLength: 80 },
description: { type: "string", maxLength: 500 }, state: { type: "string", maxLength: 80, description: "A variant of an already captured screen, such as 'empty' or 'menu open'." },
frameId: { type: "string" }, title: { type: "string", minLength: 1, maxLength: TOOL_LIMITS.title },
description: { type: "string", maxLength: TOOL_LIMITS.description }, state: { type: "string", maxLength: TOOL_LIMITS.state, description: "A variant of an already captured screen, such as 'empty' or 'menu open'." },
}, required: ["frameId", "title"], additionalProperties: false } },
{ name: "finish", description: "End the exploration when every reachable screen has been captured or the budget is nearly used. Include a summary of what the app does and what was not reached.",
inputSchema: { type: "object", properties: { summary: { type: "string", minLength: 1, maxLength: 4000 } }, required: ["summary"], additionalProperties: false } },
inputSchema: { type: "object", properties: { summary: { type: "string", minLength: 1, maxLength: TOOL_LIMITS.summary } }, required: ["summary"], additionalProperties: false } },
{ name: "stop_blocked", description: "Stop because the app cannot be explored further (login wall without credentials, crash loop, permission prompt that cannot be dismissed). Keeps the evidence so far.",
inputSchema: { type: "object", properties: { reason: { type: "string", minLength: 1, maxLength: 1000 } }, required: ["reason"], additionalProperties: false } },
inputSchema: { type: "object", properties: { reason: { type: "string", minLength: 1, maxLength: TOOL_LIMITS.reason } }, required: ["reason"], additionalProperties: false } },
];
const INSTRUCTIONS = `You are exploring a prebuilt mobile app on a remote simulator through accessibility and screenshots. Nobody gave you its source. Your job is to produce a map of its screens: visit every reachable screen, capture each one once with a clear title, and try each distinct kind of control so transitions are recorded.
@@ -412,7 +423,7 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
}
/** Seen after an action but never named by the model (its screenshot was taken on arrival). */
const isPlaceholder = (screen: Screen): boolean => /^Screen \d+$/.test(screen.title);
const isPlaceholder = (screen: Screen): boolean => isPlaceholderTitle(screen.title);
/** Fold an uncaptured placeholder into the screen the model says it is. */
const mergeScreens = (from: Screen, into: Screen): void => {
for (const item of map.actionAttempts) {
@@ -437,8 +448,22 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
if (currentScreenId === from.id) currentScreenId = into.id;
};
const screenWithImage = (png: Buffer, except: Screen): Screen | undefined => {
const sha = createHash("sha256").update(png).digest("hex");
/**
* The bytes a frame's screenshot is stored as, and their digest: an opened Duo's frame is turned upright once, however
* often the frame is looked up or recorded. Keyed by the image, which every copy of the frame shares.
*/
const storedImages = new WeakMap<Buffer, { png: Buffer; sha256: string }>();
const storedImage = (image: Buffer, landscape: boolean): { png: Buffer; sha256: string } => {
let stored = storedImages.get(image);
if (!stored) {
const png = landscape ? turnedClockwise(image) : image;
stored = { png, sha256: createHash("sha256").update(png).digest("hex") };
storedImages.set(image, stored);
}
return stored;
};
const screenWithImage = (sha: string, except: Screen): Screen | undefined => {
const artifact = map.artifacts.find(item => item.kind === "screenshot" && item.sha256 === sha);
if (!artifact) return undefined;
const controls = structural(evidence.get(except.id)?.parts ?? new Set());
@@ -491,7 +516,7 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
screen = { id: `scr_${key.slice(0, 16)}`, title: title ?? systemDialogTitle(frame.nodes) ?? `Screen ${map.screens.length + 1}`, start: map.screens.length === 0, terminal: false,
depth, firstSeenAt: frame.capturedAt, lastSeenAt: frame.capturedAt, visits: 0, variantIds: [], observationIds: [], elementIds: [] };
map.screens.push(screen); screensByKey.set(key, screen);
} else if (title && /^Screen \d+$/.test(screen.title)) screen.title = title;
} else if (title && isPlaceholderTitle(screen.title)) screen.title = title;
screen.visits += 1; screen.lastSeenAt = frame.capturedAt;
const seen = evidence.get(screen.id) ?? { parts: new Set<string>(), loadingOnly: true };
for (const part of loose) seen.parts.add(part);
@@ -529,11 +554,13 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
let screenshotArtifactId: string | undefined;
if (withImage && frame.png) {
// Hashed here rather than trusting the daemon's digest: the report validates artifacts against their bytes.
const png = frame.png, sha = createHash("sha256").update(png).digest("hex");
const { png, sha256: sha } = storedImage(frame.png, frame.landscape === true);
const extent = frame.landscape ? frameExtent(frame.nodes) : undefined;
let artifact = map.artifacts.find(item => item.kind === "screenshot" && item.sha256 === sha);
if (!artifact) {
artifact = { id: `art_${randomUUID().slice(0, 12)}`, kind: "screenshot" as const, path: `artifacts/screenshots/${sha}.png`, sha256: sha,
bytes: png.length, mimeType: "image/png", ...(frame.width ? { width: frame.width } : {}), ...(frame.height ? { height: frame.height } : {}), createdAt: frame.capturedAt };
bytes: png.length, mimeType: "image/png", ...(frame.width && frame.height ? frame.landscape ? { width: frame.height, height: frame.width } : { width: frame.width, height: frame.height } : {}),
...(extent ? { turned: extent } : {}), createdAt: frame.capturedAt };
map.artifacts.push(artifact); prepared.push({ artifact, bytes: png, destination: join(store.root, artifact.path) });
}
screenshotArtifactId = artifact.id;
@@ -599,7 +626,7 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
// The same pixels as a screen already seen, with the same controls, is that screen: a tree read while Android was
// still exposing a text node (AudioScape's playlist title missing from one read of two identical frames) is not new.
if (created && latest.png) {
const twin = screenWithImage(latest.png, screen);
const twin = screenWithImage(storedImage(latest.png, latest.landscape === true).sha256, screen);
if (twin) { mergeScreens(screen, twin); screen = twin; created = false; }
}
// Every screen gets an observation the moment it is seen, so elements and action attempts always point at a
@@ -625,7 +652,7 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
const actionFor = (input: Record<string, unknown>): { action: BlackBoxAction; kind: ActionKind; label: string; risk: ActionRisk } => {
const action = String(input.action);
const selector = input.identifier !== undefined ? { identifier: text(input.identifier, "identifier", 512) } : input.label !== undefined ? { label: text(input.label, "label", 512) } : undefined;
const selector = input.identifier !== undefined ? { identifier: text(input.identifier, "identifier", TOOL_LIMITS.selector) } : input.label !== undefined ? { label: text(input.label, "label", TOOL_LIMITS.selector) } : undefined;
const label = selector ? ("identifier" in selector ? selector.identifier : selector.label) : action;
const risk = selector ? riskOf(label, allow) : "safe";
switch (action) {
@@ -637,13 +664,13 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
}
if (!selector) throw failure("AGENT_ACTION_INVALID", "tap needs a label, an identifier, or frame coordinates.");
return { action: { action: "tap", ...selector } as BlackBoxAction, kind: "press", label, risk };
case "fill": if (!selector) throw failure("AGENT_ACTION_INVALID", "fill needs a label or identifier."); return { action: { action: "fill", text: text(input.text, "text", 2000), ...selector } as BlackBoxAction, kind: "fill", label, risk };
case "type": return { action: { action: "type", text: text(input.text, "text", 2000) }, kind: "fill", label: "type", risk: "safe" };
case "fill": if (!selector) throw failure("AGENT_ACTION_INVALID", "fill needs a label or identifier."); return { action: { action: "fill", text: text(input.text, "text", TOOL_LIMITS.text), ...selector } as BlackBoxAction, kind: "fill", label, risk };
case "type": return { action: { action: "type", text: text(input.text, "text", TOOL_LIMITS.text) }, kind: "fill", label: "type", risk: "safe" };
case "scroll": return { action: { action: "scroll", direction: input.direction as "up" }, kind: "swipe", label: `scroll ${String(input.direction)}`, risk: "safe" };
case "wait": return { action: { action: "wait", milliseconds: Number(input.milliseconds) || 1000 }, kind: "other", label: "wait", risk: "safe" };
case "back": return { action: { action: "back" }, kind: "back", label: "back", risk: "safe" };
case "dismiss-keyboard": return { action: { action: "dismiss-keyboard" }, kind: "dismiss", label: "dismiss keyboard", risk: "safe" };
case "open": return { action: { action: "open", url: text(input.url, "url", 2000) }, kind: "deep-link", label: text(input.url, "url", 2000), risk: "unknown" };
case "open": return { action: { action: "open", url: text(input.url, "url", TOOL_LIMITS.url) }, kind: "deep-link", label: text(input.url, "url", TOOL_LIMITS.url), risk: "unknown" };
default: throw failure("AGENT_ACTION_INVALID", `Unknown action ${action}.`);
}
};
@@ -707,8 +734,12 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
const role = roleOf(under);
const label = str(under.label) ?? "";
const what = role === "field" ? "field" : role === "button" || role === "tab" || role === "link" ? (cleanLabel(label) ? role : `icon ${role}`) : /\p{L}/u.test(cleanLabel(label)) ? "item" : "icon";
const row = point.y < 1 / 3 ? "top" : point.y < 2 / 3 ? "" : "bottom";
const column = point.x < 1 / 3 ? "left" : point.x < 2 / 3 ? "" : "right";
// The place is read off the screen's stored screenshot, which an opened Duo's frames are turned upright for: there the
// portrait point (x, y) shows at (1 - y, x). A frame read without an image says nothing, so the screen's screenshot is asked.
const turned = latest!.landscape || map.observations.some(item => item.screenId === currentScreenId && map.artifacts.find(artifact => artifact.id === item.screenshotArtifactId)?.turned);
const shown = turned ? { x: 1 - point.y, y: point.x } : point;
const row = shown.y < 1 / 3 ? "top" : shown.y < 2 / 3 ? "" : "bottom";
const column = shown.x < 1 / 3 ? "left" : shown.x < 2 / 3 ? "" : "right";
const where = [row, column].filter(Boolean).join(" ") || "centre";
return `${what}, ${where}`;
};
@@ -749,10 +780,10 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
const frameId = text(input.frameId, "frameId", 256);
if (!latest?.png && latest?.imageUnavailable?.code === "CAPTURE_SENSITIVE_INPUT") throw failure("CAPTURE_SENSITIVE_INPUT", latest.imageUnavailable.message, true);
if (!latest?.png || latest.frameId !== frameId) throw failure("AGENT_FRAME_STALE", "Observe with an image first; capture must name the latest frame.", true);
const title = text(input.title, "title", 80).trim();
const title = text(input.title, "title", TOOL_LIMITS.title).trim();
if (captures >= options.maxScreens) { stop("budget-screens"); abort.abort(failure("EXPLORATION_BUDGET_EXHAUSTED", "Screen budget exhausted.")); throw abort.signal.reason; }
const { screen } = locate(latest, title);
const state = str(input.state, 80);
const state = str(input.state, TOOL_LIMITS.state);
if (!state && map.observations.some(item => item.screenId === screen.id && item.screenshotArtifactId)) {
if (title) screen.title = title;
return toolResult({ ok: true, duplicate: true, screenId: screen.id, title: screen.title, message: "This screen is already captured (its title is updated); pass state to record a variant." });
@@ -832,9 +863,9 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
await persist();
return toolResult({ ok: !error, outcome, ...(error ? { error } : {}), ...view, ...progress().budget });
}
if (name === "finish") { stop("finished", text(input.summary, "summary", 4000)); abort.abort(new Error("finished")); return toolResult({ ok: true }); }
if (name === "finish") { stop("finished", text(input.summary, "summary", TOOL_LIMITS.summary)); abort.abort(new Error("finished")); return toolResult({ ok: true }); }
if (name === "stop_blocked") {
const reason = text(input.reason, "reason", 1000);
const reason = text(input.reason, "reason", TOOL_LIMITS.reason);
map.gaps.push({ id: `gap_${randomUUID().slice(0, 8)}`, kind: "coverage", status: "open", message: reason, ...(currentScreenId ? { screenId: currentScreenId } : {}), createdAt: timestamp() });
await persist();
stop("blocked", reason); abort.abort(new Error("blocked")); return toolResult({ ok: true });
@@ -856,8 +887,9 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
if (!result) stop(budget ? "budget-time" : "model-error", undefined, budget ? undefined : fatal ?? error);
}
if (result!.stopReason !== "finished" && result!.stopReason !== "blocked") {
const note = result!.stopReason === "model-error" ? undefined : options.budgetNote?.();
map.gaps.push({ id: `gap_${randomUUID().slice(0, 8)}`, kind: result!.stopReason === "model-error" ? "other" : "coverage", status: "open",
message: result!.stopReason === "model-error" ? `The exploration stopped on an error: ${result!.error instanceof Error ? result!.error.message.slice(0, 300) : "unknown"}` : `The exploration stopped at its ${result!.stopReason.replace("budget-", "")} budget with ${map.frontier.filter(item => item.status === "pending").length} controls unexplored.`,
message: result!.stopReason === "model-error" ? `The exploration stopped on an error: ${result!.error instanceof Error ? result!.error.message.slice(0, 300) : "unknown"}` : `The exploration stopped at its ${result!.stopReason.replace("budget-", "")} budget with ${map.frontier.filter(item => item.status === "pending").length} controls unexplored.${note ? ` ${note}` : ""}`,
createdAt: timestamp() });
}
for (const item of map.frontier) if (item.status === "claimed") item.status = "pending";
@@ -870,10 +902,10 @@ export async function runBlackBoxExploration(options: BlackBoxExplorerOptions):
function nameUncapturedScreens(map: ExploreMap): void {
const titles = new Map(map.screens.map(screen => [screen.id, screen.title]));
for (const screen of map.screens) {
if (!/^Screen \d+$/.test(screen.title)) continue;
if (!isPlaceholderTitle(screen.title)) continue;
const arrival = map.actionAttempts.find(attempt => attempt.targetScreenId === screen.id && attempt.outcome === "transition" && attempt.sourceScreenId !== screen.id);
const from = arrival ? titles.get(arrival.sourceScreenId) : undefined;
if (!arrival || !from || /^Screen \d+$/.test(from)) continue;
if (!arrival || !from || isPlaceholderTitle(from)) continue;
const step = arrival.action.input ? "typing" : arrival.action.label;
screen.title = `${from} → ${step}`.slice(0, 80);
titles.set(screen.id, screen.title);
+300
View File
@@ -0,0 +1,300 @@
import { readFile } from "node:fs/promises";
import { CiError } from "../shared/ci-errors.js";
import type { CiAgentSessionRunRequest, CiAgentSessionRunResult } from "../capture/agent/session.js";
import { SETUP_SELECTOR_MAX, isSetupInput, isSetupSelector } from "../capture/setup-actions.js";
import type { CaptureSetupAction } from "../config/explore-config.js";
import { explorationSecretNames } from "../config/exploration-inputs.js";
import { TOOL_LIMITS, isPlaceholderTitle } from "./explorer.js";
export type ReplayPlatform = "ios" | "android";
type Selector = { label: string; identifier?: never } | { identifier: string; label?: never };
type ReplayAction =
| ({ action: "tap" } & Selector)
| ({ action: "fill"; text: string } & Selector)
| { action: "type"; text: string }
| { action: "scroll"; direction: "up" | "down" | "left" | "right" | "top" | "bottom" }
| { action: "wait"; milliseconds: number }
| { action: "open"; url: string }
| { action: "back" }
| { action: "dismiss-keyboard" };
type ReplayCapture = { capture: string; state?: string; description?: string };
/** `platforms` limits a step to those platforms; an `optional` action that fails is skipped instead of ending the replay. */
export type ReplayStep = (ReplayCapture & { platforms?: ReplayPlatform[] }) | (ReplayAction & { platforms?: ReplayPlatform[]; optional?: boolean });
export type ReplayPlan = { steps: ReplayStep[] };
export type ReplaySession = {
readonly provider: "replay";
readonly model: string;
preflight(): Promise<void>;
run(request: CiAgentSessionRunRequest): Promise<CiAgentSessionRunResult>;
/** After a budget stop, the planned screens the run did not reach, for the explorer's gap. */
budgetNote(): string | undefined;
};
const DIRECTIONS = new Set(["up", "down", "left", "right", "top", "bottom"]);
/**
* The fields each kind of step takes. Anything else is refused: a typo (`platform`, `optinal`) would otherwise do nothing,
* and a field `act` reads another way (a coordinate tap's `frameId`, `x`, `y` and `target`) would change what is tapped.
*/
const FIELDS = new Map<string, readonly string[]>([
["capture", ["capture", "state", "description", "platforms"]],
["tap", ["action", "label", "identifier", "platforms", "optional"]],
["fill", ["action", "label", "identifier", "text", "platforms", "optional"]],
["type", ["action", "text", "platforms", "optional"]],
["scroll", ["action", "direction", "platforms", "optional"]],
["wait", ["action", "milliseconds", "platforms", "optional"]],
["open", ["action", "url", "platforms", "optional"]],
["back", ["action", "platforms", "optional"]],
["dismiss-keyboard", ["action", "platforms", "optional"]],
]);
const ACTIONS = [...FIELDS.keys()].filter(kind => kind !== "capture");
function invalid(message: string): CiError {
return new CiError({ code: "REPLAY_PLAN_INVALID", stage: "exploration", message, retryable: false,
nextAction: "Fix the replay plan: a JSON object with a steps array of actions ({\"action\":\"tap\",\"label\":\"Settings\"}) and captures ({\"capture\":\"Settings\"})." });
}
const nonEmpty = (value: unknown, max: number): value is string => typeof value === "string" && value.trim().length > 0 && value.length <= max;
function parseStep(step: unknown, at: string): ReplayStep {
if (!step || typeof step !== "object" || Array.isArray(step)) throw invalid(`${at} is not an object.`);
const entry = step as Record<string, unknown>;
const kind = "capture" in entry ? "capture" : typeof entry.action === "string" && FIELDS.has(entry.action) ? entry.action : undefined;
if (!kind) throw invalid(`${at} needs a capture title or an action (${ACTIONS.join(", ")}).`);
const fields = FIELDS.get(kind)!;
const unknown = Object.keys(entry).find(key => !fields.includes(key));
if (unknown !== undefined) throw invalid(`${at} (${kind}) has a field ${JSON.stringify(unknown.slice(0, 40))} it does not take; a ${kind} step takes ${fields.join(", ")}.`);
if (entry.platforms !== undefined && (!Array.isArray(entry.platforms) || !entry.platforms.length || entry.platforms.some(item => item !== "ios" && item !== "android"))) {
throw invalid(`${at}: platforms must list ios and/or android.`);
}
if (entry.optional !== undefined && typeof entry.optional !== "boolean") throw invalid(`${at}: optional must be true or false.`);
const extra = { ...(entry.platforms ? { platforms: [...entry.platforms] as ReplayPlatform[] } : {}), ...(entry.optional !== undefined ? { optional: entry.optional as boolean } : {}) };
if (kind === "capture") {
if (!nonEmpty(entry.capture, TOOL_LIMITS.title)) throw invalid(`${at}: capture must be a screen title of at most ${TOOL_LIMITS.title} characters.`);
if (entry.state !== undefined && !nonEmpty(entry.state, TOOL_LIMITS.state)) throw invalid(`${at}: state must be a string of at most ${TOOL_LIMITS.state} characters.`);
if (entry.description !== undefined && !nonEmpty(entry.description, TOOL_LIMITS.description)) throw invalid(`${at}: description must be a string of at most ${TOOL_LIMITS.description} characters.`);
return { capture: entry.capture.trim(), ...(entry.state !== undefined ? { state: entry.state as string } : {}),
...(entry.description !== undefined ? { description: entry.description as string } : {}), ...extra };
}
// The device selects by the setup executor's rules, which are stricter than what `act` itself accepts.
const selector = (): Selector => {
const keys = (["label", "identifier"] as const).filter(key => entry[key] !== undefined);
if (keys.length !== 1) throw invalid(`${at}: ${kind} needs exactly one of label or identifier.`);
const key = keys[0]!, value = entry[key];
if (!isSetupSelector(value) || value.length > TOOL_LIMITS.selector) throw invalid(`${at}: ${key} must be 1–${SETUP_SELECTOR_MAX} characters without control characters.`);
return key === "label" ? { label: value } : { identifier: value };
};
const input = (): string => {
if (!nonEmpty(entry.text, TOOL_LIMITS.text) || !isSetupInput(entry.text)) throw invalid(`${at}: ${kind} needs text: one line of at most ${TOOL_LIMITS.text} characters.`);
try { explorationSecretNames(entry.text); }
catch (error) { throw invalid(`${at}: ${error instanceof Error ? error.message : String(error)} Write each input as {{secret:NAME}} with an uppercase environment variable name that is not a provider, Expo, or CI credential.`); }
return entry.text;
};
switch (kind) {
case "tap": return { action: "tap", ...selector(), ...extra };
case "fill": return { action: "fill", ...selector(), text: input(), ...extra };
case "type": return { action: "type", text: input(), ...extra };
case "scroll":
if (typeof entry.direction !== "string" || !DIRECTIONS.has(entry.direction)) throw invalid(`${at}: scroll needs a direction (${[...DIRECTIONS].join(", ")}).`);
return { action: "scroll", direction: entry.direction as "up", ...extra };
case "wait":
if (!(Number.isInteger(entry.milliseconds) && Number(entry.milliseconds) >= 1 && Number(entry.milliseconds) <= TOOL_LIMITS.waitMs)) {
throw invalid(`${at}: wait needs milliseconds between 1 and ${TOOL_LIMITS.waitMs}.`);
}
return { action: "wait", milliseconds: entry.milliseconds as number, ...extra };
case "open":
if (!nonEmpty(entry.url, TOOL_LIMITS.url)) throw invalid(`${at}: open needs a url of at most ${TOOL_LIMITS.url} characters.`);
return { action: "open", url: entry.url, ...extra };
default: return { action: kind as "back" | "dismiss-keyboard", ...extra };
}
}
/** Checks a plan and keeps only the fields each step takes, so what reaches `act` is what the plan documents. */
export function parseReplayPlan(value: unknown): ReplayPlan {
if (!value || typeof value !== "object" || !Array.isArray((value as { steps?: unknown }).steps)) throw invalid("The replay plan has no steps array.");
const unknown = Object.keys(value).find(key => key !== "steps");
if (unknown !== undefined) throw invalid(`The replay plan has a field ${JSON.stringify(unknown.slice(0, 40))} it does not take; it takes only steps.`);
const steps = (value as { steps: unknown[] }).steps;
if (!steps.length) throw invalid("The replay plan has no steps.");
return { steps: steps.map((step, index) => parseStep(step, `Step ${index + 1}`)) };
}
export async function readReplayPlan(path: string): Promise<ReplayPlan> {
let text: string;
try { text = await readFile(path, "utf8"); }
catch (error) { throw invalid(`The replay plan ${path} could not be read: ${error instanceof Error ? error.message : String(error)}`); }
let value: unknown;
try { value = JSON.parse(text); }
catch (error) { throw invalid(`The replay plan ${path} is not JSON: ${error instanceof Error ? error.message : String(error)}`); }
return parseReplayPlan(value);
}
/** The plan's fill and type steps, whose {{secret:NAME}} inputs must be set before any device starts, as the hints' are. */
export function replaySetup(plan: ReplayPlan): CaptureSetupAction[] {
return plan.steps.flatMap(step => "action" in step && (step.action === "fill" || step.action === "type") ? [step as CaptureSetupAction] : []);
}
const stepsFor = (plan: ReplayPlan, platform: ReplayPlatform) => plan.steps.map((step, index) => ({ step, number: index + 1 }))
.filter(({ step }) => !step.platforms || step.platforms.includes(platform));
/**
* Refuses, before any device starts, a plan with nothing to do on this platform or one the budgets cannot hold: a run that
* would stop at its budget partway through the plan is better raised or shortened up front.
*/
export function assertReplayFits(plan: ReplayPlan, platform: ReplayPlatform, limits: { maxScreens: number; maxActions: number; maxTurns: number }): void {
const steps = stepsFor(plan, platform);
if (!steps.length) {
throw new CiError({ code: "REPLAY_PLAN_EMPTY", stage: "exploration", message: `No step of the replay plan runs on ${platform}: each one is limited to ${platform === "ios" ? "android" : "ios"} by its platforms.`,
nextAction: `Add steps for ${platform} to the plan, or run it on the platform it was written for.` });
}
const captures = steps.filter(({ step }) => "capture" in step).length;
const actions = steps.filter(({ step }) => "action" in step && step.action !== "wait").length;
// A screen counts against the budget when it is first photographed, on arrival, and a capture is refused once the budget
// is used up, so the last capture needs one screen to spare.
const over = [
...(captures >= limits.maxScreens ? [{ text: `${captures} captures, which need a screen budget of at least ${captures + 1} (it is ${limits.maxScreens})`, flag: `--max-screens ${captures + 1}` }] : []),
...(actions > limits.maxActions ? [{ text: `${actions} actions, more than the action budget of ${limits.maxActions}`, flag: `--max-actions ${actions}` }] : []),
];
if (over.length) {
throw new CiError({ code: "REPLAY_PLAN_OVER_BUDGET", stage: "exploration", message: `On ${platform} the replay plan takes ${over.map(item => item.text).join(" and ")}.`,
nextAction: `Pass ${over.map(item => item.flag).join(" ")} (or more: each screen the plan reaches without capturing it also counts), or shorten the plan.` });
}
// Each capture is an observe and a capture call, each action one act call, and the run ends with get_progress and finish.
const turns = steps.length + captures + 2;
if (turns > limits.maxTurns) {
throw new CiError({ code: "REPLAY_PLAN_OVER_BUDGET", stage: "exploration", message: `On ${platform} the replay plan takes ${turns} tool calls, more than the ${limits.maxTurns} that the screen and action budgets allow.`,
nextAction: "Raise --max-actions, or use fewer wait steps." });
}
}
function describeAction(step: ReplayAction): string {
if ("label" in step && step.label) return `${step.action} "${step.label}"`;
if ("identifier" in step && step.identifier) return `${step.action} #${step.identifier}`;
if (step.action === "scroll") return `scroll ${step.direction}`;
return step.action;
}
/** Only the fields `act` documents for that action, whatever else the step object carries. */
function actInput(step: ReplayAction): Record<string, unknown> {
const selector = "label" in step && step.label !== undefined ? { label: step.label } : "identifier" in step && step.identifier !== undefined ? { identifier: step.identifier } : {};
switch (step.action) {
case "tap": return { action: "tap", ...selector };
case "fill": return { action: "fill", ...selector, text: step.text };
case "type": return { action: "type", text: step.text };
case "scroll": return { action: "scroll", direction: step.direction };
case "wait": return { action: "wait", milliseconds: step.milliseconds };
case "open": return { action: "open", url: step.url };
default: return { action: step.action };
}
}
/** As many names as fit in `room` characters, then "and N more"; a count of them when not even the first fits. */
function listed(names: readonly string[], room: number, noun: string): string {
let text = "";
for (const [index, name] of names.entries()) {
const next = text ? `${text}, ${name}` : name;
const left = names.length - index - 1;
if (next.length + (left ? ` and ${left} more`.length : 0) > room) return text ? `${text} and ${names.length - index} more` : `${names.length} ${noun}`;
text = next;
}
return text;
}
const sameTitle = (left: string, right: string): boolean => left.toLowerCase() === right.toLowerCase();
/**
* Runs a fixed plan through the explorer's own tools in place of a model: the same observe, act, and capture calls,
* so the map and report are those of a model run, but every run takes the same steps. A step that fails does not
* fail the run: the report keeps what was captured and names the screens that were not.
*/
export function createReplaySession(plan: ReplayPlan, platform: ReplayPlatform): ReplaySession {
const steps = stepsFor(plan, platform);
const title = (step: ReplayCapture) => step.state ? `${step.capture} (${step.state})` : step.capture;
const missed: string[] = [];
let position = 0;
/** What the plan meant to capture and has not: the screens still ahead first, then the ones earlier steps missed. */
const notCaptured = () => [...steps.slice(position).flatMap(({ step }) => "capture" in step ? [title(step)] : []), ...missed];
return {
provider: "replay",
model: "replay plan",
async preflight() {},
budgetNote() {
const names = notCaptured();
return names.length ? `Not captured: ${listed(names, TOOL_LIMITS.reason, "planned screens")}.` : undefined;
},
async run(request) {
const signal = request.signal ?? new AbortController().signal;
const call = async (name: string, input: Record<string, unknown>) => {
const result = await request.callTool(name, input, signal);
const text = result.content.find(block => block.type === "text");
const value = text && "text" in text ? JSON.parse(text.text) as Record<string, unknown> : {};
return { failed: Boolean(result.isError) || value.ok === false, value };
};
const why = (value: Record<string, unknown>) => String(value.error ?? value.reason ?? value.message ?? "it failed");
// A refused finish or stop_blocked would otherwise end the session with no result, which the explorer reports as a turn budget stop.
const end = async (name: "finish" | "stop_blocked", input: Record<string, unknown>) => {
const ended = await call(name, input);
if (!ended.failed) return;
if (name === "finish") return end("stop_blocked", { reason: `The replay could not finish (${why(ended.value)}). ${String(input.summary)}`.slice(0, TOOL_LIMITS.reason) });
throw new CiError({ code: "REPLAY_STOP_REFUSED", stage: "exploration", message: `The explorer refused the replay's ${name}: ${why(ended.value)}`.slice(0, 600),
nextAction: "Report this with the replay plan; the screens captured so far are in the map." });
};
/** Screen id to the title the plan captured it under. */
const planned = new Map<string, string>();
const skipped: string[] = [];
missed.length = 0;
try {
for (const [index, { step, number }] of steps.entries()) {
position = index;
if ("capture" in step) {
const seen = await call("observe", { includeImage: true });
const frameId = typeof seen.value.frameId === "string" ? seen.value.frameId : undefined;
const known = typeof seen.value.screenId === "string" && !step.state ? planned.get(seen.value.screenId) : undefined;
// The plan already captured this screen: a second capture would only rename it, so it is not made.
if (known !== undefined) {
if (!sameTitle(known, step.capture)) missed.push(`${title(step)} (looked the same as ${known})`);
continue;
}
const captured = frameId && !seen.failed ? await call("capture", { frameId, title: step.capture,
...(step.state ? { state: step.state } : {}), ...(step.description ? { description: step.description } : {}) }) : undefined;
const screenId = typeof captured?.value.screenId === "string" ? captured.value.screenId : undefined;
if (!captured || captured.failed || !screenId) missed.push(title(step));
else if (!planned.has(screenId)) planned.set(screenId, step.capture);
continue;
}
const acted = await call("act", actInput(step));
if (!acted.failed) continue;
if (step.optional) { skipped.push(`step ${number} (${describeAction(step)})`); continue; }
const head = `Replay step ${number} (${describeAction(step)}) failed: ${why(acted.value)}`.slice(0, 600);
const names = notCaptured();
const lead = ". Not captured: ";
await end("stop_blocked", { reason: head + (names.length ? `${lead}${listed(names, TOOL_LIMITS.reason - head.length - lead.length, "planned screens")}` : "") });
return { inputTokens: 0, outputTokens: 0 };
}
position = steps.length;
// The map as it stands: screens merged into others are gone, and each keeps its latest title. One reached but never
// captured is still "Screen N" here and is named after the path to it only once the run ends, so it is counted.
const progress = await call("get_progress", {});
const screens = Array.isArray(progress.value.captured) ? progress.value.captured as Array<{ id?: unknown; title?: unknown }> : [];
const unplanned = screens.filter(screen => typeof screen.id === "string" && !planned.has(screen.id)).map(screen => typeof screen.title === "string" ? screen.title : "");
const named = unplanned.filter(name => name && !isPlaceholderTitle(name));
const unnamed = unplanned.length - named.length;
const reached = [...named, ...(unnamed ? [`${unnamed} unnamed screen${unnamed === 1 ? "" : "s"}`] : [])];
const sections = ([["Not captured", missed, "planned screens"], ["Optional steps skipped because they failed", skipped, "steps"],
["Also reached, not in the plan", reached, "screens"]] as const).filter(([, names]) => names.length);
let summary = `Replayed ${steps.length} planned steps on ${platform}.${missed.length ? "" : " Every planned screen was captured."}`;
for (const [index, [label, names, noun]] of sections.entries()) {
// Each later section keeps room for its label and a count, so a long list of misses does not crowd out the rest.
const reserve = sections.slice(index + 1).reduce((sum, [later]) => sum + later.length + 32, 0);
const lead = ` ${label}: `;
summary += `${lead}${listed(names, TOOL_LIMITS.summary - summary.length - lead.length - 1 - reserve, noun)}.`;
}
await end("finish", { summary });
} catch (error) {
if (signal.aborted) return { inputTokens: 0, outputTokens: 0 };
throw error;
}
return { inputTokens: 0, outputTokens: 0 };
},
};
}
+269 -87
View File
@@ -1,11 +1,11 @@
import { createHash, randomUUID } from "node:crypto";
import { createRequire } from "node:module";
import { lstat, mkdir, mkdtemp, readFile, realpath, rm, writeFile } from "node:fs/promises";
import { tmpdir } from "node:os";
import { join } from "node:path";
import { createCiAgentSession, type CiAgentSession, type CiAgentSessionConfig } from "../capture/agent/session.js";
import { createLocalCiAgentSession } from "../capture/agent/session-local.js";
import type { CiAgentProviderName } from "../capture/agent/provider.js";
import { assertReplayFits, createReplaySession, parseReplayPlan, replaySetup, type ReplayPlan } from "./replay.js";
import { CONTROLLER_SESSION, baseEnvironment, isRecord } from "../capture/common.js";
import { controllerData, defaultRunner, openedSimulatorDevice } from "../capture/controller.js";
import type { CommandRequest, CommandResult } from "../capture/run.js";
@@ -13,14 +13,20 @@ import { executeCaptureSetup } from "../capture/setup-actions.js";
import { SIMULATOR_DOTENV_FILE_NAME, SIMULATOR_DOTENV_HEADER, assertSimulatorSessionIdentity, assertSimulatorStopped, readSimulatorDotenvHandoff,
assertAttachedSimulatorSession, connectionController, controllerVersion, resolveAttachedSimulatorSession, simulatorOsVersion, simulatorStartFailure, type ActiveSimulatorRuntime, type SimulatorConnection } from "../capture/simulator.js";
import { argentControllerInvocation } from "../capture/controller-invocation.js";
import { agentDeviceLauncher, ensureController } from "../shared/controller-cache.js";
import { agentDeviceLauncher, ensureArgentToolServer, ensureController } from "../shared/controller-cache.js";
import { agentDeviceClaims, approveLinkScheme, createLocalSimulator, deleteLocalEmulator, deleteLocalSimulator, hostEmulatorAbi, installOnLocalEmulator, installOnLocalSimulator, startLocalArgentServer,
startLocalEmulator, stopLocalAgentDevice, stopLocalArgentServer, waitForLocalEmulator, waitForLocalSimulator, type LocalArgentServer } from "../capture/local-simulator.js";
import { openLocalCaptureRecord, sweepLocalCaptures } from "../capture/local-registry.js";
import { screenSpace } from "../capture/route-capture.js";
import { approveSessionUrlSchemes, simulatorControlActions, type EasCommand } from "../capture/simulator-control.js";
import { DEFAULT_ANDROID_DEVICE } from "../config/explore-config.js";
import { DUO_PANELS } from "../contracts/capture.js";
import { createExplorationInputs } from "../config/exploration-inputs.js";
import { ExploreStore, getCoverage } from "../map/store.js";
import { whoamiAccounts } from "../report/preview-deploy.js";
import { CiError, serializeCiError, toCiError } from "../shared/ci-errors.js";
import { packageBin } from "../shared/package-paths.js";
import { assertDecodedPng, type DecodedFrame } from "../shared/png-evidence.js";
import { assertDecodedPng, solidColour } from "../shared/png-evidence.js";
import { CONTRACT, type ExploreMap, type Platform } from "../types.js";
import { assertEmulatorCompatible, extractIosBundle, inspectArchive, type ArchiveKind, type InspectedArchive, readArchiveIcon } from "./archive.js";
import { runBlackBoxExploration, type BlackBoxExplorerResult, type BlackBoxFrame, SYSTEM_DIALOG_PACKAGES } from "./explorer.js";
@@ -50,8 +56,12 @@ export type BinaryExploreOptions = {
project?: string;
/** The session's device controller; agent-device unless Argent is chosen. explore:session takes it from the session. */
controller?: "agent-device" | "argent";
/** Where the app runs: an EAS Simulator session (the default), or a simulator or emulator this run starts on this machine, with no EAS login. */
simulator?: "eas" | "local";
environment?: NodeJS.ProcessEnv;
runner?: typeof defaultRunner;
/** Fixed steps to take instead of a model's; no model access is needed. */
replay?: ReplayPlan;
createSession?: (config: CiAgentSessionConfig, options: { env: NodeJS.ProcessEnv; auth: ModelAuth }) => CiAgentSession;
progress?: (message: string) => void;
now?: () => number;
@@ -63,7 +73,7 @@ export type BinaryExploreResult = {
archive: Omit<InspectedArchive, "path"> & { path: string };
session: { id: string; platform: Platform; device: string; owner: string; project: string };
runtime: ActiveSimulatorRuntime;
model: { provider: CiAgentProviderName; model: string; auth: "local-login" | "api-key" };
model: { provider: CiAgentProviderName | "replay"; model: string; auth: "local-login" | "api-key" | "none" };
exploration: BlackBoxExplorerResult;
durationMs: number;
};
@@ -71,8 +81,8 @@ export type BinaryExploreResult = {
/** The command and environment that reach a session's controller: agent-device's CLI with its daemon, or the Argent bridge. */
type ControllerInvocation = { command: string; environment: NodeJS.ProcessEnv };
async function argentInvocation(connection: SimulatorConnection, work: string, platform: Platform, environment: NodeJS.ProcessEnv, bundleId?: string): Promise<ControllerInvocation> {
return argentControllerInvocation({ directory: work, connection, platform, environment: baseEnvironment(environment), sourceEnvironment: environment, ...(bundleId ? { bundleId } : {}) });
async function argentInvocation(connection: SimulatorConnection, work: string, platform: Platform, environment: NodeJS.ProcessEnv, bundleId?: string, device?: string): Promise<ControllerInvocation> {
return argentControllerInvocation({ directory: work, connection, platform, environment: baseEnvironment(environment), sourceEnvironment: environment, ...(bundleId ? { bundleId } : {}), ...(device ? { device } : {}) });
}
/** The daemon refuses a second session on a busy device; the caller's own `agent-device` session (from install/open) is reused instead. */
@@ -157,8 +167,8 @@ export function resolveSessionApp(runner: typeof defaultRunner, connection: Simu
* App-test inputs: each {{secret:NAME}} in the hints names an environment variable whose value Explore types for the
* model. Read before any session starts, so a missing one costs nothing.
*/
export function hintInputs(hints: string | undefined, environment: NodeJS.ProcessEnv): ReturnType<typeof createExplorationInputs> {
try { return createExplorationInputs({ instructions: hints ?? "", environment }); }
export function hintInputs(hints: string | undefined, environment: NodeJS.ProcessEnv, replay?: ReplayPlan): ReturnType<typeof createExplorationInputs> {
try { return createExplorationInputs({ instructions: hints ?? "", ...(replay ? { setup: replaySetup(replay) } : {}), environment }); }
catch (error) {
if (!(error instanceof CiError) || !error.code.startsWith("EXPLORATION_SECRET_")) throw error;
const name = typeof error.details?.secretName === "string" ? error.details.secretName : "NAME";
@@ -189,6 +199,36 @@ function modelSession(provider: CiAgentProviderName, auth: ModelAuth, options: M
return { session, provider, model: config.model, auth };
}
/** The turns the model gets: three for each action and two for each screen, plus a few to look around and finish. */
const turnBudget = (limits: { maxScreens: number; maxActions: number }): number => limits.maxActions * 3 + limits.maxScreens * 2 + 20;
/**
* A replay plan is checked here as well as by the CLI, for callers that pass one directly. It takes the model's place, so
* the options that only steer a model would do nothing; the plan's own fill and type steps name the secrets it types.
*/
function replayPlan(options: Pick<BinaryExploreOptions, "replay" | "provider" | "model" | "hints">): ReplayPlan | undefined {
if (!options.replay) return undefined;
const ignored = (["provider", "model", "hints"] as const).filter(name => options[name] !== undefined);
if (ignored.length) {
throw binaryError("REPLAY_OPTIONS_IGNORED", `--replay takes the model's place, so ${ignored.map(name => `--${name}`).join(" and ")} would do nothing.`,
`Drop ${ignored.map(name => `--${name}`).join(" and ")}. A {{secret:NAME}} in the plan's fill and type steps is read from the environment, as one in --hints is.`);
}
const plan = parseReplayPlan(options.replay);
for (const [index, step] of plan.steps.entries()) {
if (!("action" in step) || step.action !== "open") continue;
try { assertAppUrl(step.url); }
catch (error) { throw new CiError({ code: "REPLAY_PLAN_INVALID", stage: "exploration", message: `Step ${index + 1}: ${error instanceof Error ? error.message : String(error)}`, nextAction: "Open only the app's own URLs in the replay plan." }); }
}
return plan;
}
/** A replay plan takes the model's place: no model access is checked or used. */
function replayModel(plan: ReplayPlan, platform: Platform, limits: { maxScreens: number; maxActions: number }) {
assertReplayFits(plan, platform, { ...limits, maxTurns: turnBudget(limits) });
const session = createReplaySession(plan, platform);
return { session, budgetNote: session.budgetNote, provider: "replay" as const, model: "replay plan", auth: "none" as const };
}
/** Local `claude`/`codex` login unless an API key is present; preflight proves access before a simulator is paid for. */
export function selectModelSession(options: ModelSessionOptions, environment: NodeJS.ProcessEnv) {
const provider: CiAgentProviderName = options.provider ?? (environment.OPENAI_API_KEY?.trim() && !environment.ANTHROPIC_API_KEY?.trim() ? "openai" : "anthropic");
@@ -255,7 +295,7 @@ export type ConnectedSessionExploration = {
limits: { maxDurationMinutes: number; maxScreens: number; maxActions: number };
store: ExploreStore;
map: ExploreMap;
model: { session: CiAgentSession };
model: { session: Pick<CiAgentSession, "run">; budgetNote?: () => string | undefined };
/** `eas simulator:get --json` for the attached session; descriptive (OS version) only. */
sessionJson: () => string;
/** agent-device session name on the daemon; explore:session reuses the caller's session so the device is not "in use" by another. */
@@ -267,11 +307,18 @@ export type ConnectedSessionExploration = {
urlSchemes?: readonly string[];
/** The app's icon from the archive for the Viewer header. */
appIcon?: Buffer;
/**
* A simulator or emulator on this machine instead of an EAS session: the controller drives it by its UDID or adb
* serial (agent-device with a daemon of its own in `stateDirectory`, or a local Argent tool-server pinned to it).
*/
local?: { deviceId: string; osVersion: string; stateDirectory: string };
/** Runs once the app is installed and before anything opens it, as an EAS session's link approval must. */
afterInstall?: () => void;
};
/** Drives an agent-device daemon we already hold a connection to: optional install, open, then the model-led exploration. */
export async function exploreConnectedSession(ctx: ConnectedSessionExploration): Promise<{ runtime: ActiveSimulatorRuntime; exploration: BlackBoxExplorerResult }> {
const { connection, platform, device, bundleId, app, install, work, runner, environment, secrets, inputs, say, now, limits, store, map, model, sessionJson, launchUrl, hints, allowDestructive } = ctx;
const { connection, platform, device, bundleId, app, install, work, runner, environment, secrets, inputs, say, now, limits, store, map, model, sessionJson, launchUrl, hints, allowDestructive, local } = ctx;
const controllerSession = ctx.controllerSession ?? CONTROLLER_SESSION;
const controllerKind = connectionController(connection);
const controllerName = controllerKind === "argent" ? "Argent" : "agent-device";
@@ -279,15 +326,18 @@ export async function exploreConnectedSession(ctx: ConnectedSessionExploration):
let agentDevice: string;
let controllerEnvironment: NodeJS.ProcessEnv;
if (controllerKind === "argent") {
({ command: agentDevice, environment: controllerEnvironment } = await argentInvocation(connection, work, platform, environment, bundleId));
({ command: agentDevice, environment: controllerEnvironment } = await argentInvocation(connection, work, platform, environment, bundleId, local?.deviceId));
} else {
const controllerConfig = join(work, "agent-device.json");
await writeFile(controllerConfig, "{}\n", { flag: "wx", mode: 0o400 });
controllerEnvironment = { ...baseEnvironment(environment), AGENT_DEVICE_CONFIG: controllerConfig, AGENT_DEVICE_DAEMON_BASE_URL: connection.daemonBaseUrl,
AGENT_DEVICE_DAEMON_AUTH_TOKEN: connection.daemonAuthToken, AGENT_DEVICE_SESSION: controllerSession, AGENT_DEVICE_SESSION_LOCK: "reject", AGENT_DEVICE_PLATFORM: platform };
// A local device has no EAS daemon: agent-device runs its own, in a state directory of this run's.
controllerEnvironment = { ...baseEnvironment(environment), AGENT_DEVICE_CONFIG: controllerConfig,
...(local ? { AGENT_DEVICE_STATE_DIR: local.stateDirectory } : { AGENT_DEVICE_DAEMON_BASE_URL: connection.daemonBaseUrl, AGENT_DEVICE_DAEMON_AUTH_TOKEN: connection.daemonAuthToken }),
AGENT_DEVICE_SESSION: controllerSession, AGENT_DEVICE_SESSION_LOCK: "reject", AGENT_DEVICE_PLATFORM: platform };
agentDevice = agentDeviceLauncher(environment);
}
const deviceArgs = platform === "ios" ? ["--device", device] : [];
// A local device is named for Explore, so it is picked by its UDID or serial rather than its model name.
const deviceArgs = local ? [platform === "ios" ? "--udid" : "--serial", local.deviceId] : platform === "ios" ? ["--device", device] : [];
const controller = (args: string[], timeoutMs = 90_000): CommandResult => {
// Session flags go before a `--` separator so model-chosen text (`type -h`, a URL, a label) is never read as a flag.
const separator = args.indexOf("--");
@@ -304,16 +354,18 @@ export async function exploreConnectedSession(ctx: ConnectedSessionExploration):
try { controller(["install", install.path, "--json"], 15 * 60_000); }
catch (error) { throw binaryError("APP_INSTALL_FAILED", `${controllerName} could not install the app: ${failureText(error)}`, "Check that the archive is a simulator build for this platform and under the size limit, then rerun.", { sessionId: connection.sessionId }); }
}
const opened = controller(["open", bundleId, "--level", "full", "--json"], 2 * 60_000);
ctx.afterInstall?.();
// On a local device, the first open also builds agent-device's iOS runner.
const opened = controller(["open", bundleId, "--level", "full", "--json"], local ? 15 * 60_000 : 2 * 60_000);
const openedDevice = openedSimulatorDevice(opened.stdout, bundleId, "initial app open", platform, controllerSession);
const data = controllerData(opened.stdout, "initial app open");
const deviceData = data.device && typeof data.device === "object" ? data.device as Record<string, unknown> : {};
const observed = (key: string): string => { const value = deviceData[key] ?? data[key]; return typeof value === "string" && value.trim() ? value : "unknown"; };
const runtime = { provider: "eas-simulator", controller: controllerKind, controllerPackageVersion: controllerVersion(controllerKind), easCliVersion: "unknown", platform, controllerDeviceId: openedDevice.deviceId,
deviceModel: openedDevice.deviceModel, osVersion: observed("osVersion"), locale: observed("locale"), orientation: observed("orientation") === "landscape" ? "landscape" : "portrait",
const runtime = { provider: local ? "local-simulator" : "eas-simulator", controller: controllerKind, controllerPackageVersion: controllerVersion(controllerKind), easCliVersion: "unknown", platform, controllerDeviceId: openedDevice.deviceId,
deviceModel: local ? device : openedDevice.deviceModel, osVersion: local ? local.osVersion : observed("osVersion"), locale: observed("locale"), orientation: observed("orientation") === "landscape" ? "landscape" : "portrait",
colorScheme: observed("colorScheme") === "dark" ? "dark" : "light" } as ActiveSimulatorRuntime;
// Argent's `open` reports the OS version itself (list-devices); agent-device's comes from the session's preview.
if (runtime.osVersion === "unknown" || controllerKind !== "argent") try { runtime.osVersion = await simulatorOsVersion(sessionJson(), runtime.controllerDeviceId, fetch, platform); } catch { /* descriptive only */ }
if (!local && (runtime.osVersion === "unknown" || controllerKind !== "argent")) try { runtime.osVersion = await simulatorOsVersion(sessionJson(), runtime.controllerDeviceId, fetch, platform); } catch { /* descriptive only */ }
map.run.deviceId = runtime.controllerDeviceId;
map.run.environment = { deviceModel: runtime.deviceModel, osVersion: runtime.osVersion, locale: runtime.locale, orientation: runtime.orientation, colorScheme: runtime.colorScheme };
map.run.status = "exploring";
@@ -357,6 +409,20 @@ export async function exploreConnectedSession(ctx: ConnectedSessionExploration):
catch { return undefined; }
};
const foregroundElsewhere = (): boolean => { const state = foregroundState(); return state ? leftForeground(state, bundleId) : false; };
let unmatchedPanel = false;
/** Whether the frame is an opened Duo's, to be stored turned upright; the run then records the landscape it shows. */
const landscapeOnPortrait = (nodes: readonly Record<string, unknown>[], png: Buffer): boolean => {
const width = png.readUInt32BE(16), height = png.readUInt32BE(20);
const turn = duoScreenshotTurn({ platform, deviceModel: runtime.deviceModel, nodes, width, height });
if (turn === "unmatched" && !unmatchedPanel) {
unmatchedPanel = true;
const root = nodes.find(node => node.type === "Application")?.rect as { width?: unknown; height?: unknown } | undefined;
say(`an iPhone Duo frame (${width}×${height} px, root ${root ? `${String(root.width)}×${String(root.height)} pt` : "missing"}) shows neither of its panels; its screenshot is stored as taken, which may be sideways`);
}
if (turn !== "turn") return false;
if (runtime.orientation !== "landscape") { runtime.orientation = "landscape"; if (map.run.environment) map.run.environment.orientation = "landscape"; }
return true;
};
const observe = async (includeImage: boolean): Promise<BlackBoxFrame> => {
let nodes: Record<string, unknown>[];
let leftApp = false;
@@ -394,7 +460,7 @@ export async function exploreConnectedSession(ctx: ConnectedSessionExploration):
assertDecodedPng(png, "Exploration screenshot");
if (png.length > 10 * 1024 * 1024) return { nodes, capturedAt, imageUnavailable: { code: "SCREENSHOT_TOO_LARGE", message: "The screenshot exceeded 10 MB." } };
if (solidColour(png)) return { nodes, capturedAt, imageUnavailable: { code: "SCREENSHOT_BLANK", message: BLANK_SCREENSHOT_MESSAGE }, ...(leftApp ? { leftApp } : {}) };
return { nodes, capturedAt, frameId: randomUUID(), png, imageSha256: createHash("sha256").update(png).digest("hex"), width: png.readUInt32BE(16), height: png.readUInt32BE(20), ...(leftApp ? { leftApp } : {}) };
return { nodes, capturedAt, frameId: randomUUID(), png, imageSha256: createHash("sha256").update(png).digest("hex"), width: png.readUInt32BE(16), height: png.readUInt32BE(20), ...(landscapeOnPortrait(nodes, png) ? { landscape: true } : {}), ...(leftApp ? { leftApp } : {}) };
} finally { await rm(path, { force: true }); }
};
const setupOptions = { command: async (args: string[]) => { const result = controller(guardPositionals([...args, "--json"]), 120_000); return controllerData(result.stdout, args[0] ?? "command"); },
@@ -427,7 +493,8 @@ export async function exploreConnectedSession(ctx: ConnectedSessionExploration):
}
const exploration = blockedAtLaunch ?? await runBlackBoxExploration({ session: model.session, store, map, app: { name: app.name, bundleId, platform, ...(app.version ? { version: app.version } : {}) },
inputs, ...(hints ? { hints: hints } : {}), ...(launchUrl ? { launchUrl: launchUrl } : {}), ...(allowDestructive ? { allowDestructive: allowDestructive } : {}), ...(ctx.appIcon ? { appIcon: ctx.appIcon } : {}),
deadlineMs, maxScreens: limits.maxScreens, maxActions: limits.maxActions, maxTurns: limits.maxActions * 3 + limits.maxScreens * 2 + 20, now, observe, progress: say,
...(model.budgetNote ? { budgetNote: model.budgetNote } : {}),
deadlineMs, maxScreens: limits.maxScreens, maxActions: limits.maxActions, maxTurns: turnBudget(limits), now, observe, progress: say,
act: async action => {
const perform = async () => {
if (action.action === "tap" && "frameId" in action) {
@@ -491,7 +558,8 @@ export async function runSessionExploration(options: SessionExploreOptions): Pro
const environment = options.environment ?? process.env;
const runner = options.runner ?? defaultRunner;
const limits = validatedLimits(options);
const inputs = hintInputs(options.hints, environment);
const replay = replayPlan(options);
const inputs = hintInputs(options.hints, environment, replay);
if (options.bundleId !== undefined && !appIdPattern.test(options.bundleId)) throw binaryError("BUNDLE_ID_INVALID", `${options.bundleId} is not a bundle or package id.`, "Pass --bundle-id with the app's bundle identifier (iOS) or package name (Android).");
let connection: SimulatorConnection;
try { connection = await resolveAttachedSimulatorSession(environment, options.envFile, options.session); }
@@ -555,7 +623,7 @@ export async function runSessionExploration(options: SessionExploreOptions): Pro
map.run.environment = { deviceModel: device };
session = { id: connection.sessionId, platform, device, name: info.name, ...(info.deviceRunSessionUrl ? { url: info.deviceRunSessionUrl } : {}) };
const model = await preflightModelSession(options, environment, say);
const model = replay ? replayModel(replay, platform, limits) : await preflightModelSession(options, environment, say);
const outputDirectory = await realpath(await mkdir(options.output, { recursive: true }).then(() => options.output));
const store = new ExploreStore(outputDirectory);
if (await store.exists()) throw binaryError("OUTPUT_NOT_EMPTY", `${outputDirectory} already holds an Explore map.`, "Pass an empty --output directory.");
@@ -572,6 +640,75 @@ export async function runSessionExploration(options: SessionExploreOptions): Pro
}
}
type LocalDevice = NonNullable<ConnectedSessionExploration["local"]>;
/**
* Starts a simulator or emulator on this machine for one run, installs the app, and hands the explorer a connection
* to it; everything is stopped afterwards. They are the devices `capture --simulator local` starts, recorded the same
* way, so the next local run stops what a run stopped hard left behind.
*/
async function withLocalDevice(options: { archive: InspectedArchive; installPath: string; platform: Platform; device: string; bundleId: string;
controller: "agent-device" | "argent"; tag: string; work: string; runner: typeof defaultRunner; environment: NodeJS.ProcessEnv; secrets: readonly string[]; say: (message: string) => void; now?: () => number },
explore: (connection: SimulatorConnection, local: LocalDevice) => Promise<void>): Promise<void> {
const { archive, installPath, platform, device, bundleId, controller, tag, work, runner, say } = options;
const environment = baseEnvironment(options.environment);
const agentDevice = agentDeviceLauncher(options.environment);
const stateDirectory = join(work, "agent-device-state");
const record = await openLocalCaptureRecord(options.environment, work);
let udid: string | undefined, serial: string | undefined, argent: LocalArgentServer | undefined;
let primaryError: unknown;
try {
let osVersion: string;
if (platform === "ios") {
say(`starting a local ${device} simulator`);
const created = await createLocalSimulator({ runner, device, name: `Explore ${device} (binary ${tag})`, cwd: work, environment,
created: async id => { udid = id; await record.update({ udid: id }); } });
osVersion = created.osVersion;
waitForLocalSimulator({ runner, udid: created.udid, cwd: work, environment });
// Links in the app's own schemes open without iOS asking first.
for (const scheme of archive.urlSchemes) approveLinkScheme({ runner, udid: created.udid, scheme, bundleId, cwd: work, environment });
say("installing the app");
installOnLocalSimulator({ runner, udid: created.udid, app: installPath, cwd: work, environment });
} else {
say(`starting a local ${device} emulator`);
const name = `explore-binary-${tag}`;
const started = await startLocalEmulator({ runner, device, name, directory: work, cwd: work, environment,
...(controller === "agent-device" ? { claimed: (candidate: string) => agentDeviceClaims({ runner, agentDevice, serial: candidate, cwd: work, environment }) } : {}),
started: async id => { serial = id; await record.update({ serial: id, avd: name }); } });
osVersion = waitForLocalEmulator({ runner, serial: started.serial, device, apiLevel: started.apiLevel, directory: work, cwd: work, environment, ...(options.now ? { now: options.now } : {}) });
say("installing the APK");
installOnLocalEmulator({ runner, serial: started.serial, apk: installPath, cwd: work, environment });
}
const deviceId = (udid ?? serial)!;
let connection: SimulatorConnection = { sessionId: `local-${deviceId}`, daemonBaseUrl: "", daemonAuthToken: "" };
if (controller === "argent") {
// Argent drives the device through a tool-server of this run's own.
const cli = await ensureArgentToolServer({ environment: options.environment, progress: say });
argent = await startLocalArgentServer({ runner, cli, directory: work, cwd: work, environment, ...(options.now ? { now: options.now } : {}) });
await record.update({ argent: { pid: argent.pid, cli: argent.cli, home: argent.home } });
connection = { ...connection, daemonBaseUrl: argent.url, daemonAuthToken: argent.token, controller: "argent" };
}
await explore(connection, { deviceId, osVersion, stateDirectory });
} catch (error) { primaryError = error; throw error; }
finally {
try {
if (argent) stopLocalArgentServer(argent);
// Its own agent-device daemon lets go of the device before it goes.
if (controller === "agent-device") runner({ command: agentDevice, args: ["close", "--session", CONTROLLER_SESSION, "--json"], cwd: work,
environment: { ...environment, AGENT_DEVICE_STATE_DIR: stateDirectory }, timeoutMs: 60_000, allowFailure: true });
await stopLocalAgentDevice(stateDirectory);
if (serial) { say(`stopping the local emulator ${serial}`); await deleteLocalEmulator({ runner, serial, cwd: work, environment }); }
if (udid) { say(`deleting the local simulator ${udid}`); deleteLocalSimulator({ runner, udid, cwd: work, environment }); }
// Everything is stopped; a cleanup that threw leaves the record for the next run's sweep.
await record.close();
} catch (error) {
throw new CiError({ code: "SESSION_CLEANUP_FAILED", stage: "cleanup", message: `The local ${serial ? "emulator" : "simulator"} Explore started could not be stopped.`,
nextAction: serial ? `Stop it with: adb -s ${serial} emu kill` : udid ? `Delete it with: xcrun simctl delete ${udid}` : "Rerun explore:binary --simulator local; it stops what this run left.",
details: { ...(serial ? { serial } : {}), ...(udid ? { udid } : {}), ...(primaryError ? { originalFailure: serializeCiError(primaryError, { secrets: options.secrets }) } : {}) }, cause: error });
}
}
}
export async function runBinaryExploration(options: BinaryExploreOptions): Promise<BinaryExploreResult> {
const startedAt = Date.now();
const now = options.now ?? Date.now;
@@ -579,21 +716,25 @@ export async function runBinaryExploration(options: BinaryExploreOptions): Promi
const environment = options.environment ?? process.env;
const runner = options.runner ?? defaultRunner;
const limits = validatedLimits(options);
const inputs = hintInputs(options.hints, environment);
const replay = replayPlan(options);
const inputs = hintInputs(options.hints, environment, replay);
const local = options.simulator === "local";
// A local emulator runs this Mac's ABI (arm64-v8a on Apple silicon); EAS's runs x86_64.
const emulatorAbi = local ? hostEmulatorAbi() : "x86_64";
say("inspecting the archive");
const archive = await inspectArchive(options.archive);
assertEmulatorCompatible(archive);
assertEmulatorCompatible(archive, [emulatorAbi]);
if (options.platform && options.platform !== archive.platform) throw binaryError("ARCHIVE_PLATFORM_MISMATCH", `The archive is an ${archive.platform} build but --platform ${options.platform} was requested.`, "Drop --platform or pass a matching archive.");
if (!environment.EXPO_TOKEN?.trim() && !environment.EXPO_SESSION_SECRET?.trim() && !(await easLoggedIn(runner, environment))) {
if (!local && !environment.EXPO_TOKEN?.trim() && !environment.EXPO_SESSION_SECRET?.trim() && !(await easLoggedIn(runner, environment))) {
throw binaryError("EXPO_LOGIN_REQUIRED", "No EAS login was found.", "Run `eas login` or set EXPO_TOKEN, then rerun explore:binary.");
}
const platform = archive.platform;
const bundleId = options.bundleId ?? archive.bundleId;
const device = options.device ?? (platform === "ios" ? DEFAULT_IOS_DEVICE : DEFAULT_ANDROID_DEVICE);
say(`${archive.appName} ${archive.version ?? ""} (${bundleId}, ${platform}, ${(archive.bytes / 1048576).toFixed(0)} MB, sha256 ${archive.sha256.slice(0, 12)})`);
if (archive.missingLibs?.length) say(`warning: lib/x86_64 lacks ${archive.missingLibs.slice(0, 4).join(", ")}${archive.missingLibs.length > 4 ? ` and ${archive.missingLibs.length - 4} more` : ""} that another ABI ships; the app may stop with UnsatisfiedLinkError on the emulator`);
if (archive.missingLibs?.length && emulatorAbi === "x86_64") say(`warning: lib/x86_64 lacks ${archive.missingLibs.slice(0, 4).join(", ")}${archive.missingLibs.length > 4 ? ` and ${archive.missingLibs.length - 4} more` : ""} that another ABI ships; the app may stop with UnsatisfiedLinkError on the emulator`);
const model = await preflightModelSession(options, environment, say);
const model = replay ? replayModel(replay, platform, limits) : await preflightModelSession(options, environment, say);
const outputDirectory = await realpath(await mkdir(options.output, { recursive: true }).then(() => options.output));
const store = new ExploreStore(outputDirectory);
@@ -601,7 +742,7 @@ export async function runBinaryExploration(options: BinaryExploreOptions): Promi
const work = await realpath(await mkdtemp(join(tmpdir(), "expo-explore-binary-")));
const easEnvironment = { ...baseEnvironment(environment), ...(environment.EXPO_TOKEN ? { EXPO_TOKEN: environment.EXPO_TOKEN } : {}), ...(environment.HOME ? { HOME: environment.HOME } : {}), CI: "1", EXPO_NO_TELEMETRY: "1" };
const easBin = packageBin("eas-cli", "eas");
await lstat(easBin).catch(() => { throw binaryError("EXPLORE_TOOLS_MISSING", "The EAS CLI that Explore pins is not installed.", "Reinstall @expo/explore with optional dependencies."); });
if (!local) await lstat(easBin).catch(() => { throw binaryError("EXPLORE_TOOLS_MISSING", "The EAS CLI that Explore pins is not installed.", "Reinstall @expo/explore with optional dependencies."); });
const eas = (args: string[], timeoutMs = 5 * 60_000): CommandResult & { status: number } => {
try { return { ...runner({ command: process.execPath, args: [easBin, ...args], cwd: work, environment: easEnvironment, timeoutMs }), status: 0 }; }
catch (error) { return { stdout: "", stderr: error instanceof Error ? error.message : String(error), status: 1 }; }
@@ -615,6 +756,11 @@ export async function runBinaryExploration(options: BinaryExploreOptions): Promi
const controller = options.controller ?? "agent-device";
// Fetched before the paid session starts, so a blocked download costs nothing.
if (!options.runner) await ensureController(controller, { environment, progress: say });
if (local) {
// What an earlier local run stopped hard left running (a simulator, an emulator, a daemon) goes first.
const swept = await sweepLocalCaptures({ environment, runner, cwd: work }).catch(() => [] as string[]);
if (swept.length) say(`stopped what an interrupted local run left running: ${swept.join(", ")}`);
}
const map: ExploreMap = { contract: CONTRACT, revision: 0, run: { id: randomUUID(), name: archive.appName, platform, controller, exploration: "black-box", bundleId, status: "draft",
createdAt: new Date(now()).toISOString(), updatedAt: new Date(now()).toISOString(),
build: { ...(archive.version ? { version: archive.version } : {}), ...(archive.buildNumber ? { number: archive.buildNumber } : {}), hash: archive.sha256, source: "local" },
@@ -622,57 +768,83 @@ export async function runBinaryExploration(options: BinaryExploreOptions): Promi
try {
let installPath = archive.path;
if (archive.kind === "ios-tgz" || archive.kind === "ios-ipa") { say("unpacking the iOS bundle"); installPath = await extractIosBundle(archive, join(work, "bundle")); }
if (!owner) {
const who = eas(["whoami"], 60_000);
owner = who.status === 0 ? whoamiAccounts(who.stdout)[0] ?? "" : "";
if (!owner) throw binaryError("EXPO_LOGIN_REQUIRED", "EAS did not report an account for this login.", "Run `eas login` (or set EXPO_TOKEN) and rerun, or pass --owner.");
}
await writeFile(join(work, "package.json"), JSON.stringify({ name: project, version: "0.0.0", private: true }), { mode: 0o600 });
await writeFile(join(work, "app.json"), JSON.stringify({ expo: { name: project, slug: project, owner, web: { output: "server" } } }), { mode: 0o600 });
const init = eas(["init", "--force", "--non-interactive", "--json"], 2 * 60_000);
const appJson = JSON.parse(await readFile(join(work, "app.json"), "utf8")) as { expo?: { extra?: { eas?: { projectId?: string } } } };
const projectId = appJson.expo?.extra?.eas?.projectId;
if (init.status !== 0 || !projectId) throw binaryError("EXPLORE_PROJECT_FAILED", `EAS could not create or link @${owner}/${project}: ${lastLine(init.stderr || init.stdout)}`,
"Give this login permission to create projects under the owner, or pass --project with an existing slug.", { owner, project });
const projectIdentity = { name: project, slug: project, owner, projectId };
say(`starting an EAS Simulator session (${device}, ${limits.maxDurationMinutes} min)`);
const sessionName = `explore-binary-${createHash("sha256").update(`${archive.sha256}:${map.run.id}`).digest("hex").slice(0, 16)}`;
const dotenvPath = join(work, SIMULATOR_DOTENV_FILE_NAME);
await writeFile(dotenvPath, SIMULATOR_DOTENV_HEADER, { flag: "wx", mode: 0o600 });
let startError: unknown;
try {
const started = eas(["simulator:start", "--name", sessionName, "--platform", platform, "--device", device, "--type", controller, "--package-version", controllerVersion(controller), "--public-link",
"--max-duration-minutes", String(limits.maxDurationMinutes + 5), "--out-config-type", "dotenv", "--non-interactive"], 20 * 60_000);
if (started.status !== 0) throw new Error(started.stderr || started.stdout || "eas simulator:start failed");
} catch (error) { startError = error; }
const handoff = await readSimulatorDotenvHandoff(dotenvPath);
const secrets = [environment.EXPO_TOKEN ?? "", ...(handoff.connection ? [handoff.connection.daemonAuthToken] : []), ...inputs.values].filter(Boolean);
const startFailure = startError ? simulatorStartFailure(startError, secrets) : undefined;
if (handoff.sessionId) {
try { assertSimulatorSessionIdentity(eas(["simulator:get", "--id", handoff.sessionId, "--json", "--non-interactive"], 2 * 60_000).stdout, handoff.sessionId, sessionName, projectIdentity, handoff.connection, platform); }
catch (error) { throw new CiError({ code: "SESSION_HANDOFF_UNVERIFIED", stage: "session", message: "The EAS handoff could not prove ownership of the exact simulator session. The private handoff was preserved.",
nextAction: "Inspect the EAS session list; confirm ownership before stopping the reported session.", details: { handoffDirectory: work, reportedSessionId: handoff.sessionId, ...(startFailure ? { startFailure } : {}) }, cause: error }); }
sessionId = handoff.sessionId;
}
if (!sessionId) throw new CiError({ code: startFailure ? "SIMULATOR_START_FAILED" : "SESSION_HANDOFF_UNVERIFIED", stage: "session",
message: startFailure ? `EAS simulator:start failed: ${startFailure.message}` : "EAS did not provide a verifiable simulator session.",
nextAction: "Inspect the EAS start failure and session list; verify a session's exact ID and ownership before cleanup.", details: { handoffDirectory: work, ...(startFailure ? { startFailure } : {}) }, cause: startError });
if (startFailure) throw new CiError({ code: "SIMULATOR_START_FAILED", stage: "session", message: `EAS simulator:start failed after the session was verified: ${startFailure.message}`,
nextAction: "Inspect the start failure; Explore stops the verified session before returning.", details: { sessionId, startFailure }, cause: startError });
if (handoff.validationError) throw handoff.validationError;
if (!handoff.connection) throw new Error("Simulator handoff is missing its controller connection");
await rm(dotenvPath, { force: true });
const appIcon = await readArchiveIcon(archive);
({ runtime, exploration } = await exploreConnectedSession({ connection: handoff.connection, platform, device, bundleId, ...(appIcon ? { appIcon } : {}), app: { name: archive.appName, ...(archive.version ? { version: archive.version } : {}) },
install: { path: installPath, kind: archive.kind }, work, runner, environment, secrets, inputs, say, now, limits, store, map, model, urlSchemes: archive.urlSchemes,
sessionJson: () => eas(["simulator:get", "--id", handoff.connection!.sessionId, "--json", "--non-interactive"], 2 * 60_000).stdout,
...(options.launchUrl ? { launchUrl: options.launchUrl } : {}), ...(options.hints ? { hints: options.hints } : {}), ...(options.allowDestructive ? { allowDestructive: options.allowDestructive } : {}) }));
const exploreOn = (connection: SimulatorConnection, sessionJson: () => string, localDevice?: LocalDevice, afterInstall?: () => void) => exploreConnectedSession({ connection, platform, device, bundleId,
...(afterInstall ? { afterInstall } : {}),
...(appIcon ? { appIcon } : {}), app: { name: archive.appName, ...(archive.version ? { version: archive.version } : {}) },
// A local device has the app installed by simctl or adb before the controller starts.
...(localDevice ? { local: localDevice } : { install: { path: installPath, kind: archive.kind } }),
work, runner, environment, secrets: [environment.EXPO_TOKEN ?? "", connection.daemonAuthToken, ...inputs.values].filter(Boolean), inputs, say, now, limits, store, map, model,
urlSchemes: archive.urlSchemes, sessionJson,
...(options.launchUrl ? { launchUrl: options.launchUrl } : {}), ...(options.hints ? { hints: options.hints } : {}), ...(options.allowDestructive ? { allowDestructive: options.allowDestructive } : {}) });
if (local) {
const tag = map.run.id.replace(/-/g, "").slice(0, 8);
await withLocalDevice({ archive, installPath, platform, device, bundleId, controller, tag, work, runner, environment, say, secrets: inputs.values, ...(options.now ? { now: options.now } : {}) },
async (connection, localDevice) => {
sessionId = connection.sessionId;
({ runtime, exploration } = await exploreOn(connection, () => "", localDevice));
});
} else {
if (!owner) {
const who = eas(["whoami"], 60_000);
owner = who.status === 0 ? whoamiAccounts(who.stdout)[0] ?? "" : "";
if (!owner) throw binaryError("EXPO_LOGIN_REQUIRED", "EAS did not report an account for this login.", "Run `eas login` (or set EXPO_TOKEN) and rerun, or pass --owner.");
}
await writeFile(join(work, "package.json"), JSON.stringify({ name: project, version: "0.0.0", private: true }), { mode: 0o600 });
await writeFile(join(work, "app.json"), JSON.stringify({ expo: { name: project, slug: project, owner, web: { output: "server" } } }), { mode: 0o600 });
const init = eas(["init", "--force", "--non-interactive", "--json"], 2 * 60_000);
const appJson = JSON.parse(await readFile(join(work, "app.json"), "utf8")) as { expo?: { extra?: { eas?: { projectId?: string } } } };
const projectId = appJson.expo?.extra?.eas?.projectId;
if (init.status !== 0 || !projectId) throw binaryError("EXPLORE_PROJECT_FAILED", `EAS could not create or link @${owner}/${project}: ${lastLine(init.stderr || init.stdout)}`,
"Give this login permission to create projects under the owner, or pass --project with an existing slug.", { owner, project });
const projectIdentity = { name: project, slug: project, owner, projectId };
say(`starting an EAS Simulator session (${device}, ${limits.maxDurationMinutes} min)`);
const sessionName = `explore-binary-${createHash("sha256").update(`${archive.sha256}:${map.run.id}`).digest("hex").slice(0, 16)}`;
const dotenvPath = join(work, SIMULATOR_DOTENV_FILE_NAME);
await writeFile(dotenvPath, SIMULATOR_DOTENV_HEADER, { flag: "wx", mode: 0o600 });
let startError: unknown;
try {
const started = eas(["simulator:start", "--name", sessionName, "--platform", platform, "--device", device, "--type", controller, "--package-version", controllerVersion(controller), "--public-link",
"--max-duration-minutes", String(limits.maxDurationMinutes + 5), "--out-config-type", "dotenv", "--non-interactive"], 20 * 60_000);
if (started.status !== 0) throw new Error(started.stderr || started.stdout || "eas simulator:start failed");
} catch (error) { startError = error; }
const handoff = await readSimulatorDotenvHandoff(dotenvPath);
const secrets = [environment.EXPO_TOKEN ?? "", ...(handoff.connection ? [handoff.connection.daemonAuthToken] : []), ...inputs.values].filter(Boolean);
const startFailure = startError ? simulatorStartFailure(startError, secrets) : undefined;
if (handoff.sessionId) {
try { assertSimulatorSessionIdentity(eas(["simulator:get", "--id", handoff.sessionId, "--json", "--non-interactive"], 2 * 60_000).stdout, handoff.sessionId, sessionName, projectIdentity, handoff.connection, platform); }
catch (error) { throw new CiError({ code: "SESSION_HANDOFF_UNVERIFIED", stage: "session", message: "The EAS handoff could not prove ownership of the exact simulator session. The private handoff was preserved.",
nextAction: "Inspect the EAS session list; confirm ownership before stopping the reported session.", details: { handoffDirectory: work, reportedSessionId: handoff.sessionId, ...(startFailure ? { startFailure } : {}) }, cause: error }); }
sessionId = handoff.sessionId;
}
if (!sessionId) throw new CiError({ code: startFailure ? "SIMULATOR_START_FAILED" : "SESSION_HANDOFF_UNVERIFIED", stage: "session",
message: startFailure ? `EAS simulator:start failed: ${startFailure.message}` : "EAS did not provide a verifiable simulator session.",
nextAction: "Inspect the EAS start failure and session list; verify a session's exact ID and ownership before cleanup.", details: { handoffDirectory: work, ...(startFailure ? { startFailure } : {}) }, cause: startError });
if (startFailure) throw new CiError({ code: "SIMULATOR_START_FAILED", stage: "session", message: `EAS simulator:start failed after the session was verified: ${startFailure.message}`,
nextAction: "Inspect the start failure; Explore stops the verified session before returning.", details: { sessionId, startFailure }, cause: startError });
if (handoff.validationError) throw handoff.validationError;
if (!handoff.connection) throw new Error("Simulator handoff is missing its controller connection");
await rm(dotenvPath, { force: true });
// The app's own links open without iOS asking first, when the session's host can approve them (serve-sim 0.10.0).
const verifiedSession = sessionId;
const sessionEas: EasCommand = (args, commandOptions) => {
const result = eas(args, commandOptions?.timeoutMs);
if (result.status !== 0 && !commandOptions?.allowFailure) throw new Error(lastLine(result.stderr || result.stdout) || `eas ${args[0]} failed`);
return result;
};
const afterInstall = platform === "ios" && archive.urlSchemes.length ? () => {
const approved = approveSessionUrlSchemes(sessionEas, verifiedSession, simulatorControlActions(sessionEas, verifiedSession), bundleId, archive.urlSchemes);
if (approved.length) say(`approved ${approved.map(item => `${item}://`).join(", ")} links on the session`);
} : undefined;
({ runtime, exploration } = await exploreOn(handoff.connection, () => eas(["simulator:get", "--id", handoff.connection!.sessionId, "--json", "--non-interactive"], 2 * 60_000).stdout, undefined, afterInstall));
}
} catch (error) {
primaryError = toCiError(error, { code: "EXPLORE_BINARY_FAILED", stage: "exploration", nextAction: "Inspect the diagnostic, correct the failing prerequisite, and rerun explore:binary." });
} finally {
if (sessionId) {
// A local device is stopped by withLocalDevice; only the work directory is left.
if (local) await rm(work, { recursive: true, force: true });
else if (sessionId) {
try {
say(`stopping simulator session ${sessionId}`);
assertSimulatorStopped(eas(["simulator:stop", "--id", sessionId, "--json", "--non-interactive"], 2 * 60_000).stdout, sessionId);
@@ -734,18 +906,7 @@ export function assertAppUrl(url: string, appSchemes: readonly string[] = []): v
export const BLANK_SCREENSHOT_MESSAGE = "The app blocks screen capture on this screen (Android FLAG_SECURE), so its screenshot came back blank.";
/** A capture-protected screen comes back as a single flat colour (black on Android); every 97th pixel is enough to tell. */
export function solidColour(png: Buffer): boolean {
let frame: DecodedFrame;
try { frame = createRequire(import.meta.url.replace(/\?.*$/, ""))("pngjs").PNG.sync.read(png, { checkCRC: false }) as DecodedFrame; }
catch { return false; }
const { data } = frame;
if (data.length < 4) return false;
for (let offset = 0; offset + 3 < data.length; offset += 4 * 97) {
if (data[offset] !== data[0] || data[offset + 1] !== data[1] || data[offset + 2] !== data[2]) return false;
}
return true;
}
export { solidColour };
/** Setup commands whose last positional is free text (`fill <ref> <text>`, `type <text>`): move the flags ahead of a `--` separator. */
export function guardPositionals(args: readonly string[]): string[] {
@@ -762,6 +923,27 @@ export function guardPositionals(args: readonly string[]): string[] {
const KEYBOARD_RETURN_KEY = /^(return|next|done|go|continue|join)$/i;
/**
* How a frame's screenshot is stored. An opened (or half-open) iPhone Duo lights its inner panel, whose native axes are
* portrait, and shows a landscape UI on it: the tree and taps stay on the portrait axes (a 669×951 pt root, the transpose
* of `DUO_PANELS.inner`), and a screenshot on those axes is sideways to a person, so it is turned ("turn"). Local
* agent-device runs returned exactly that (2007×2853 px). Capture's Argent screenshots on EAS, taken with Argent's
* LandscapeRight rotation, came back upright (2853×2007); an image that is already the root's transpose is left as it is.
* The root is the full-screen Application node capture takes its coordinate space from, matched to within a pixel and a
* point. A Duo frame on neither panel is "unmatched": it is stored as taken, and the caller says so. An empty tree, read
* mid-draw, has nothing to judge.
*/
export function duoScreenshotTurn(frame: { platform: Platform; deviceModel: string; nodes: readonly Record<string, unknown>[]; width: number; height: number }): "turn" | "keep" | "unmatched" {
if (frame.platform !== "ios" || !/\bduo\b/i.test(frame.deviceModel) || !frame.nodes.length) return "keep";
const space = screenSpace(frame.nodes, frame, 1);
if (!space) return "unmatched";
const near = (value: number, points: number) => Math.abs(value - points) < 1;
const portraitAxes = near(space.width, DUO_PANELS.inner.height) && near(space.height, DUO_PANELS.inner.width);
if (portraitAxes && !space.transposed && frame.width < frame.height) return "turn";
const panel = Object.values(DUO_PANELS).some(({ width, height }) => (near(space.width, width) && near(space.height, height)) || (near(space.width, height) && near(space.height, width)));
return panel ? "keep" : "unmatched";
}
/** A node that is not the app's UI: another package drew it (Android's system UI over the app), or it is Explore's own route marker, which an app
* instrumented for PR runs renders as a 2x2 hidden text. */
export function foreignNode(node: Record<string, unknown>, bundleId: string): boolean {
+9 -1
View File
@@ -1,4 +1,4 @@
import type { CaptureOutcome } from "../contracts/capture.js";
import type { CaptureBundle, CaptureOutcome } from "../contracts/capture.js";
export type ColorScheme = "light" | "dark";
/** `auto` follows the app's `userInterfaceStyle`; a list captures exactly those schemes, first one primary. */
@@ -118,3 +118,11 @@ export function appearancePairs(routeId: string, headSchemes: readonly ColorSche
baseId: baseSchemes.includes(scheme) ? schemeScreenId(routeId, scheme === baseSchemes[0] ? undefined : scheme) : null,
}));
}
/** A `<id>@<scheme>` outcome recording that its whole appearance pass failed (`sessionFailures`), whatever the primary did. */
export function failedPassOutcome(bundle: Pick<CaptureBundle, "outcomes" | "sessionFailures">, id: string): boolean {
const { scheme } = routeIdOfScreen(id);
const own = bundle.outcomes[id];
return scheme !== undefined && own?.status === "failed" && (bundle.sessionFailures ?? []).some(failure => failure.scheme === scheme
&& failure.error.code === own.error?.code && failure.error.message === own.error?.message);
}
+101
View File
@@ -0,0 +1,101 @@
/**
* Runs the Argent bridge's commands on a worker thread of this process, through a synchronous runner, so the
* capture's command-at-a-time control flow keeps one bridge and one tool-server connection for the whole run
* instead of starting a process per command. Commands for any other executable go to the fallback runner.
*/
import { Worker } from "node:worker_threads";
import { commandOutcome } from "./controller.js";
import type { CommandRequest, CommandResult, CommandRunner } from "./run.js";
import { BRIDGE_HEADER_BYTES, BRIDGE_KIND_UNAVAILABLE, BRIDGE_SLOT_KIND, BRIDGE_SLOT_LENGTH, BRIDGE_SLOT_READY, type BridgeRequest, type BridgeWorkerData } from "./argent-bridge-worker.js";
/** Room for one answer: an accessibility tree of a large screen is a few megabytes. */
const BRIDGE_BUFFER_BYTES = 64 * 1024 * 1024;
/** How long a command gets when its request names no timeout. */
const DEFAULT_COMMAND_TIMEOUT_MS = 10 * 60_000;
/** After a timeout, how long the thread gets to confirm it dropped the command. */
const ABORT_GRACE_MS = 15_000;
export type ArgentBridgeRunner = {
runner: CommandRunner;
/** Whether commands still go to the thread (false after it proved unavailable or stopped answering). */
readonly active: boolean;
close(): void;
};
export function startArgentBridgeRunner(options: {
/** The bridge executable the callers name; its commands run on the thread. */
command: string;
/** The bridge's environment, as argentControllerInvocation builds it. */
environment: NodeJS.ProcessEnv;
fallback: CommandRunner;
/** Tests: a module exporting `createArgentClient` in place of the pinned Argent release's client. */
clientModule?: string;
progress?: (message: string) => void;
}): ArgentBridgeRunner {
const shared = new SharedArrayBuffer(BRIDGE_HEADER_BYTES + BRIDGE_BUFFER_BYTES);
const header = new Int32Array(shared, 0, BRIDGE_HEADER_BYTES / 4);
const bytes = new Uint8Array(shared, BRIDGE_HEADER_BYTES);
const workerData: BridgeWorkerData = { shared, ...(options.clientModule ? { clientModule: options.clientModule } : {}) };
let worker: Worker | undefined;
// The environment the thread runs with: a copy, so a caller that changes the session's (a restarted local
// tool-server's URL and token) is seen by the next command, which starts a thread with the new one.
let environment: NodeJS.ProcessEnv = {};
const start = (from: NodeJS.ProcessEnv) => {
environment = { ...from };
try {
const started = new Worker(new URL("./argent-bridge-worker.js", import.meta.url), { workerData, env: environment });
// A thread that crashes stops serving; the commands after it run per command.
started.on("error", error => { if (worker === started) stop(`the Argent bridge thread failed (${error.message}); running the bridge per command`); });
worker = started;
} catch (error) { options.progress?.(`the Argent bridge thread could not start (${(error as Error).message}); running the bridge per command`); }
};
let sequence = 0;
const stop = (reason: string) => {
if (!worker) return;
options.progress?.(reason);
void worker.terminate();
worker = undefined;
};
const changed = (next: NodeJS.ProcessEnv) => {
const keys = new Set([...Object.keys(environment), ...Object.keys(next)]);
return [...keys].some(key => environment[key] !== next[key]);
};
const waitFor = (id: number, timeoutMs: number): boolean => {
const deadline = Date.now() + timeoutMs;
for (;;) {
// The value waited on is the one just compared: an answer stored after the comparison wakes the wait at once.
const ready = Atomics.load(header, BRIDGE_SLOT_READY);
if (ready === id) return true;
const remaining = deadline - Date.now();
if (remaining <= 0) return false;
Atomics.wait(header, BRIDGE_SLOT_READY, ready, remaining);
}
};
start(options.environment);
const runner: CommandRunner = (request: CommandRequest): CommandResult => {
// A recording runs per command, in the session's own directory: Argent's client saves one under the project its
// working directory is in, and this thread's is the caller's checkout. The state file carries it between hosts.
if (!worker || request.command !== options.command || request.args[0] === "record") return options.fallback(request);
if (changed(request.environment)) { void worker.terminate(); worker = undefined; start(request.environment); }
if (!worker) return options.fallback(request);
const id = ++sequence;
Atomics.store(header, BRIDGE_SLOT_READY, 0);
worker.postMessage({ id, argv: request.args, ...(request.timeoutMs !== undefined ? { timeoutMs: request.timeoutMs } : {}) } satisfies BridgeRequest);
const timeoutMs = request.timeoutMs ?? DEFAULT_COMMAND_TIMEOUT_MS;
if (!waitFor(id, timeoutMs)) {
worker.postMessage({ abort: id } satisfies BridgeRequest);
if (!waitFor(id, ABORT_GRACE_MS)) stop("the Argent bridge thread stopped answering; running the bridge per command");
throw Object.assign(new Error(`${request.command} ${request.args[0] ?? ""} timed out after ${timeoutMs} ms`), { code: "ETIMEDOUT" });
}
const text = Buffer.from(bytes.subarray(0, Atomics.load(header, BRIDGE_SLOT_LENGTH))).toString("utf8");
if (Atomics.load(header, BRIDGE_SLOT_KIND) === BRIDGE_KIND_UNAVAILABLE) {
stop(`the Argent bridge thread is unavailable (${text}); running the bridge per command`);
return options.fallback(request);
}
const result = JSON.parse(text) as { exitCode: number; stdout: string; notes?: string[] };
// The thread cannot reach the capture's log while this one waits for it, so its lines come with the answer.
for (const note of result.notes ?? []) options.progress?.(note);
return commandOutcome(request, { status: result.exitCode, stdout: result.stdout, stderr: "" });
};
return { runner, get active() { return worker !== undefined; }, close: () => { if (worker) { void worker.terminate(); worker = undefined; } } };
}
+159
View File
@@ -0,0 +1,159 @@
/**
* The Argent bridge on a worker thread: one bridge and one tool-server connection for a whole capture, instead of a
* process (and a TLS handshake) per command. The capture thread posts a command and blocks on the shared buffer;
* this thread runs it through Argent's own client library (the code `argent run` uses) and writes the answer back.
*/
import { isMainThread, parentPort, workerData } from "node:worker_threads";
import { copyFile, mkdir, open, writeFile } from "node:fs/promises";
import { dirname, join } from "node:path";
import { pathToFileURL } from "node:url";
import { ArgentCallError, type ArgentCall } from "./argent-driver.js";
import { argentBridgeSettings, createArgentBridge, type ArgentBridge } from "./argent-bridge.js";
import { ensureController } from "../shared/controller-cache.js";
/** The shared buffer: four Int32 header slots, then the UTF-8 answer. */
export const BRIDGE_HEADER_BYTES = 16;
export const BRIDGE_SLOT_READY = 0;
export const BRIDGE_SLOT_LENGTH = 1;
export const BRIDGE_SLOT_KIND = 2;
export const BRIDGE_KIND_RESULT = 0;
/** The bridge cannot run on this thread (Argent's client did not load); the host falls back to the process bridge. */
export const BRIDGE_KIND_UNAVAILABLE = 1;
export type BridgeWorkerData = { shared: SharedArrayBuffer; clientModule?: string };
export type BridgeRequest = { id: number; argv: readonly string[]; timeoutMs?: number } | { abort: number };
const ARGENT_CALL_TIMEOUT_MS = 180_000;
/** Argent's programmatic client, as `dist/client.js` of the pinned release exports it. */
type ArgentClient = { callTool(name: string, args: Record<string, unknown>, options?: { signal?: AbortSignal }): Promise<{ data: unknown; note?: string }> };
/** Whether some bytes begin as a PNG or a JPEG does: `--out` is a screenshot, and nothing else is taken for one. */
function isImage(head: Uint8Array): boolean {
const png = [0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a], jpeg = [0xff, 0xd8, 0xff];
return [png, jpeg].some(signature => signature.every((byte, index) => head[index] === byte));
}
/** Whether a path names a file that is an image by its content. */
async function imageFile(path: string): Promise<boolean> {
const file = await open(path, "r").catch(() => undefined);
if (!file) return false;
try {
if (!(await file.stat()).isFile()) return false;
const head = new Uint8Array(8);
const { bytesRead } = await file.read(head, 0, head.length, 0);
return isImage(head.subarray(0, bytesRead));
} finally { await file.close(); }
}
/** The first string in a tool's answer that names an image file: the materialized screenshot (`path` first). */
async function materializedFile(data: unknown): Promise<string | undefined> {
const candidates: string[] = [];
const walk = (value: unknown, key?: string) => {
if (typeof value === "string") { if (key === "path") candidates.unshift(value); else candidates.push(value); }
else if (Array.isArray(value)) value.forEach(item => walk(item));
else if (value && typeof value === "object") Object.entries(value).forEach(([k, v]) => walk(v, k));
};
walk(data);
for (const candidate of candidates) {
if (candidate.startsWith("/") && await imageFile(candidate)) return candidate;
}
return undefined;
}
/**
* An error's message with its cause's, as Argent's client words a failed fetch: "fetch failed: connect ECONNREFUSED …".
* A host that resolves to several addresses (localhost: ::1 and 127.0.0.1) fails with an AggregateError whose message is
* empty, so the causes' codes are named too.
*/
export function failureText(error: unknown): string {
if (!(error instanceof Error)) return String(error);
const cause = error.cause;
if (!(cause instanceof Error)) return error.message;
const codes = [cause, ...(cause instanceof AggregateError ? cause.errors : [])]
.map(item => (item as { code?: unknown } | undefined)?.code).filter((code): code is string => typeof code === "string" && !cause.message.includes(code));
return [`${error.message}:`, cause.message, ...(codes.length ? [`(${[...new Set(codes)].join(", ")})`] : [])].filter(Boolean).join(" ");
}
async function loadClient(environment: NodeJS.ProcessEnv, clientModule?: string): Promise<ArgentClient> {
const module = clientModule ?? join(dirname(await ensureController("argent", { environment })), "client.js");
const loaded = await import(pathToFileURL(module).href) as { createArgentClient?: () => ArgentClient };
if (typeof loaded.createArgentClient !== "function") throw new Error(`${module} does not export createArgentClient`);
return loaded.createArgentClient();
}
/** The in-thread Argent call: the client's `callTool`, with `--out` served from the artifact it materialized. */
export function argentClientCall(client: ArgentClient, signal: () => AbortSignal): ArgentCall {
return async (tool, args, options) => {
const active = signal();
if (active.aborted) throw new ArgentCallError(`argent run ${tool} was not sent: the command was given up.`, false);
let answer: { data: unknown };
const timeoutMs = Math.max(1, Math.ceil(Math.min(ARGENT_CALL_TIMEOUT_MS, options?.timeoutMs ?? ARGENT_CALL_TIMEOUT_MS)));
try { answer = await client.callTool(tool, args, { signal: AbortSignal.any([active, AbortSignal.timeout(timeoutMs)]) }); }
catch (error) { throw new ArgentCallError(`argent run ${tool} failed: ${failureText(error).slice(0, 800)}`, true); }
if (!options?.out) return answer.data;
const file = await materializedFile(answer.data);
if (!file) {
const url = (answer.data as { url?: unknown } | null)?.url;
if (typeof url !== "string") throw new ArgentCallError(`argent run ${tool} did not write ${options.out}`, true);
const response = await fetch(url, { signal: signal() });
if (!response.ok) throw new ArgentCallError(`argent run ${tool} did not write ${options.out}: ${response.status}`, true);
const body = Buffer.from(await response.arrayBuffer());
if (!isImage(body)) throw new ArgentCallError(`argent run ${tool} did not write ${options.out}: the answer is not an image`, true);
await mkdir(dirname(options.out), { recursive: true });
await writeFile(options.out, body);
return { path: options.out };
}
await mkdir(dirname(options.out), { recursive: true });
if (file !== options.out) await copyFile(file, options.out);
return { path: options.out };
};
}
if (!isMainThread && parentPort && (workerData as Partial<BridgeWorkerData> | null)?.shared instanceof SharedArrayBuffer) {
const port = parentPort;
const { shared, clientModule } = workerData as BridgeWorkerData;
const header = new Int32Array(shared, 0, BRIDGE_HEADER_BYTES / 4);
const bytes = new Uint8Array(shared, BRIDGE_HEADER_BYTES);
const answer = (id: number, kind: number, text: string) => {
let encoded = Buffer.from(text, "utf8");
if (encoded.length > bytes.length) {
encoded = Buffer.from(JSON.stringify({ exitCode: 1, stdout: JSON.stringify({ success: false,
error: { code: "COMMAND_FAILED", message: `The controller's answer (${encoded.length} bytes) exceeds the bridge buffer.` } }) }));
}
bytes.set(encoded);
Atomics.store(header, BRIDGE_SLOT_LENGTH, encoded.length);
Atomics.store(header, BRIDGE_SLOT_KIND, kind);
Atomics.store(header, BRIDGE_SLOT_READY, id);
Atomics.notify(header, BRIDGE_SLOT_READY);
};
const controllers = new Map<number, AbortController>();
let current: AbortController | undefined;
const settings = argentBridgeSettings(process.env);
const bridge: Promise<ArgentBridge> = settings && settings.toolsUrl
? loadClient(process.env, clientModule).then(client => createArgentBridge(settings, argentClientCall(client, () => current?.signal ?? new AbortController().signal), "thread"))
: Promise.reject(new Error("The Argent bridge needs EXPLORE_ARGENT_STATE, EXPLORE_ARGENT_PLATFORM and ARGENT_TOOLS_URL."));
bridge.catch(() => undefined);
let queue: Promise<unknown> = Promise.resolve();
port.on("message", (request: BridgeRequest) => {
if ("abort" in request) { controllers.get(request.abort)?.abort(new Error("The capture gave up waiting for this command.")); return; }
const controller = new AbortController();
controllers.set(request.id, controller);
queue = queue.then(async () => {
current = controller;
try {
const ready = await bridge.catch(error => { throw Object.assign(error as Error, { unavailable: true }); });
if (controller.signal.aborted) throw controller.signal.reason;
const result = await ready.run(request.argv);
answer(request.id, BRIDGE_KIND_RESULT, JSON.stringify(result));
} catch (error) {
const message = error instanceof Error ? error.message : String(error);
if ((error as { unavailable?: boolean }).unavailable) answer(request.id, BRIDGE_KIND_UNAVAILABLE, message);
else answer(request.id, BRIDGE_KIND_RESULT, JSON.stringify({ exitCode: 1, stdout: JSON.stringify({ success: false, error: { code: "COMMAND_FAILED", message } }) }));
} finally {
controllers.delete(request.id);
current = undefined;
}
});
});
}
+154
View File
@@ -0,0 +1,154 @@
/**
* The Argent bridge behind both of its hosts: the command-line worker Explore spawns per command, and the
* in-process worker thread that keeps one bridge (and one tool-server connection) for a whole capture. Each command
* reads the shared state file, runs through `runBridgeCommand`, and writes the state back, so the two hosts can
* take turns on one session.
*/
import { spawnSync } from "node:child_process";
import { appendFile, copyFile, mkdir, readFile, realpath, rename, rm, writeFile } from "node:fs/promises";
import { dirname, join } from "node:path";
import { createRequire } from "node:module";
import type { DecodedFrame } from "../shared/png-evidence.js";
import { runBridgeCommand, type ArgentBridgeState, type ArgentCall, type ArgentPlatform, type ArgentTree, type BridgeResult } from "./argent-driver.js";
import { ARGENT_VERSION } from "../shared/argent-version.js";
export type ArgentBridgeSettings = {
environment: NodeJS.ProcessEnv;
statePath: string;
platform: ArgentPlatform;
toolsUrl: string | undefined;
};
/** Reads the settings the bridge takes from its environment; absent when they are incomplete. */
export function argentBridgeSettings(environment: NodeJS.ProcessEnv): ArgentBridgeSettings | undefined {
const statePath = environment.EXPLORE_ARGENT_STATE;
const platform = environment.EXPLORE_ARGENT_PLATFORM;
if (!statePath || (platform !== "ios" && platform !== "android")) return undefined;
return { environment, statePath, platform, toolsUrl: environment.ARGENT_TOOLS_URL };
}
async function readState(path: string): Promise<ArgentBridgeState> {
try { return JSON.parse(await readFile(path, "utf8")) as ArgentBridgeState; }
catch { return { generation: 0 }; }
}
/** Written whole and renamed into place, so a command stopped mid-write never leaves a torn state. */
async function writeState(path: string, state: ArgentBridgeState): Promise<void> {
const temporary = `${path}.${process.pid}.tmp`;
await writeFile(temporary, JSON.stringify(state), { mode: 0o600 });
await rename(temporary, path);
}
/** A call's arguments for the debug log: text that could be typed into a field is logged by length only. */
function loggedArgs(args: Record<string, unknown>): Record<string, unknown> {
return Object.fromEntries(Object.entries(args).map(([key, value]) => key === "text" && typeof value === "string" ? [key, `<${value.length} characters>`]
: key === "steps" && Array.isArray(value) ? [key, `<${value.length} steps>`] : [key, value]));
}
/** What a call answered, for the debug log: a tree's shape, not its contents. */
function summarize(tool: string, value: unknown): unknown {
if (!value || typeof value !== "object") return value;
if (tool !== "ui-tree") return Object.keys(value);
const tree = value as ArgentTree;
let nodes = 0;
const walk = (node: { children?: unknown[] }) => { nodes++; (node.children ?? []).forEach(child => walk(child as { children?: unknown[] })); };
(tree.roots ?? []).forEach(walk);
return { foregroundApp: tree.foregroundApp, alertVisible: tree.alertVisible, keyboardVisible: tree.keyboardVisible, truncated: tree.truncated,
screen: tree.screen, nodes, roots: (tree.roots ?? []).map(root => [root.type ?? null, root.bundleId ?? null, (root.children ?? []).length]) };
}
/**
* What a failed `argent run` printed, for its error: the last few lines, and ahead of them the line saying the
* tool-server is not serving its tools yet, which the bridge waits out (argent-driver's TOOLS_NOT_SERVED) and which a
* stack trace can push out of the last lines.
*/
export function argentRunFailureDetail(stderr: string, stdout: string): string {
const lines = `${stderr}\n${stdout}`.trim().split("\n").filter(Boolean);
const notServed = lines.find(line => /GET \/tools failed: \d{3}|ECONNREFUSED/.test(line));
return [...new Set([...(notServed ? [notServed] : []), ...lines.slice(-4)])].join(" | ").slice(0, 800);
}
/** A command's answer, with the lines it has for the capture's log (`notes`), which its host prints. */
export type ArgentBridgeResult = BridgeResult & { notes?: string[] };
export type ArgentBridge = {
/** Runs one agent-device-shaped command against the session. */
run(argv: readonly string[]): Promise<ArgentBridgeResult>;
};
/**
* One bridge over the given Argent call. `host` names the process running it in the debug log.
*/
export function createArgentBridge(settings: ArgentBridgeSettings, call: ArgentCall, host: "process" | "thread"): ArgentBridge {
const { environment, statePath, platform, toolsUrl } = settings;
const logPath = environment.EXPLORE_ARGENT_LOG_DIR
? join(environment.EXPLORE_ARGENT_LOG_DIR, `argent-${platform}-${environment.EXPLORE_ARGENT_LOG_NAME ?? "bridge"}.jsonl`) : undefined;
return {
async run(argv) {
// EXPLORE_ARGENT_LOG_DIR (opt-in): every Argent call this bridge makes, how long it took, and the shape of what it read.
const log = async (entry: Record<string, unknown>) => {
if (!logPath) return;
await appendFile(logPath, `${JSON.stringify({ at: new Date().toISOString(), host, command: argv[0], ...entry })}\n`, { mode: 0o600 }).catch(() => undefined);
};
const state = await readState(statePath);
const notes: string[] = [];
const result = await runBridgeCommand(argv, {
state, platform, version: ARGENT_VERSION,
progress: message => { notes.push(message); void log({ note: message }); },
...(environment.EXPLORE_ARGENT_BUNDLE_ID ? { bundleId: environment.EXPLORE_ARGENT_BUNDLE_ID } : {}),
...(environment.EXPLORE_ARGENT_DEVICE ? { device: environment.EXPLORE_ARGENT_DEVICE } : {}),
// An opened foldable's inner panel shows its UI in landscape (Argent's fold: "Unfolded, the UI is landscape").
...(platform === "ios" && ["open", "half-open"].includes(environment.EXPLORE_ARGENT_POSE ?? "") ? { landscape: true } : {}),
call: async (tool, args, options) => {
const started = Date.now();
try {
const value = await call(tool, args, options);
await log({ tool, args: loggedArgs(args), ms: Date.now() - started, summary: summarize(tool, value) });
return value;
} catch (error) {
await log({ tool, args: loggedArgs(args), ms: Date.now() - started, error: (error instanceof Error ? error.message : String(error)).slice(0, 400) });
throw error;
}
},
reachable: async () => {
let ok = false;
try {
const token = environment.ARGENT_AUTH_TOKEN;
// The same URL Argent's own client checks: ARGENT_TOOLS_URL may carry a path, which /tools is appended to.
const response = await fetch(`${(toolsUrl ?? "").replace(/\/+$/, "")}/tools`, { headers: token ? { authorization: `Bearer ${token}` } : {}, signal: AbortSignal.timeout(5_000) });
await response.body?.cancel();
// A tool-server still starting can answer 404 to /tools (euxy main baseline run 39506374824): it is not serving yet.
ok = response.status < 500 && response.status !== 404;
} catch { ok = false; }
if (!ok) await log({ tool: "reachable", ok });
return ok;
},
sleep: milliseconds => new Promise(resolve => setTimeout(resolve, milliseconds)),
orientVideo: async path => {
const probed = spawnSync("ffprobe", ["-v", "error", "-select_streams", "v:0", "-show_entries", "stream=width,height", "-of", "csv=p=0:s=x", path], { encoding: "utf8", timeout: 30_000 });
const [width, height] = (probed.stdout ?? "").trim().split("x").map(Number);
if (probed.status !== 0 || !width || !height || width >= height) return;
const turned = `${path}.upright.mp4`;
const ffmpeg = spawnSync("ffmpeg", ["-nostdin", "-y", "-loglevel", "error", "-i", path, "-vf", "transpose=clock", "-an", turned], { encoding: "utf8", timeout: 120_000 });
if (ffmpeg.status === 0) await rename(turned, path);
else await rm(turned, { force: true });
},
scratchImage: { path: join(dirname(statePath), "argent-scratch.png"), read: async path => {
const { PNG } = createRequire(import.meta.url)("pngjs") as { PNG: { sync: { read(bytes: Buffer, options: { checkCRC: boolean }): DecodedFrame } } };
return PNG.sync.read(await readFile(path), { checkCRC: false });
} },
moveFile: async (from, to) => {
const source = await realpath(from);
const roots = await Promise.all([environment.HOME, process.cwd()].filter((root): root is string => !!root).map(root => realpath(root).catch(() => root)));
if (!roots.some(root => source.startsWith(`${root}/`))) throw new Error(`Argent returned a recording outside its own directories: ${from}`);
await mkdir(dirname(to), { recursive: true });
try { await rename(from, to); }
catch { await copyFile(from, to); await rm(from, { force: true }); }
},
});
await writeState(statePath, state);
await log({ result: result.exitCode === 0 ? "ok" : result.stdout.slice(0, 300) });
return notes.length ? { ...result, notes } : result;
},
};
}
+13 -110
View File
@@ -2,71 +2,30 @@
* The Argent controller's command line: Explore runs it where it would run agent-device, with the same
* arguments. Each command becomes `argent run <tool> --args <json>` calls, as EAS Simulator recommends
* driving an `argent` session, with the session's ARGENT_TOOLS_URL and ARGENT_AUTH_TOKEN in the environment.
* A capture keeps the same bridge on a worker thread instead (argent-bridge-host); this entry serves the callers
* that run one command at a time.
*/
import { spawnSync } from "node:child_process";
import { appendFile, copyFile, mkdir, readFile, realpath, rename, rm, stat, writeFile } from "node:fs/promises";
import { dirname, join } from "node:path";
import { createRequire } from "node:module";
import type { DecodedFrame } from "../shared/png-evidence.js";
import { ArgentCallError, runBridgeCommand, type ArgentBridgeState, type ArgentPlatform, type ArgentTree } from "./argent-driver.js";
import { ARGENT_VERSION } from "../shared/argent-version.js";
import { stat } from "node:fs/promises";
import { ArgentCallError } from "./argent-driver.js";
import { argentBridgeSettings, argentRunFailureDetail, createArgentBridge } from "./argent-bridge.js";
import { ensureController } from "../shared/controller-cache.js";
const ARGENT_CALL_TIMEOUT_MS = 180_000;
async function readState(path: string): Promise<ArgentBridgeState> {
try { return JSON.parse(await readFile(path, "utf8")) as ArgentBridgeState; }
catch { return { generation: 0 }; }
}
/** Written whole and renamed into place, so a command stopped mid-write never leaves a torn state. */
async function writeState(path: string, state: ArgentBridgeState): Promise<void> {
const temporary = `${path}.${process.pid}.tmp`;
await writeFile(temporary, JSON.stringify(state), { mode: 0o600 });
await rename(temporary, path);
}
/** A call's arguments for the debug log: text that could be typed into a field is logged by length only. */
function loggedArgs(args: Record<string, unknown>): Record<string, unknown> {
return Object.fromEntries(Object.entries(args).map(([key, value]) => key === "text" && typeof value === "string" ? [key, `<${value.length} characters>`]
: key === "steps" && Array.isArray(value) ? [key, `<${value.length} steps>`] : [key, value]));
}
/** What a call answered, for the debug log: a tree's shape, not its contents. */
function summarize(tool: string, value: unknown): unknown {
if (!value || typeof value !== "object") return value;
if (tool !== "ui-tree") return Object.keys(value);
const tree = value as ArgentTree;
let nodes = 0;
const walk = (node: { children?: unknown[] }) => { nodes++; (node.children ?? []).forEach(child => walk(child as { children?: unknown[] })); };
(tree.roots ?? []).forEach(walk);
return { foregroundApp: tree.foregroundApp, alertVisible: tree.alertVisible, keyboardVisible: tree.keyboardVisible, truncated: tree.truncated,
screen: tree.screen, nodes, roots: (tree.roots ?? []).map(root => [root.type ?? null, root.bundleId ?? null, (root.children ?? []).length]) };
}
async function main(): Promise<number> {
const environment = process.env;
const statePath = environment.EXPLORE_ARGENT_STATE;
const platform = environment.EXPLORE_ARGENT_PLATFORM as ArgentPlatform | undefined;
const toolsUrl = environment.ARGENT_TOOLS_URL;
if (!statePath || (platform !== "ios" && platform !== "android")) {
const settings = argentBridgeSettings(environment);
if (!settings) {
process.stdout.write(JSON.stringify({ success: false, error: { code: "INVALID_ARGS", message: "The Argent controller needs EXPLORE_ARGENT_STATE and EXPLORE_ARGENT_PLATFORM." } }));
return 1;
}
const argv = process.argv.slice(2);
if (argv[0] !== "--version" && !toolsUrl) {
if (argv[0] !== "--version" && !settings.toolsUrl) {
// Without a URL the Argent client would start a local tool-server; Explore only drives the session's own.
process.stdout.write(JSON.stringify({ success: false, error: { code: "INVALID_ARGS", message: "ARGENT_TOOLS_URL is not set for the Argent controller." } }));
return 1;
}
// EXPLORE_ARGENT_LOG_DIR (opt-in): every Argent call this bridge makes, how long it took, and the shape of what it read.
const logPath = environment.EXPLORE_ARGENT_LOG_DIR
? join(environment.EXPLORE_ARGENT_LOG_DIR, `argent-${platform}-${environment.EXPLORE_ARGENT_LOG_NAME ?? "bridge"}.jsonl`) : undefined;
const log = async (entry: Record<string, unknown>) => {
if (!logPath) return;
await appendFile(logPath, `${JSON.stringify({ at: new Date().toISOString(), command: argv[0], ...entry })}\n`, { mode: 0o600 }).catch(() => undefined);
};
const state = await readState(statePath);
let cli: string;
try { cli = argv[0] === "--version" ? "" : await ensureController("argent", { environment, progress: message => process.stderr.write(`[explore] ${message}\n`) }); }
catch (error) {
@@ -75,13 +34,12 @@ async function main(): Promise<number> {
return 1;
}
const callArgent = async (tool: string, args: Record<string, unknown>, options?: { out?: string }): Promise<unknown> => {
const callArgent = async (tool: string, args: Record<string, unknown>, options?: { out?: string; timeoutMs?: number }): Promise<unknown> => {
const run = spawnSync(process.execPath, [cli, "run", tool, "--args", JSON.stringify(args), ...(options?.out ? ["--out", options.out] : [])],
{ env: environment, encoding: "utf8", maxBuffer: 64 * 1024 * 1024, timeout: ARGENT_CALL_TIMEOUT_MS });
{ env: environment, encoding: "utf8", maxBuffer: 64 * 1024 * 1024, timeout: Math.min(ARGENT_CALL_TIMEOUT_MS, options?.timeoutMs ?? ARGENT_CALL_TIMEOUT_MS) });
if (run.error) throw new ArgentCallError(`argent run ${tool}: ${run.error.message}`, true);
if (run.status !== 0) {
const detail = `${run.stderr ?? ""}\n${run.stdout ?? ""}`.trim().split("\n").filter(Boolean).slice(-4).join(" | ").slice(0, 800);
throw new ArgentCallError(`argent run ${tool} failed: ${detail}`, true);
throw new ArgentCallError(`argent run ${tool} failed: ${argentRunFailureDetail(run.stderr ?? "", run.stdout ?? "")}`, true);
}
// With --out the client saves the tool's file and prints where ("Wrote: <path>"), not the tool's JSON.
if (options?.out) {
@@ -93,65 +51,10 @@ async function main(): Promise<number> {
catch { throw new ArgentCallError(`argent run ${tool} did not answer with JSON`, true); }
};
const result = await runBridgeCommand(argv, {
state, platform, version: ARGENT_VERSION,
...(environment.EXPLORE_ARGENT_BUNDLE_ID ? { bundleId: environment.EXPLORE_ARGENT_BUNDLE_ID } : {}),
...(environment.EXPLORE_ARGENT_DEVICE ? { device: environment.EXPLORE_ARGENT_DEVICE } : {}),
// An opened foldable's inner panel shows its UI in landscape (Argent's fold: "Unfolded, the UI is landscape").
...(platform === "ios" && ["open", "half-open"].includes(environment.EXPLORE_ARGENT_POSE ?? "") ? { landscape: true } : {}),
call: async (tool, args, options) => {
const started = Date.now();
try {
const value = await callArgent(tool, args, options);
await log({ tool, args: loggedArgs(args), ms: Date.now() - started, summary: summarize(tool, value) });
return value;
} catch (error) {
await log({ tool, args: loggedArgs(args), ms: Date.now() - started, error: (error instanceof Error ? error.message : String(error)).slice(0, 400) });
throw error;
}
},
reachable: async () => {
let ok = false;
try {
const token = environment.ARGENT_AUTH_TOKEN;
// The same URL Argent's own client checks: ARGENT_TOOLS_URL may carry a path, which /tools is appended to.
const response = await fetch(`${(toolsUrl ?? "").replace(/\/+$/, "")}/tools`, { headers: token ? { authorization: `Bearer ${token}` } : {}, signal: AbortSignal.timeout(5_000) });
await response.body?.cancel();
ok = response.status < 500;
} catch { ok = false; }
if (!ok) await log({ tool: "reachable", ok });
return ok;
},
sleep: milliseconds => new Promise(resolve => setTimeout(resolve, milliseconds)),
orientVideo: async path => {
// Turned with ffmpeg, which observing a launch needs anyway; without it the recording is left as it is.
const probed = spawnSync("ffprobe", ["-v", "error", "-select_streams", "v:0", "-show_entries", "stream=width,height", "-of", "csv=p=0:s=x", path], { encoding: "utf8", timeout: 30_000 });
const [width, height] = (probed.stdout ?? "").trim().split("x").map(Number);
if (probed.status !== 0 || !width || !height || width >= height) return;
const turned = `${path}.upright.mp4`;
const ffmpeg = spawnSync("ffmpeg", ["-nostdin", "-y", "-loglevel", "error", "-i", path, "-vf", "transpose=clock", "-an", turned], { encoding: "utf8", timeout: 120_000 });
if (ffmpeg.status === 0) await rename(turned, path);
else await rm(turned, { force: true });
},
scratchImage: { path: join(dirname(statePath), "argent-scratch.png"), read: async path => {
const { PNG } = createRequire(import.meta.url)("pngjs") as { PNG: { sync: { read(bytes: Buffer, options: { checkCRC: boolean }): DecodedFrame } } };
return PNG.sync.read(await readFile(path), { checkCRC: false });
} },
moveFile: async (from, to) => {
// Argent's client downloads a recording under its HOME or the working directory; nothing else is moved.
const source = await realpath(from);
const roots = await Promise.all([environment.HOME, process.cwd()].filter((root): root is string => !!root).map(root => realpath(root).catch(() => root)));
if (!roots.some(root => source.startsWith(`${root}/`))) throw new Error(`Argent returned a recording outside its own directories: ${from}`);
await mkdir(dirname(to), { recursive: true });
try { await rename(from, to); }
catch { await copyFile(from, to); await rm(from, { force: true }); }
},
});
await writeState(statePath, state);
await log({ result: result.exitCode === 0 ? "ok" : result.stdout.slice(0, 300) });
const result = await createArgentBridge(settings, callArgent, "process").run(argv);
for (const note of result.notes ?? []) process.stderr.write(`[explore] ${note}\n`);
process.stdout.write(result.stdout);
return result.exitCode;
}
process.exitCode = await main();
+144 -20
View File
@@ -5,6 +5,7 @@
* injected, so the translation is tested against recorded Argent output.
*/
import { CONTROLLER_SESSION } from "./common.js";
import { INPUT_METHOD_PACKAGES, KEYBOARD_WINDOW_IDENTIFIERS } from "./keyboard.js";
import type { DecodedFrame } from "../shared/png-evidence.js";
export type ArgentPlatform = "ios" | "android";
@@ -14,7 +15,8 @@ export class ArgentCallError extends Error {
constructor(message: string, readonly sent: boolean) { super(message); }
}
export type ArgentCall = (tool: string, args: Record<string, unknown>, options?: { out?: string }) => Promise<unknown>;
/** `timeoutMs` shortens the call's own timeout, so a wait for a tool-server still starting stays within its window. */
export type ArgentCall = (tool: string, args: Record<string, unknown>, options?: { out?: string; timeoutMs?: number }) => Promise<unknown>;
type Rect = { x: number; y: number; width: number; height: number };
@@ -30,6 +32,8 @@ export type ArgentBridgeState = {
/** iOS asks "Open in “<app>”?" for the first custom-scheme link of a session; later links open directly. */
deepLinkPromptAccepted?: boolean;
recording?: string;
/** The tool-server has answered a call this session, so a GET /tools it fails is no longer taken for it still starting. */
toolsServed?: boolean;
};
export type ArgentBridgeContext = {
@@ -42,6 +46,8 @@ export type ArgentBridgeContext = {
device?: string;
version: string;
sleep: (milliseconds: number) => Promise<void>;
/** A monotonic clock in milliseconds (performance.now when absent), which bounds the wait for a tool-server to start. */
now?: () => number;
/** Moves a file the Argent client downloaded (a screen recording) to where Explore asked for it. */
moveFile: (from: string, to: string) => Promise<void>;
/** Whether the tool-server answers, checked before a UI action is sent (see `remoteUnavailable`). */
@@ -56,6 +62,8 @@ export type ArgentBridgeContext = {
landscape?: boolean;
/** Turns a portrait screen recording a quarter clockwise, to the landscape UI it shows. */
orientVideo?: (path: string) => Promise<void>;
/** A line for the capture's log: the bridge waiting on a tool-server that is still starting. */
progress?: (message: string) => void;
};
export type BridgeResult = { exitCode: number; stdout: string };
@@ -88,7 +96,7 @@ export function parseCommand(argv: readonly string[]): { positional: string[]; f
}
type ArgentNode = { role?: string; type?: string; label?: string; identifier?: string; value?: string; contentDescription?: string; disabled?: boolean;
frame?: Rect; traits?: string[]; hidden?: boolean; selected?: boolean; password?: boolean; bundleId?: string; children?: ArgentNode[] };
frame?: Rect; traits?: string[]; hidden?: boolean; selected?: boolean; focused?: boolean; password?: boolean; bundleId?: string; children?: ArgentNode[] };
export type ArgentTree = { schemaVersion?: number; source?: string; screen?: { width: number; height: number }; roots?: ArgentNode[];
truncated?: boolean; alertVisible?: boolean; keyboardVisible?: boolean; foregroundApp?: string };
@@ -137,7 +145,11 @@ export function snapshotFromTree(tree: ArgentTree, platform: ArgentPlatform, gen
if (root || !node.frame || ![node.frame.x, node.frame.y, node.frame.width, node.frame.height].every(finite)) return clip;
return clip && intersect(node.frame, clip);
};
const visit = (node: ArgentNode, depth: number, parentIndex: number | undefined, appRoot: boolean, clip: Rect | undefined, root: boolean) => {
const keyboard = platform === "ios" ? keyboardWindows(roots) : new Set<ArgentNode>();
/** `owner` marks every node of a window that is the system keyboard's, as agent-device marks an input method's nodes. */
type Owner = { bundleId?: string; systemKeyboard?: true };
const visit = (node: ArgentNode, depth: number, parentIndex: number | undefined, appRoot: boolean, clip: Rect | undefined, root: boolean, owner: Owner) => {
if (keyboard.has(node)) owner = { ...owner, systemKeyboard: true };
const frame = node.frame;
const inner = clipOf(node, clip, root);
const rect = frame && [frame.x, frame.y, frame.width, frame.height].every(finite) ? toPoints(frame) : undefined;
@@ -159,19 +171,42 @@ export function snapshotFromTree(tree: ArgentTree, platform: ArgentPlatform, gen
rect: appRoot ? { x: 0, y: 0, width: screen.width, height: screen.height } : rect,
...(label ? { label } : {}), ...(node.identifier ? { identifier: node.identifier } : {}),
...(node.value !== undefined ? { value: node.value } : {}), enabled, hittable: enabled && visible && !appRoot,
...(visible ? {} : { visible: false }), ...(node.selected || traits.includes("selected") ? { selected: true } : {}) });
...(visible ? {} : { visible: false }), ...(node.selected || traits.includes("selected") ? { selected: true } : {}),
// The field with the keyboard focus, as agent-device marks it, so capture can wait for the keyboard it raises.
...(node.focused === true || traits.includes("isEditing") ? { focused: true } : {}),
...owner });
refs[ref] = { rect: appRoot ? { x: 0, y: 0, width: screen.width, height: screen.height } : rect, type,
...(node.value !== undefined ? { value: node.value } : {}), ...(type === "SecureTextField" ? { secure: true } : {}) };
}
for (const child of node.children ?? []) visit(child, depth + 1, index, false, inner, false);
for (const child of node.children ?? []) visit(child, depth + 1, index, false, inner, false, owner);
};
for (const root of roots) {
const isAppRoot = platform === "ios" ? root.type === "Application" || nodes.length === 0 : root.bundleId === app;
visit(root, 0, undefined, isAppRoot, { x: 0, y: 0, width: 1, height: 1 }, true);
// An Android input method's window names its package on each node, as agent-device's do, which marks the keyboard.
const inputMethod = platform === "android" && root.bundleId && INPUT_METHOD_PACKAGES.has(root.bundleId) ? root.bundleId : undefined;
visit(root, 0, undefined, isAppRoot, { x: 0, y: 0, width: 1, height: 1 }, true, inputMethod ? { bundleId: inputMethod } : {});
}
return { nodes, refs, screen: { width: screen.width, height: screen.height } };
}
/**
* The windows iOS draws the software keyboard in. Argent lists each window as a child of its root, the app's first. A
* later window is the keyboard's when it holds XCTest's `Keyboard` element, which no app can give itself; an identifier
* never decides it alone, since an app can set any identifier on its own views (a second window of the app's with a
* `testID="inputView"` stays the app's). While such a window is up, UIKit also hosts the keyboard's input view
* placeholder in a window of its own: a later window whose only named node is an `inputView` belongs to it too.
*/
function keyboardWindows(roots: readonly ArgentNode[]): Set<ArgentNode> {
const holds = (node: ArgentNode, test: (node: ArgentNode) => boolean): boolean => test(node) || (node.children ?? []).some(child => holds(child, test));
const later = roots.flatMap(root => (root.children ?? []).slice(1));
const keyboard = later.filter(window => holds(window, node => node.type === "Keyboard"));
if (!keyboard.length) return new Set();
const named = (node: ArgentNode) => Boolean(node.label?.trim() || node.contentDescription?.trim() || node.value !== undefined
|| (node.identifier && !KEYBOARD_WINDOW_IDENTIFIERS.has(node.identifier)));
const placeholders = later.filter(window => !keyboard.includes(window) && holds(window, node => node.identifier === "inputView") && !holds(window, named));
return new Set([...keyboard, ...placeholders]);
}
function intersect(a: Rect, b: Rect): Rect | undefined {
const x = Math.max(a.x, b.x), y = Math.max(a.y, b.y);
const right = Math.min(a.x + a.width, b.x + b.width), bottom = Math.min(a.y + a.height, b.y + b.height);
@@ -216,6 +251,35 @@ const KEYBOARD_DISMISS_LABELS = ["Hide keyboard", "Dismiss keyboard", "Done"];
const KEYBOARD_TOP = 0.4;
const ALERT_PREFERRED_BUTTONS = ["Open", "OK", "Allow", "Allow While Using App", "Continue"];
const READ_TOOLS = new Set(["ui-tree", "list-devices", "screenshot", "foreground-app", "await-screen-idle", "describe"]);
/**
* Argent 0.29's answer when a service a tool depends on (the device's simulator-server, its accessibility service) did
* not start: the registry resolves a tool's services before it calls the tool, and wraps a start failure as
* "[Tool:<id>] Service dependency failed: …", so the tool itself never ran and resending it is safe, a UI action too.
* The tools the bridge sends that resolve a service while they run (keyboard, fill, restart-app, screen-recording-start)
* resolve it before they act, and run-sequence answers a step's failure in its result rather than raising it. A failed
* start leaves the service in the registry's ERROR state, so the next call starts it afresh.
*/
const SERVICE_DEPENDENCY_FAILED = "Service dependency failed:";
/**
* In brentvatne/euxy#172 (review run 39348956443) capture restarted SpringBoard and folded the iPhone Duo open again 8 s
* later; that fold's simulator-server did not become ready within Argent's 30 s, and the whole target failed. The resend
* comes after a pause, so a device still coming back from the restart has the time it needs.
*/
const DEPENDENCY_RETRY_DELAYS_MS = [3_000, 6_000] as const;
/**
* A tool-server that is not serving its tools yet. Argent's client asks GET /tools before every tool call and stops with
* "GET /tools failed: <status>" when that is not answered, so the tool itself was never sent; a refused connection
* reached nothing at all. Either is safe to send again, a UI action too. In brentvatne/euxy's main baseline run
* 39506374824 a replacement EAS session's tool-server answered 404 to /tools for the first list-devices of its install,
* and the whole target failed. An unknown tool's 404 comes from POST /tools/<name> instead, and still fails at once.
*/
const TOOLS_NOT_SERVED = /GET \/tools failed: (?:404|503)\b|ECONNREFUSED/;
/**
* How long a tool-server that has not answered yet gets to start serving its tools, requests included: a deadline, not
* a number of tries, since each try can itself take until the call's timeout. The pauses between tries grow to 10 s.
*/
const STARTUP_WINDOW_MS = 75_000;
const STARTUP_RETRY_DELAYS_MS = [1_000, 2_000, 4_000, 8_000, 10_000] as const;
const SPRINGBOARD = "com.apple.springboard";
const ANDROID_SYSTEM_UI = "com.android.systemui";
const OPEN_IN_APP_PROMPT = /^Open in “.+”\?$/;
@@ -299,13 +363,49 @@ export async function runBridgeCommand(argv: readonly string[], context: ArgentB
const { positional, flags } = parseCommand(rest);
const bundleFlag = typeof flags["bundle-id"] === "string" ? flags["bundle-id"] : context.bundleId;
/** Reads are resent after a transport failure (they change nothing); a UI action is sent once, after a health check. */
/**
* Reads are resent after a transport failure (they change nothing); a UI action is sent once, after a health check.
* Any tool, a UI action too, is resent when Argent could not start a service it depends on, which it raises before
* the tool runs (see SERVICE_DEPENDENCY_FAILED), and while a tool-server that has not answered yet is still starting
* (see TOOLS_NOT_SERVED).
*/
const call = async (tool: string, args: Record<string, unknown> = {}, options?: { out?: string }): Promise<Record<string, unknown>> => {
const withDevice = tool === "list-devices" ? args : { udid: state.udid, ...args };
for (let attempt = 0; ; attempt++) {
try { return record(await context.call(tool, withDevice, options)); }
catch (error) {
if (READ_TOOLS.has(tool) && attempt < 2) { await context.sleep(1_000 * (attempt + 1)); continue; }
// Tools that name devices their own way (or none) take no udid.
const withDevice = tool === "list-devices" || tool === "stop-all-simulator-servers" ? args : { udid: state.udid, ...args };
const clock = context.now ?? (() => performance.now());
// The window opens with the first call to a tool-server that has not answered yet, and closes 75 s later. A device
// already named came from the tool-server's own answer.
const served = () => state.toolsServed === true || state.udid !== undefined;
const startupDeadline = served() ? undefined : clock() + STARTUP_WINDOW_MS;
for (let readRetry = 0, dependencyRetry = 0, startupRetry = 0; ;) {
const starting = !served() && startupDeadline !== undefined;
try {
// A call to a tool-server still starting gets only the time left in the window.
const answer = record(await context.call(tool, withDevice, starting ? { ...options, timeoutMs: Math.max(1, Math.ceil(startupDeadline - clock())) } : options));
state.toolsServed = true;
return answer;
} catch (error) {
// Until the session's tool-server has answered once, a GET /tools it does not serve is it still starting.
if (starting && error instanceof ArgentCallError) {
const notServed = TOOLS_NOT_SERVED.test(error.message);
const left = startupDeadline - clock();
// A try the window's end cut short is the tool-server not starting in time, whatever the request reported.
if (left <= 0) {
throw new BridgeFailure("TOOL_SERVER_NOT_READY", `Argent's tool-server did not start serving its tools within ${STARTUP_WINDOW_MS / 1_000} s: ${error.message}`);
}
if (notServed) {
const delay = Math.min(STARTUP_RETRY_DELAYS_MS[Math.min(startupRetry++, STARTUP_RETRY_DELAYS_MS.length - 1)]!, left);
context.progress?.(`Argent's tool-server is not serving its tools yet (${error.message}); sending ${tool} again in ${Math.ceil(delay / 1_000)} s (${Math.ceil(left / 1_000)} s left of ${STARTUP_WINDOW_MS / 1_000} s)`);
await context.sleep(delay);
continue;
}
}
if (error instanceof ArgentCallError && error.message.includes(`[Tool:${tool}] ${SERVICE_DEPENDENCY_FAILED}`)
&& dependencyRetry < DEPENDENCY_RETRY_DELAYS_MS.length) {
await context.sleep(DEPENDENCY_RETRY_DELAYS_MS[dependencyRetry++]!);
continue;
}
if (READ_TOOLS.has(tool) && readRetry < 2) { await context.sleep(1_000 * ++readRetry); continue; }
throw error;
}
}
@@ -364,8 +464,9 @@ export async function runBridgeCommand(argv: readonly string[], context: ArgentB
return target;
};
/**
* A UI action is only sent when the tool-server answers first, so a failure after this point is never resent. A
* session's tool-server can still be starting when EAS reports the session ready, so it is given about 25 s to answer.
* A UI action is only sent when the tool-server answers first, so a failure after this point is never resent unless
* the tool provably never ran (see call). A session's tool-server can still be starting when EAS reports the session
* ready, so it is given about 25 s to answer.
*/
const beforeAction = async () => {
for (const delay of [1_000, 2_000, 3_000, 4_000, 5_000, 5_000, 5_000]) {
@@ -413,7 +514,9 @@ export async function runBridgeCommand(argv: readonly string[], context: ArgentB
converted.nodes.push({ depth: 1, index: converted.nodes.length, parentIndex: 0, type: "Alert", ref: `e${converted.nodes.length + 1}`,
rect: { x: 0, y: 0, width: converted.screen.width, height: converted.screen.height }, label: "System alert", enabled: true, hittable: false });
}
// Whether a keyboard shows, which Argent's Android tree does not hold: the input method's window is not in it.
return success({ nodes: converted.nodes, truncated: current.truncated === true, refsGeneration: state.generation,
...(typeof current.keyboardVisible === "boolean" ? { keyboardVisible: current.keyboardVisible } : {}),
...(current.foregroundApp ? { appBundleId: current.foregroundApp, appName: current.foregroundApp } : {}),
snapshotQuality: { state: "healthy", backend: "argent" } });
}
@@ -619,6 +722,10 @@ export async function runBridgeCommand(argv: readonly string[], context: ArgentB
const posture = positional[0];
if (posture !== "closed" && posture !== "half-open" && posture !== "open") throw new BridgeFailure("INVALID_ARGS", "fold needs closed, half-open or open.");
await beforeAction();
// Helpers the tool-server started before a fold can leave the panel it lights black and unread: on a fresh EAS
// session, whose tool-server started them at boot, the first fold opened to a black inner panel every time
// without this, and read the panel at once with it. A local tool-server is Explore's own and starts clean.
await call("stop-all-simulator-servers", { devices: [state.udid] });
const folded = await call("fold", { posture });
// The other panel lights at another size: every stored coordinate and element reference is stale.
delete state.screen;
@@ -630,16 +737,33 @@ export async function runBridgeCommand(argv: readonly string[], context: ArgentB
const lit = record(folded.screen);
if (finite(lit.width) && finite(lit.height) && lit.height > 0) {
const litRatio = lit.width / lit.height;
let followed = false;
for (let poll = 0; poll < 10 && !followed; poll++) {
const shape = (await tree()).screen;
followed = !!shape && shape.height > 0 && Math.abs(shape.width / shape.height - litRatio) < 0.004;
if (!followed) await context.sleep(1_000);
const follows = async () => {
for (let poll = 0; poll < 10; poll++) {
const shape = (await tree()).screen;
if (shape && shape.height > 0 && Math.abs(shape.width / shape.height - litRatio) < 0.004) return true;
await context.sleep(1_000);
}
return false;
};
// A reader still stuck on the dark panel is reset once more (the device's helpers, as reset-reader does): it is
// cheap and leaves the app alone. If that is not enough, capture restarts SpringBoard and folds again.
if (!await follows() && !await call("stop-all-simulator-servers", { devices: [state.udid] }).then(follows, () => false)) {
throw new BridgeFailure("CONTROLLER_FOLD_UNVERIFIED", `Argent kept reading the other panel after folding the device ${posture}, and after its device helpers were reset.`);
}
if (!followed) throw new BridgeFailure("CONTROLLER_FOLD_UNVERIFIED", `Argent kept reading the other panel after folding the device ${posture}.`);
}
return success({ action: "fold", pose: posture, ...(folded.panel !== undefined ? { panel: folded.panel } : {}) });
}
case "reset-reader": {
// Stops the device's Argent helpers, among them the accessibility service describe reads through; the
// tool-server starts fresh ones on the next call. It is how a reader stuck on the panel a fold turned off is
// reset where the tool-server itself cannot be restarted, as on EAS (it started it; never stop it there).
await ensureDevice();
const stopped = await call("stop-all-simulator-servers", { devices: [state.udid] });
delete state.screen;
delete state.refs;
state.generation += 1;
return success({ action: "reset-reader", stopped: Array.isArray(stopped.stopped) ? stopped.stopped.filter(item => typeof item === "string") : [] });
}
case "apps": return success({ apps: [] });
case "session": return success({ sessions: [] });
// TODO(argent#1509): map `settings appearance <scheme>` to Argent's `set-appearance` once a release with it is pinned.
+1 -1
View File
@@ -1,6 +1,6 @@
import { CiError } from "../shared/ci-errors.js";
export const EAS_CLI_VERSION = "25.0.0";
export const EAS_CLI_VERSION = "25.2.0";
export { AGENT_DEVICE_VERSION } from "../shared/agent-device-version.js";
export const CONTROLLER_SESSION = "expo-explore-runtime";
+8 -4
View File
@@ -5,6 +5,7 @@ import type {
ElementRole,
InputElement,
} from "../types.js";
import { INPUT_METHOD_PACKAGES, keyboardNodes } from "./keyboard.js";
export type ControllerParseResult = {
controller: Controller;
@@ -221,10 +222,7 @@ function nodeToElement(node: Record<string, unknown>): InputElement | undefined
const SYSTEM_UI_PACKAGES = new Set([
"android",
"com.android.systemui",
"com.google.android.inputmethod.latin",
"com.samsung.android.honeyboard",
"com.touchtype.swiftkey",
"com.microsoft.swiftkey",
...INPUT_METHOD_PACKAGES,
]);
export function isSystemUiNode(node: Record<string, unknown>): boolean {
@@ -240,6 +238,12 @@ function collectJsonElements(value: unknown, output: InputElement[], seen: Set<u
return;
}
const record = value as Record<string, unknown>;
// iOS's software keyboard is system UI too, like Android's: its keys and its Emoji and Dictate buttons are not the
// app's elements. A keyboard that rose on one side only made "Added button “Dictate”" lead a review whose real
// change was a field's label (brentvatne/euxy#172, review run 39362642343).
if (Array.isArray(record.nodes)) {
for (const node of keyboardNodes(record.nodes.filter((node): node is Record<string, unknown> => !!node && typeof node === "object" && !Array.isArray(node)))) seen.add(node);
}
const element = isSystemUiNode(record) ? undefined : nodeToElement(record);
if (element) output.push(element);
for (const [key, child] of Object.entries(record)) {
+10 -4
View File
@@ -13,21 +13,27 @@ export function defaultRunner(request: CommandRequest): CommandResult {
maxBuffer: 50 * 1024 * 1024,
...(request.timeoutMs ? { timeout: request.timeoutMs } : {}),
});
if (result.error) throw result.error;
// A command that timed out or could not run keeps what it printed on its error: a hosted repack's start names its run there.
if (result.error) throw Object.assign(result.error, { stdout: result.stdout ?? "", stderr: result.stderr ?? "" });
return commandOutcome(request, { status: result.status, stdout: result.stdout ?? "", stderr: result.stderr ?? "" });
}
/** A finished command's output, or the error its exit status and output describe. */
export function commandOutcome(request: CommandRequest, result: { status: number | null; stdout: string; stderr: string }): CommandResult {
if (result.status !== 0 && !request.allowFailure) {
const sessionIndex = request.args.indexOf("--session");
// Our own session or a caller's session attached by explore:session: both answer --json failures with a structured error.
const structuredFailure = (sessionIndex >= 0 && request.args[sessionIndex + 1] === CONTROLLER_SESSION) || request.args.includes("--json")
? controllerCommandFailure(result.stdout ?? "", request.args[0] ?? "command") : undefined;
? controllerCommandFailure(result.stdout, request.args[0] ?? "command") : undefined;
if (structuredFailure) throw structuredFailure;
const subcommand = typeof request.args[0] === "string" ? ` ${request.args[0]}` : "";
const detail = failureDetail(`${result.stderr ?? ""}\n${result.stdout ?? ""}`);
const detail = failureDetail(`${result.stderr}\n${result.stdout}`);
throw new Error(
`${request.command}${subcommand} exited with status ${result.status ?? "unknown"}`
+ (detail ? `: ${detail}` : ""),
);
}
return { stdout: result.stdout ?? "", stderr: result.stderr ?? "" };
return { stdout: result.stdout, stderr: result.stderr };
}
/** Keep the CLI tool host responsive while a remote controller request is in flight. */
+474 -24
View File
@@ -1,8 +1,11 @@
import { cp, lstat, realpath, mkdtemp, mkdir, writeFile, readFile, rename, rm } from "node:fs/promises";
import { createHash } from "node:crypto";
import { cp, lstat, realpath, mkdtemp, mkdir, writeFile, readFile, readdir, readlink, rename, rm } from "node:fs/promises";
import { tmpdir } from "node:os";
import { isAbsolute, basename, dirname, resolve, join } from "node:path";
import { isAbsolute, basename, dirname, resolve, join, relative, sep } from "node:path";
import { setTimeout as delay } from "node:timers/promises";
import type { SourceRoute } from "../contracts/source-route.js";
import { CiError } from "../shared/ci-errors.js";
import { EXPLORE_VERSION } from "../shared/package-paths.js";
import { prepareCaptureObserver, captureObserverParameterKeys } from "./observer.js";
import { isRecord, parseJson, nonEmpty } from "./common.js";
import { isWithin } from "./fs-safety.js";
@@ -110,12 +113,22 @@ export function nativeBuildFrom(stdout: string): EasBuildDescriptor & { fingerpr
};
}
/** The message Explore's hosted repacks give their builds (hostedRepackWorkflow), with or without a share marker. */
const EXPLORE_REPACK_MESSAGE = /^expo-explore (?:ios|android) capture: route observer repack\b/;
/**
* The finished build to repack from `eas build:list --fingerprint-hash`: the newest one Explore did not repack itself,
* else the newest. A repack keeps its source build's native code, and may carry its fingerprint; preferring the build
* it came from has every job of a run pick the same one, which a shared repack's key needs (RepackShare), instead of a
* repack another job finished a minute earlier.
*/
export function existingBuild(stdout: string): EasBuildDescriptor | undefined {
const value = parseJson(stdout, "eas build:list");
if (!Array.isArray(value)) throw new Error("eas build:list did not return a build list");
if (value.length === 0) return undefined;
const id = buildIdFrom(value, "eas build:list");
const candidate = value[0];
const index = Math.max(0, value.findIndex(build => !(isRecord(build) && typeof build.message === "string" && EXPLORE_REPACK_MESSAGE.test(build.message))));
const candidate = value[index];
const id = buildIdFrom([candidate], "eas build:list");
return isRecord(candidate) ? easBuildDescriptor(candidate, id, "eas build:list") : { id };
}
@@ -395,48 +408,438 @@ export function hostedRepackWorkflow(input: {
].join("\n");
}
/** The repacked build id from `eas workflow:run --json`, or the job's failure as a readable error. */
export function hostedRepackBuildId(stdout: string, stderr: string, platform: "ios" | "android"): string {
const logs = /See logs: (https:\/\/expo\.dev\/\S+)/.exec(stderr)?.[1];
const where = logs ? ` Logs: ${logs}` : "";
/**
* Why a hosted repack failed, as far as its run tells:
* - `app`: the app's own code broke it: the "Repack" step failed, or the logs show a bundling error. A rerun repeats it.
* - `configuration`: EAS or a package registry refused a credential (not logged in, unauthorized, a 401 or 403).
* - `canceled`: someone cancelled the run.
* - `timeout`: the build ran out its EAS build time, which a rerun would wait out again, or Explore's time for the repack ran out.
* - `unconfirmed`: Explore could not learn whether the run ended; it asked EAS to cancel it and starts no other beside it.
* - `infrastructure`: EAS failed around the app's code: a step that sets up the job, downloads or uploads failed, or EAS
* ended the build (brentvatne/euxy#176).
* - `unknown`: nothing says why.
*/
export type HostedRepackFailure = "app" | "configuration" | "canceled" | "timeout" | "unconfirmed" | "infrastructure" | "unknown";
const APP_REPACK_STEP = /^Repack(?: app)?$/i;
// Metro's, Babel's and Expo's words for the app's own code or files: `Unable to resolve "react-native-reanimated" from
// "index.js"`, a missing module, a file an asset step could not read.
const APP_REPACK_FAILURE = /Unable to resolve\b|SyntaxError|TransformError|Error while bundling|Bundling failed|Cannot find module|could not read \.{0,2}\/|no such file or directory, open '\.{0,2}\//i;
// A file of EAS's own worker missing (`Cannot find module '/opt/eas/...'`) is EAS's failure, whatever the step.
const WORKER_REPACK_FAILURE = /['"]\/opt\/eas\//;
const CONFIGURATION_REPACK_FAILURE = /Not logged in|\bunauthori[sz]ed\b|\bnot authorized\b|\bE40[13]\b|\b40[13] (?:Unauthorized|Forbidden)\b|authentication (?:failed|required)/i;
const TIMEOUT_REPACK_FAILURE = /\bEAS_BUILD_TIMEOUT_ERROR\b/;
// The steps EAS runs around the app's own: setting up the job, fetching the build to repack, saving the result.
const INFRASTRUCTURE_REPACK_STEP = /^(?:Spin up|Set up|Download|Upload)\b/i;
// EAS's and eas-cli's own words for a failure on their side. In brentvatne/euxy#176 (review run 39506405687) EAS refused
// the "Upload artifact" step with "Turtle build is not in progress", and the build reported "Failed to upload application archive.".
const INFRASTRUCTURE_REPACK_FAILURE = /Turtle build is not in progress|Failed to upload application archive|\bSERVER_ERROR\b|Failed to upload project sources|Failed to start the workflow with the API|Unexpected response from server \(5\d\d\)|\b(?:ECONNRESET|ECONNREFUSED|ETIMEDOUT|ENOTFOUND|EAI_AGAIN)\b|socket hang up|fetch failed/;
/** How a failure reads: from the build's error, the job's errors, the failed job's logs and the CLI's output. */
export function hostedRepackFailure(evidence: string, status?: unknown): HostedRepackFailure {
if (status === "CANCELED") return "canceled";
if (CONFIGURATION_REPACK_FAILURE.test(evidence)) return "configuration";
if (WORKER_REPACK_FAILURE.test(evidence)) return "infrastructure";
const step = [...evidence.matchAll(/Build step "([^"]+)" failed/g)].at(-1)?.[1];
if ((step && APP_REPACK_STEP.test(step)) || APP_REPACK_FAILURE.test(evidence)) return "app";
if (TIMEOUT_REPACK_FAILURE.test(evidence)) return "timeout";
if ((step && INFRASTRUCTURE_REPACK_STEP.test(step)) || INFRASTRUCTURE_REPACK_FAILURE.test(evidence)) return "infrastructure";
return "unknown";
}
/** How many fresh repacks a failure earns: two after EAS's own failure, one after one nothing explains, none otherwise. */
export function hostedRepackRetries(error: unknown): number {
const failure = error instanceof CiError && error.code === "EAS_REPACK_FAILED" ? error.details?.failure : undefined;
return failure === "infrastructure" ? 2 : failure === "unknown" ? 1 : 0;
}
const EXPO_URL = /^https:\/\/expo\.dev\/\S+$/;
const EAS_ID = /^[A-Za-z0-9-]{1,128}$/;
/**
* The repacked build id from a finished run's JSON (`eas workflow:view --json`), or the job's failure as a readable
* error. `evidence` is the rest of what the run said: the CLI's stderr and the failed job's logs.
*/
export function hostedRepackBuildId(stdout: string, evidence: string, platform: "ios" | "android"): string {
let logs = /See logs: (https:\/\/expo\.dev\/\S+)/.exec(evidence)?.[1];
let value: unknown;
try {
value = parseJson(stdout, "eas workflow:run");
} catch (error) {
const tail = evidence.trim().split("\n").slice(-3).join(" | ").slice(0, 600);
throw new CiError({
code: "EAS_REPACK_FAILED",
stage: "instrumentation",
message: `eas workflow:run did not report the hosted ${platform} repack: ${stderr.trim().split("\n").slice(-3).join(" | ").slice(0, 600) || (error instanceof Error ? error.message : String(error))}`,
nextAction: `Check that EXPO_TOKEN can run workflows for this EAS project and retry.${where}`,
message: `eas workflow:run did not report the hosted ${platform} repack: ${tail || (error instanceof Error ? error.message : String(error))}`,
nextAction: `Check that EXPO_TOKEN can run workflows for this EAS project and retry.${logs ? ` Logs: ${logs}` : ""}`,
details: { failure: hostedRepackFailure(evidence), ...(logs ? { logs } : {}) },
cause: error,
});
}
for (const field of ["url", "logURL"]) {
const link = isRecord(value) ? value[field] : undefined;
if (typeof link === "string" && EXPO_URL.test(link)) logs = link;
}
const where = logs ? ` Logs: ${logs}` : "";
const jobs = isRecord(value) && Array.isArray(value.jobs) ? value.jobs.filter(isRecord) : [];
const job = jobs.find(entry => entry.key === "repack") ?? jobs[0];
if (!job || job.status !== "SUCCESS") {
const build = job && isRecord(job.turtleBuild) ? job.turtleBuild : undefined;
const detail = build && isRecord(build.error) && typeof build.error.message === "string" ? build.error.message.trim() : undefined;
const planLimit = easPlanLimitError(new Error(`${detail ?? ""}\n${stderr}`), platform);
const buildError = build && isRecord(build.error) ? build.error : undefined;
const detail = typeof buildError?.message === "string" ? buildError.message.trim() : undefined;
const planLimit = easPlanLimitError(new Error(`${detail ?? ""}\n${evidence}`), platform);
if (planLimit) throw planLimit;
const jobErrors = Array.isArray(job?.errors) ? job.errors.filter(isRecord).map(entry => `${String(entry.title ?? "")} ${String(entry.message ?? "")}`) : [];
const failure = hostedRepackFailure([detail ?? "", typeof buildError?.errorCode === "string" ? buildError.errorCode : "", ...jobErrors, evidence].join("\n"),
job?.status ?? (isRecord(value) ? value.status : undefined));
throw new CiError({
code: "EAS_REPACK_FAILED",
stage: "instrumentation",
message: `The hosted EAS repack of the ${platform} build ${job ? `finished with status ${String(job.status)}` : "did not run"}${detail ? `: ${detail.slice(0, 600)}` : "."}`,
nextAction: `Open the EAS workflow run and fix the repack job; a broken app build environment usually surfaces in its "Repack" step.${where}`,
details: { failure, ...(logs ? { logs } : {}) },
});
}
const outputs = isRecord(job.outputs) ? job.outputs : {};
const buildId = outputs.build_id;
if (typeof buildId !== "string" || !/^[A-Za-z0-9-]{1,128}$/.test(buildId)) {
if (typeof buildId !== "string" || !EAS_ID.test(buildId)) {
throw new Error("eas workflow:run did not return the repacked build id");
}
return buildId;
}
/**
* The run `eas workflow:run --json` (without `--wait`) started: its id and link, from its JSON or, when the JSON is
* missing, from the "See logs:" link it printed once the run existed. With neither, no run started, and the failure
* says why.
*/
export function hostedRepackRunStart(stdout: string, stderr: string, platform: "ios" | "android"): { id: string; url?: string } {
const printed = /See logs: (https:\/\/expo\.dev\/\S+)/.exec(stderr)?.[1];
let value: unknown;
try { value = JSON.parse(stdout); } catch { value = undefined; }
const url = isRecord(value) && typeof value.url === "string" && EXPO_URL.test(value.url) ? value.url : printed;
const id = isRecord(value) && typeof value.id === "string" ? value.id : url ? /\/workflows\/([^/?#\s]+)$/.exec(url)?.[1] : undefined;
if (id && EAS_ID.test(id)) return { id, ...(url ? { url } : {}) };
const tail = stderr.trim().split("\n").slice(-3).join(" | ").slice(0, 600);
throw new CiError({ code: "EAS_REPACK_FAILED", stage: "instrumentation",
message: `eas workflow:run did not start the hosted ${platform} repack: ${tail || "it printed no run"}`,
nextAction: "Check that EXPO_TOKEN can run workflows for this EAS project and retry.",
details: { failure: hostedRepackFailure(stderr) } });
}
/**
* The most time preparing the app may take once the build to repack is known: waiting for another job's repack,
* following it, and every repack this job starts, each run's wait included. It replaces a 30-minute wait per run, which
* with a share wait and two retries let one capture prepare for over an hour; a capture whose app takes longer than
* the session's allowance (SESSION_PREPARATION_MINUTES in run.ts) gets a fresh session.
*/
export const HOSTED_REPACK_BUDGET_MS = 30 * 60_000;
/** A repack starts only with this much of the budget left: a typical one takes about five minutes. */
const HOSTED_REPACK_MIN_ATTEMPT_MS = 5 * 60_000;
/** How often a running repack's state is read, and how many unreadable reads in a row mean Explore cannot tell. */
const HOSTED_REPACK_POLL_MS = 20_000;
const HOSTED_REPACK_UNREADABLE_READS = 6;
const TERMINAL_RUN_STATUSES = new Set(["SUCCESS", "FAILURE", "CANCELED"]);
/** What the hosted repack waits and reports with: the end of its budget, the clock and pause (tests replace both), and an abort. */
type RepackClock = { deadline: number; now: () => number; sleep?: ((milliseconds: number) => void) | undefined; signal?: AbortSignal | undefined };
async function pause(milliseconds: number, clock: RepackClock): Promise<void> {
if (clock.sleep) clock.sleep(milliseconds);
else {
try { await delay(milliseconds, undefined, clock.signal ? { signal: clock.signal } : {}); }
catch (error) { clock.signal?.throwIfAborted(); throw error; }
}
clock.signal?.throwIfAborted();
}
function timeoutError(platform: "ios" | "android", detail: string, logs?: string): CiError {
return new CiError({ code: "EAS_REPACK_FAILED", stage: "instrumentation",
message: `The hosted EAS repack of the ${platform} build did not finish within the ${HOSTED_REPACK_BUDGET_MS / 60_000} minutes a capture has to prepare its app: ${detail}.`,
nextAction: `Check the EAS workflow run for a long queue or a stuck step, then rerun Explore.${logs ? ` Logs: ${logs}` : ""}`,
details: { failure: "timeout", ...(logs ? { logs } : {}) } });
}
/**
* The failed step's logs from `eas workflow:logs --json` (the step that logged `Build step "…" failed`, else the last
* with any lines), so that a warning another step printed does not decide the failure; nothing when unreadable.
*/
function workflowLogText(stdout: string): string {
try {
const value: unknown = JSON.parse(stdout);
if (!isRecord(value)) return "";
const steps = Object.values(value).map(lines => Array.isArray(lines) ? lines.filter(isRecord).map(line => String(line.msg ?? "")) : []).filter(lines => lines.length);
const failed = [...steps].reverse().find(lines => lines.some(line => /Build step "[^"]+" failed/.test(line))) ?? steps.at(-1) ?? [];
return failed.slice(-400).join("\n");
} catch { return ""; }
}
type EasCall = (args: string[], timeoutMs: number, options?: { cwd?: string; strict?: boolean }) => { stdout: string; stderr: string };
/** What a command that threw still printed: defaultRunner keeps a timed-out or signalled command's output on its error. */
function printedBy(error: unknown): { stdout: string; stderr: string } {
const value = error as { stdout?: unknown; stderr?: unknown } | undefined;
return { stdout: typeof value?.stdout === "string" ? value.stdout : "", stderr: typeof value?.stderr === "string" ? value.stderr : "" };
}
/**
* One hosted repack run: started without `--wait`, so Explore always knows the run and waits on it within its own
* budget. eas-cli's `--wait` gives up after six failed status reads with exit code 13 and leaves the run going, and
* Explore then started another beside it. A run whose state stays unreadable, that outlasts the budget, or whose
* capture is stopped is cancelled (`eas workflow:cancel`) and not replaced: only a run EAS reports finished is
* followed by another. A start that timed out or was stopped is waited on when it printed its run, and otherwise
* not replaced, since it may have started one.
*/
async function hostedRepackRun(options: { eas: EasCall; staged: string; platform: "ios" | "android"; clock: RepackClock }): Promise<string> {
const { clock, platform } = options;
let started: { stdout: string; stderr: string };
let startFailure: unknown;
try { started = options.eas(["workflow:run", HOSTED_REPACK_WORKFLOW, "--json", "--non-interactive"], Math.min(10 * 60_000, clock.deadline - clock.now()), { cwd: options.staged }); }
catch (error) { started = printedBy(error); startFailure = error; }
let run: { id: string; url?: string };
try { run = hostedRepackRunStart(started.stdout, started.stderr, platform); }
catch (error) {
if (startFailure === undefined) throw error;
throw new CiError({ code: "EAS_REPACK_FAILED", stage: "instrumentation",
message: `eas workflow:run did not finish starting the hosted ${platform} repack (${startFailure instanceof Error ? startFailure.message : String(startFailure)}), and printed no run: one may have started, so Explore does not start another.`,
nextAction: "Check the project's workflow runs on expo.dev, cancel a repack still running, and rerun Explore.",
details: { failure: "unconfirmed" }, cause: startFailure });
}
const cancel = (): boolean => {
try { options.eas(["workflow:cancel", run.id, "--non-interactive"], 2 * 60_000, { strict: true }); return true; } catch { return false; }
};
try {
let unreadable = 0;
for (;;) {
clock.signal?.throwIfAborted();
let view = { stdout: "", stderr: "" };
try { view = options.eas(["workflow:view", run.id, "--json", "--non-interactive"], 2 * 60_000); } catch { /* Counted as an unreadable read below. */ }
let value: unknown;
try { value = JSON.parse(view.stdout); } catch { value = undefined; }
const status = isRecord(value) && typeof value.status === "string" ? value.status : undefined;
if (status && TERMINAL_RUN_STATUSES.has(status)) {
const jobs = isRecord(value) && Array.isArray(value.jobs) ? value.jobs.filter(isRecord) : [];
const job = jobs.find(entry => entry.key === "repack") ?? jobs[0];
let logs = "";
if (status === "FAILURE" && typeof job?.id === "string" && EAS_ID.test(job.id)) {
try { logs = workflowLogText(options.eas(["workflow:logs", job.id, "--json", "--non-interactive"], 2 * 60_000).stdout); } catch { /* The build's error still classifies it. */ }
}
return hostedRepackBuildId(view.stdout, `${started.stderr}\n${logs}`, platform);
}
unreadable = status ? 0 : unreadable + 1;
if (unreadable >= HOSTED_REPACK_UNREADABLE_READS) {
const cancelled = cancel();
throw new CiError({ code: "EAS_REPACK_FAILED", stage: "instrumentation",
message: `EAS did not report the state of the hosted ${platform} repack ${HOSTED_REPACK_UNREADABLE_READS} times in a row; ${cancelled ? "Explore cancelled the run" : "Explore could not cancel the run"} and did not start another beside it.`,
nextAction: `Check the EAS status and the run${cancelled ? "" : ", and cancel it if it is still going"}, then rerun Explore.${run.url ? ` Logs: ${run.url}` : ""}`,
details: { failure: "unconfirmed", ...(run.url ? { logs: run.url } : {}) } });
}
const left = clock.deadline - clock.now();
if (left <= 0) {
const cancelled = cancel();
throw timeoutError(platform, `run ${run.id} was still ${status ?? "unreadable"}; ${cancelled ? "it was cancelled" : "it could not be cancelled"}`, run.url);
}
await pause(Math.min(HOSTED_REPACK_POLL_MS, left), clock);
}
} catch (error) {
// A capture stopped while its repack runs (a cancelled job, Ctrl+C) leaves no run behind it.
if (clock.signal?.aborted) cancel();
throw error;
}
}
/**
* Where the matrix jobs of one GitHub run attempt meet to make one hosted repack of an app instead of one each: `scope`
* names the run attempt (repository, run id and attempt), `leader` the first target in config order that repacks this
* app, `target` this one, `rank` its place among them (the leader's is 0), and `rerun` an attempt after the first, which
* runs only the jobs it re-runs. In brentvatne/euxy#176 (review run 39506405687) four iOS targets each repacked EAS
* build a81d5d80 with the same commit's JavaScript at the same time, and one of the four failed.
*/
export type RepackShare = { scope: string; leader: string; target: string; rank: number; rerun?: boolean };
/** How long the target after the leader waits for a repack to appear on EAS before it repacks itself. */
const SHARED_REPACK_APPEAR_MS = 3 * 60_000;
/**
* Each later target waits this much longer, so that when the leader makes no repack (its baseline covered every route,
* or `max-parallel` kept it out of the first jobs) the next target's repack has appeared before the one after it would
* start another: the targets join one repack instead of all starting their own at the same moment.
*/
const SHARED_REPACK_STAGGER_MS = 60_000;
/** How long a target whose followed repack failed looks for the repack that replaces it (the leader's retry) before repacking itself. */
const SHARED_REPACK_GRACE_MS = 60_000;
/** Polls while no repack is followed, and while one is. */
const SHARED_REPACK_APPEAR_POLL_MS = 15_000;
const SHARED_REPACK_FOLLOW_POLL_MS = 30_000;
/** The recent builds of the platform a look reads (eas-cli's most): one this run made minutes ago is among them. */
const SHARED_REPACK_LIST_LIMIT = 50;
/**
* Every byte the hosted repack would upload, as one digest: the staged checkout's tracked files and the generated
* observer, path by path, with their kind, executable bit and content (a symlink by its target).
*/
export async function stagedProjectDigest(root: string): Promise<string> {
const hash = createHash("sha256");
const entries = (await readdir(root, { recursive: true, withFileTypes: true }))
.map(entry => ({ entry, path: relative(root, join(entry.parentPath, entry.name)).split(sep).join("/") }))
.sort((a, b) => a.path < b.path ? -1 : a.path > b.path ? 1 : 0);
for (const { entry, path } of entries) {
const full = join(root, path);
if (entry.isSymbolicLink()) hash.update(`link\0${path}\0${await readlink(full)}\0`);
else if (entry.isDirectory()) hash.update(`directory\0${path}\0`);
else if (entry.isFile()) {
const executable = ((await lstat(full)).mode & 0o111) !== 0;
hash.update(`file\0${path}\0${executable ? "x" : "-"}\0${createHash("sha256").update(await readFile(full)).digest("hex")}\0`);
} else throw new Error(`The staged copy holds ${path}, which is neither a file, a directory nor a symlink`);
}
return hash.digest("hex");
}
/**
* The key a shared repack's build carries in its message: everything the repack is made from, so two jobs share one only
* when each would have uploaded the same bytes for the same native build, profile, EAS environment (by name and by a
* snapshot of its variables) and Explore release, in the same run attempt.
*/
export function hostedRepackShareKey(input: { scope: string; platform: "ios" | "android"; nativeBuildId: string; easProfile: string;
environment: string; environmentVariables: string; project: string }): string {
return createHash("sha256").update(JSON.stringify({ contract: "expo-explore.repack-share.v2", explore: EXPLORE_VERSION, scope: input.scope,
platform: input.platform, nativeBuildId: input.nativeBuildId, easProfile: input.easProfile, environment: input.environment,
environmentVariables: input.environmentVariables, project: input.project })).digest("hex").slice(0, 32);
}
/** The marker a shared repack's build message ends with. */
export function hostedRepackShareMarker(key: string): string {
return `[explore share ${key}]`;
}
type ListedBuild = { id: string; status: string };
function sharedRepackBuild(build: unknown, marker: string, platform: "ios" | "android"): ListedBuild | undefined {
return isRecord(build) && typeof build.id === "string" && EAS_ID.test(build.id) && typeof build.status === "string"
&& typeof build.message === "string" && build.message.endsWith(marker) && String(build.platform).toLowerCase() === platform
? { id: build.id, status: build.status } : undefined;
}
/** The builds `eas build:list --json` lists that carry `marker` for `platform`, or undefined when the list is unreadable. */
export function sharedRepackBuilds(stdout: string, marker: string, platform: "ios" | "android"): ListedBuild[] | undefined {
let value: unknown;
try { value = JSON.parse(stdout); } catch { return undefined; }
if (!Array.isArray(value)) return undefined;
return value.flatMap(build => sharedRepackBuild(build, marker, platform) ?? []);
}
/** One build from `eas build:view --json`: null when it no longer carries `marker` for `platform`, undefined when unreadable. */
export function sharedRepackBuildView(stdout: string, marker: string, platform: "ios" | "android"): ListedBuild | null | undefined {
let value: unknown;
try { value = JSON.parse(stdout); } catch { return undefined; }
return isRecord(value) ? sharedRepackBuild(value, marker, platform) ?? null : undefined;
}
/**
* A snapshot of the EAS environment's variables the repack job reads: `eas env:list --format long` of the project's
* and the account's variables, which names each with its value (masked when sensitive or secret) and when it was last
* updated, so an edited value changes it. Undefined when either list fails.
*/
function easEnvironmentSnapshot(list: (scope: "project" | "account") => string | undefined): string | undefined {
const project = list("project"), account = list("account");
if (project === undefined || account === undefined) return undefined;
return createHash("sha256").update(JSON.stringify([project, account])).digest("hex");
}
const LIVE_BUILD_STATUSES = new Set(["NEW", "IN_QUEUE", "IN_PROGRESS"]);
/**
* How long a target looks for another's repack before it repacks itself: the leader not at all, and each target after it
* SHARED_REPACK_STAGGER_MS longer than the one before, from SHARED_REPACK_APPEAR_MS. A re-run attempt's leader may not
* run again, so there the stagger starts at once. After a followed repack failed, the wait is SHARED_REPACK_GRACE_MS,
* staggered the same way.
*/
export function sharedRepackWaitMs(share: Pick<RepackShare, "rank" | "rerun">, afterFailure = false): number {
const later = Math.max(0, share.rank - 1) * SHARED_REPACK_STAGGER_MS;
if (afterFailure) return SHARED_REPACK_GRACE_MS + later;
if (share.rank === 0) return 0;
return share.rerun ? share.rank * SHARED_REPACK_STAGGER_MS : SHARED_REPACK_APPEAR_MS + later;
}
/**
* A repack another job of this run attempt made, or is making, of exactly this app: its EAS build id once it finished,
* or undefined when this job should repack itself.
* - The leader looks once. Another target looks every 15 s for as long as sharedRepackWaitMs gives it, and joins any
* repack with the key it finds, whichever job started it.
* - A repack it finds is then read by its id every 30 s (a list of recent builds can lose it on a busy project).
* - One that fails, or no longer carries the key, is set aside, and for SHARED_REPACK_GRACE_MS the target looks for
* another with the key (the leader's retry, at once in brentvatne/euxy#176) before it repacks itself.
* - Looking stops while the budget still has time for a repack of this job's own, and following stops at the end of
* the budget. Three unreadable reads in a row also end the wait: it never waits on a job that cannot answer.
*/
async function sharedHostedRepack(options: { share: RepackShare; marker: string; platform: "ios" | "android"; list: () => string;
view: (id: string) => string; say: (message: string) => void; clock: RepackClock }): Promise<string | undefined> {
const { clock, say } = options;
const start = clock.now();
const lookLatest = clock.deadline - HOSTED_REPACK_MIN_ATTEMPT_MS;
const firstWait = sharedRepackWaitMs(options.share);
let lookUntil = Math.min(start + firstWait, lookLatest);
let followed: string | undefined;
const setAside = new Set<string>();
let unreadable = 0;
let announced = false;
for (;;) {
clock.signal?.throwIfAborted();
let read: ListedBuild[] | null | undefined;
try {
if (followed) {
const build = sharedRepackBuildView(options.view(followed), options.marker, options.platform);
read = build === undefined ? undefined : build === null ? [] : [build];
} else read = sharedRepackBuilds(options.list(), options.marker, options.platform);
} catch { read = undefined; }
if (!read) {
unreadable += 1;
if (unreadable >= 3) { say("EAS did not answer three times in a row; repacking here instead of waiting for another target's repack"); return undefined; }
} else {
unreadable = 0;
const builds = read.filter(build => !setAside.has(build.id));
const finished = builds.find(build => build.status === "FINISHED");
if (finished) {
say(`using EAS build ${finished.id}, the hosted repack of this commit another target of this run made, instead of repacking it again`);
return finished.id;
}
const live = builds.find(build => LIVE_BUILD_STATUSES.has(build.status));
if (followed && !live) {
say(`the shared repack (EAS build ${followed}) ${read[0] ? `ended ${read[0].status}` : "no longer carries this commit's key"}; looking for another for up to ${sharedRepackWaitMs(options.share, true) / 1000} s before repacking here`);
setAside.add(followed);
followed = undefined;
lookUntil = Math.min(clock.now() + sharedRepackWaitMs(options.share, true), lookLatest);
} else if (!followed && live) {
followed = live.id;
say(`waiting for EAS build ${live.id}, the hosted repack of this commit another target of this run started, instead of repacking it again`);
}
}
const now = clock.now();
if (followed && now >= clock.deadline) {
say(`the shared repack (EAS build ${followed}) did not finish in the time this capture has to prepare its app`);
return undefined;
}
if (!followed && now >= lookUntil) {
if (setAside.size) say("no other repack of this commit appeared; repacking here");
else if (firstWait) say(`no repack of this commit from ${options.share.leader} or another target appeared within ${Math.round((now - start) / 60_000)} minutes; repacking here`);
return undefined;
}
if (!followed && !announced && firstWait) {
announced = true;
say(`waiting up to ${Math.round(firstWait / 60_000)} minutes for the hosted repack ${options.share.leader} or another target makes of this commit, instead of repacking it again`);
}
const until = followed ? clock.deadline : lookUntil;
await pause(Math.max(0, Math.min(followed ? SHARED_REPACK_FOLLOW_POLL_MS : SHARED_REPACK_APPEAR_POLL_MS, until - now)), clock);
}
}
/**
* Repack on EAS instead of the runner: the tracked files of the clean checkout are staged beside the
* observer and a one-job workflow, uploaded with `eas workflow:run`, and the job's `build_id` comes back.
* No build tools, no `eas upload`, and the profile's EAS environment applies as it does for any build.
* With a `share`, a repack another job of this run attempt makes of the same staged bytes is used instead
* (sharedHostedRepack), and this job's own repack carries the key that lets the others use it. A repack that fails
* for a reason a rerun can fix starts again (hostedRepackRetries): in brentvatne/euxy#176 EAS ended a build that had
* repacked before it could save the app. All of it stays within the capture's budget (`clock.deadline`).
*/
async function prepareHostedRepack(options: {
platform: "ios" | "android";
@@ -451,7 +854,11 @@ async function prepareHostedRepack(options: {
provider: SourceRoute["provider"];
parameterKeys: readonly string[];
root: string;
share?: RepackShare | undefined;
say: (message: string) => void;
clock: RepackClock;
}): Promise<string> {
const { clock, platform } = options;
const staged = join(options.root, "project");
await mkdir(staged, { mode: 0o700 });
// Paths relative to the app directory, which is what is staged. With --full-name they were relative to the
@@ -481,19 +888,51 @@ async function prepareHostedRepack(options: {
message: error instanceof Error ? error.message : "The route observer could not be prepared.",
nextAction: "Check the installed Expo and navigation package compatibility for automatic capture instrumentation.", cause: error });
}
const privateEnvironment = { ...options.easEnvironment, TMPDIR: options.root, TMP: options.root, TEMP: options.root };
// Reads may fail and say so in their output; `strict` makes a failed command throw (a cancel that did not happen).
const eas: EasCall = (args, timeoutMs, { cwd = options.project, strict = false } = {}) => options.runner({ command: options.easCommand, args, cwd,
environment: cwd === staged ? { ...privateEnvironment, EAS_NO_VCS: "1" } : privateEnvironment, timeoutMs, ...(strict ? {} : { allowFailure: true }) });
let message = `expo-explore ${platform} capture: route observer repack`;
if (options.share) {
const environment = options.repackProfile.easEnvironment;
const variables = easEnvironmentSnapshot(scope => {
try {
const listedVariables = options.runner({ command: options.easCommand, args: ["env:list", "--environment", environment, "--scope", scope, "--format", "long"],
cwd: options.project, environment: privateEnvironment, timeoutMs: 2 * 60_000 });
return listedVariables.stdout;
} catch { return undefined; }
});
if (variables === undefined) options.say(`EAS did not list the ${environment} environment's variables; repacking here without sharing the repack`);
else {
// The digest is taken before the workflow file is written: the file's message carries the key made from it.
const marker = hostedRepackShareMarker(hostedRepackShareKey({ scope: options.share.scope, platform, nativeBuildId: options.nativeBuildId,
easProfile: options.easProfile, environment, environmentVariables: variables, project: await stagedProjectDigest(staged) }));
const shared = await sharedHostedRepack({ share: options.share, marker, platform, say: options.say, clock,
list: () => eas(["build:list", "--platform", platform, "--limit", String(SHARED_REPACK_LIST_LIMIT), "--json", "--non-interactive"], 2 * 60_000).stdout,
view: id => eas(["build:view", id, "--json"], 2 * 60_000).stdout });
if (shared) return shared;
message = `${message} ${marker}`;
}
}
const workflowPath = join(staged, HOSTED_REPACK_WORKFLOW);
await mkdir(dirname(workflowPath), { recursive: true });
await writeFile(workflowPath, hostedRepackWorkflow({ buildId: options.nativeBuildId, profile: options.easProfile,
environment: options.repackProfile.easEnvironment, message: `expo-explore ${options.platform} capture: route observer repack` }), { flag: "wx", mode: 0o600 });
const run = options.runner({
command: options.easCommand,
args: ["workflow:run", HOSTED_REPACK_WORKFLOW, "--wait", "--json", "--non-interactive"],
cwd: staged,
environment: { ...options.easEnvironment, EAS_NO_VCS: "1", TMPDIR: options.root, TMP: options.root, TEMP: options.root },
timeoutMs: 30 * 60_000,
allowFailure: true,
});
return hostedRepackBuildId(run.stdout, run.stderr, options.platform);
environment: options.repackProfile.easEnvironment, message }), { flag: "wx", mode: 0o600 });
let retries = 0;
for (;;) {
if (clock.deadline - clock.now() < HOSTED_REPACK_MIN_ATTEMPT_MS) throw timeoutError(platform, retries ? "too little time was left for another repack" : "too little time was left for a repack of its own");
try {
return await hostedRepackRun({ eas, staged, platform, clock });
} catch (error) {
// Only a run EAS reported finished gets here with a failure that earns another (hostedRepackRun), so none is left beside it.
const allowed = hostedRepackRetries(error);
if (retries >= allowed || clock.deadline - clock.now() < HOSTED_REPACK_MIN_ATTEMPT_MS) throw error;
retries += 1;
const logs = error instanceof CiError && typeof error.details?.logs === "string" ? error.details.logs : undefined;
options.say(`the hosted EAS repack${logs ? ` ${logs}` : ""} failed (${error instanceof Error ? error.message : String(error)}); starting another repack (${retries}/${allowed})`);
clock.signal?.throwIfAborted();
}
}
}
/**
@@ -520,6 +959,16 @@ export async function prepareExactSourceBuild(options: {
easProfile?: string | undefined;
/** Where to leave the instrumented app on this machine, for the capture to install; it must not exist yet. */
keepApp: string;
/** The other matrix jobs of this run that repack the same app, to make one hosted repack with (RepackShare). */
share?: RepackShare | undefined;
/** Says what the hosted repack is doing when it waits for another job's repack or starts another one. */
progress?: ((message: string) => void) | undefined;
/** When preparing the app must be done by (HOSTED_REPACK_BUDGET_MS after the session started beside it), on `now`'s clock. */
deadline?: number | undefined;
/** The clock and pause the waits and retries use; tests replace them so that minutes are not real time. */
now?: (() => number) | undefined;
sleep?: ((milliseconds: number) => void) | undefined;
signal?: AbortSignal | undefined;
}): Promise<PreparedBuild | undefined> {
const platform = options.platform ?? "ios";
const root = await mkdtemp(join(tmpdir(), "expo-explore-repack-"));
@@ -535,7 +984,8 @@ export async function prepareExactSourceBuild(options: {
const hosted = await prepareHostedRepack({ platform, runner: options.runner, easCommand: options.easCommand, project: options.project,
nativeBuildId: options.nativeBuildId, easProfile: options.easProfile, repackProfile: options.repackProfile,
easEnvironment: options.easEnvironment, restrictedEnvironment: options.repackEnvironment, provider: providers[0]!,
parameterKeys: captureObserverParameterKeys(options.routes), root });
parameterKeys: captureObserverParameterKeys(options.routes), root, share: options.share, say: options.progress ?? (() => undefined),
clock: { deadline: options.deadline ?? (options.now ?? Date.now)() + HOSTED_REPACK_BUDGET_MS, now: options.now ?? Date.now, sleep: options.sleep, signal: options.signal } });
{
// The hosted job leaves its app on EAS; the capture installs a copy downloaded here.
const downloaded = await downloadedAppPath(options.runner({ command: options.easCommand,
+12 -2
View File
@@ -1,5 +1,5 @@
import type { CaptureBundle, CaptureOutcome } from "../contracts/capture.js";
import { bundleColorSchemes, routeIdOfScreen, schemeScreenId } from "./appearance.js";
import { bundleColorSchemes, failedPassOutcome, routeIdOfScreen, schemeScreenId } from "./appearance.js";
import { CiError } from "../shared/ci-errors.js";
type CaptureSide = CaptureBundle["side"];
@@ -59,8 +59,10 @@ export function captureHealthIssues(bundle: CaptureBundle, expectedIds: readonly
for (const id of [...extras].sort()) {
// A secondary frame is expected evidence; its own outcome is the failure of that appearance pass, or its verified
// frame when the primary pass has none (the route is then published from this appearance alone).
// A pass that failed as a whole (sessionFailures) records its failure under every route's `<target>@<scheme>`.
const primaryCaptured = bundle.outcomes[routeIdOfScreen(id).routeId]?.status === "captured";
if (!expected.has(id) && !(expectedSecondary.has(id) && !selected.has(id) && hasVerifiedOutcome(bundle.outcomes[id]) !== primaryCaptured)) {
if (!expected.has(id) && !(expectedSecondary.has(id) && !selected.has(id)
&& (hasVerifiedOutcome(bundle.outcomes[id]) !== primaryCaptured || failedPassOutcome(bundle, id)))) {
add(id, "CAPTURE_TARGET_UNEXPECTED", "Capture evidence contains a target absent from the source plan.", inventoryAction);
}
}
@@ -121,6 +123,14 @@ export function captureHealthIssues(bundle: CaptureBundle, expectedIds: readonly
const INTEGRITY_CODES = new Set(["CAPTURE_SIDE_MISMATCH", "CAPTURE_COMMIT_INVALID", "CAPTURE_SELECTION_DUPLICATE", "CAPTURE_TARGET_UNEXPECTED",
"CAPTURE_TARGET_MISSING", "CAPTURE_OUTCOME_MISSING", "CAPTURE_IMAGE_BINDING_INVALID", "CAPTURE_IMAGE_PROVENANCE_INVALID"]);
/**
* Whether a failure casts doubt on everything a capture took, so its other appearances cannot be kept as a partial
* report: a checkout that changed under it, or evidence that failed its integrity checks.
*/
export function capturedEvidenceInDoubt(code: string): boolean {
return /^(?:SOURCE|CHECKOUT)_/.test(code) || code === "CAPTURE_INCOMPLETE" || INTEGRITY_CODES.has(code);
}
export function sortedCaptureHealthIssues(inputs: readonly CaptureHealthInput[]): CaptureHealthIssue[] {
const compare = (a: string, b: string) => a < b ? -1 : a > b ? 1 : 0;
return inputs.flatMap(({ bundle, selectedIds, side }) => captureHealthIssues(bundle, selectedIds, side))
+146
View File
@@ -0,0 +1,146 @@
/**
* The system software keyboard in a controller snapshot (agent-device's flat `nodes`, which the Argent bridge also
* returns), and how long capture waits for it. Recognition is deliberately narrow: dropping one of the app's own elements
* hides a real change, which is worse than the keyboard noise this removes. A node is the keyboard's only when it, or a
* node above it, provably belongs to the system keyboard; nothing is inferred from where an element sits on the screen,
* and a `Key` alone proves nothing (React Native's `accessibilityRole="keyboardkey"` gives an app's PIN pad that type).
*/
type Rect = { x: number; y: number; width: number; height: number };
type SnapshotNode = Record<string, unknown>;
/** Android input methods: their windows name the IME's package, not the app's. */
export const INPUT_METHOD_PACKAGES: ReadonlySet<string> = new Set([
"com.google.android.inputmethod.latin",
"com.android.inputmethod.latin",
"com.samsung.android.honeyboard",
"com.touchtype.swiftkey",
"com.microsoft.swiftkey",
]);
/**
* UIKit's identifiers for the input view placeholder and the prediction bar. They mark a window as the keyboard's (see
* the Argent bridge's `keyboardWindows`); inside the app's own window an element with one of them is the app's.
*/
export const KEYBOARD_WINDOW_IDENTIFIERS: ReadonlySet<string> = new Set(["inputView", "SystemInputAssistantView"]);
/**
* The keyboard's native view classes, which agent-device's host accessibility bridge reports as `role`. The dock's Emoji
* and Dictate buttons (UIKeyboardDockItemButton) can sit outside the key plane's bounds, so they are named here.
*/
const KEYBOARD_VIEW_CLASSES = new Set(["UIKeyboardDockItemButton", "UIAccessibilityElementKBKey", "UIKeyboardLayoutStar",
"TUIPredictionView", "TUIPredictionViewCell", "_UIRemoteKeyboardPlaceholderView", "TUISystemInputAssistantView",
"TUISystemInputAssistantPageView"]);
const EDITABLE_TYPES = new Set(["textfield", "securetextfield", "searchfield", "textview"]);
const text = (value: unknown) => typeof value === "string" ? value : "";
function rectOf(node: SnapshotNode): Rect | undefined {
const rect = node.rect as Partial<Rect> | undefined;
return rect && [rect.x, rect.y, rect.width, rect.height].every(Number.isFinite) && rect.width! > 0 && rect.height! > 0 ? rect as Rect : undefined;
}
/**
* A node that belongs to the system keyboard by what it is, not where it is: XCTest's `Keyboard` element (both
* controllers), a node of a window the Argent bridge found to be the keyboard's (`systemKeyboard`), one of the keyboard's
* native view classes, or a node of an Android input method's window.
*/
function keyboardRoot(node: SnapshotNode): boolean {
return node.systemKeyboard === true || text(node.type).toLowerCase() === "keyboard" || KEYBOARD_VIEW_CLASSES.has(text(node.role))
|| INPUT_METHOD_PACKAGES.has(text(node.bundleId ?? node.packageName ?? node.package));
}
/** Every node that belongs to the system keyboard: those `keyboardRoot` names and the nodes beneath them. */
export function keyboardNodes(nodes: readonly SnapshotNode[]): Set<SnapshotNode> {
const owned = new Set<SnapshotNode>();
if (!nodes.some(keyboardRoot)) return owned;
// Only an explicit parent links a node to a keyboard above it: a node without one is no node's descendant.
const byIndex = new Map(nodes.filter(node => typeof node.index === "number").map(node => [node.index as number, node]));
for (const node of nodes) {
const seen = new Set<SnapshotNode>();
for (let current: SnapshotNode | undefined = node; current && !seen.has(current);
current = typeof current.parentIndex === "number" ? byIndex.get(current.parentIndex) : undefined) {
seen.add(current);
if (keyboardRoot(current)) { owned.add(node); break; }
}
}
return owned;
}
function union(rects: readonly Rect[]): Rect | undefined {
if (!rects.length) return undefined;
const x = Math.min(...rects.map(rect => rect.x)), y = Math.min(...rects.map(rect => rect.y));
return { x, y, width: Math.max(...rects.map(rect => rect.x + rect.width)) - x, height: Math.max(...rects.map(rect => rect.y + rect.height)) - y };
}
/**
* Where the visible keyboard is, from the nodes that span it (the `Keyboard` element, the keyboard window's input view and
* prediction bar, the native views, an input method's window), or undefined when none shows. Containers inside the
* keyboard can report a scroll extent past the screen, so not every node it owns counts.
*/
export function keyboardBounds(nodes: readonly SnapshotNode[]): Rect | undefined {
return union(nodes.filter(node => node.visible !== false && keyboardRoot(node) && (node.systemKeyboard !== true
|| text(node.type).toLowerCase() === "keyboard" || KEYBOARD_WINDOW_IDENTIFIERS.has(text(node.identifier)))).flatMap(node => rectOf(node) ?? []));
}
/**
* A text input that has the keyboard focus: agent-device reports `focused` from XCTest's keyboard focus and Android's
* focused flag; the Argent bridge passes on Argent's own.
*/
export function focusedTextInput(nodes: readonly SnapshotNode[]): boolean {
return nodes.some(node => node.focused === true && node.visible !== false
&& (EDITABLE_TYPES.has(text(node.type).toLowerCase()) || /EditText$/.test(text(node.type)) || node.editable === true));
}
/** A snapshot as the wait reads it: Argent also says whether a keyboard shows, which its Android tree cannot. */
export type KeyboardRead = { nodes: readonly SnapshotNode[]; keyboardVisible?: boolean };
/** Where the keyboard is in one read: its bounds, or only that one shows when the controller cannot place it. */
function keyboardPosition(read: KeyboardRead): string | undefined {
const bounds = keyboardBounds(read.nodes);
return bounds ? JSON.stringify(bounds) : read.keyboardVisible === true ? "visible" : undefined;
}
/**
* How capture waits for a focused field's keyboard. Reads are 250 ms apart at most, and at least two follow the first,
* so a controller whose snapshot takes a second or more still sees the keyboard rise. Waiting for a keyboard that has not
* appeared ends after a second; one that has appeared but has not yet held still is read until it does, up to six reads
* or four seconds, so a slow controller cannot keep the capture waiting.
*/
export const KEYBOARD_POLL_MS = 250;
const KEYBOARD_WAIT_MS = 1_000;
const KEYBOARD_CAP_MS = 4_000;
const KEYBOARD_MIN_READS = 2;
const KEYBOARD_MAX_READS = 6;
export type KeyboardSettle = {
/** `shown`: in the same place in two reads running. `unfocused`: the field let go and no keyboard shows. */
readonly outcome: "shown" | "moving" | "absent" | "unfocused";
readonly reads: number;
/** Whether to read the screen again, `elapsedMs` after the first read. */
more(elapsedMs: number): boolean;
read(next: KeyboardRead): void;
};
/**
* The wait for a screen whose text field has the focus, or undefined when none has it. A keyboard already up counts only
* once a later read finds it in the same place: one caught rising or going down is still moving.
*/
export function keyboardSettle(first: KeyboardRead): KeyboardSettle | undefined {
if (!focusedTextInput(first.nodes)) return undefined;
let previous = keyboardPosition(first), reads = 0;
let outcome: KeyboardSettle["outcome"] = previous ? "moving" : "absent";
return {
get outcome() { return outcome; },
get reads() { return reads; },
more: elapsedMs => (outcome === "moving" || outcome === "absent") && reads < KEYBOARD_MAX_READS
&& (reads < KEYBOARD_MIN_READS || (elapsedMs < KEYBOARD_CAP_MS && (outcome === "moving" || elapsedMs < KEYBOARD_WAIT_MS))),
read: next => {
reads += 1;
const position = keyboardPosition(next);
outcome = position && position === previous ? "shown" : position ? "moving" : focusedTextInput(next.nodes) ? "absent" : "unfocused";
previous = position;
},
};
}
+16 -3
View File
@@ -92,7 +92,7 @@ export function approveLinkScheme(options: { runner: CommandRunner; udid: string
}
/** The pixel size of a PNG, from its header. */
function pngSize(bytes: Buffer): { width: number; height: number } | undefined {
export function pngSize(bytes: Buffer): { width: number; height: number } | undefined {
if (bytes.length < 24 || bytes.toString("latin1", 12, 16) !== "IHDR") return undefined;
return { width: bytes.readUInt32BE(16), height: bytes.readUInt32BE(20) };
}
@@ -186,6 +186,18 @@ export async function assertLocalEmulatorCompatible(apk: string, abi: string): P
nextAction: `Build an APK that includes ${abi} (a universal build does), or capture Android on EAS Simulator with --simulator eas.` });
}
/**
* Whether agent-device still holds a claim on an emulator serial, one an interrupted run left behind: it refuses
* such a serial even when it reports the claim as stale, so startLocalEmulator skips that port.
*/
export function agentDeviceClaims(options: { runner: CommandRunner; agentDevice: string; serial: string; cwd: string; environment: NodeJS.ProcessEnv }): boolean {
const status = options.runner({ command: options.agentDevice, args: ["device", "status", "--platform", "android", "--serial", options.serial, "--json"],
cwd: options.cwd, environment: options.environment, timeoutMs: 30_000, allowFailure: true });
let data: { claims?: unknown[]; hiddenStaleClaims?: number } | undefined;
try { data = (JSON.parse(status.stdout) as { data?: typeof data }).data; } catch { return false; }
return (data?.claims?.length ?? 0) + (data?.hiddenStaleClaims ?? 0) > 0;
}
/**
* Creates an AVD of the configured device profile (such as pixel_7) and starts it without a window. It boots in the
* background while a capture repacks the app; waitForLocalEmulator waits for it afterwards.
@@ -282,8 +294,9 @@ export async function reservePort(port: number, environment: NodeJS.ProcessEnv):
const abandoned = async (path: string) => {
const info = await stat(path).catch(() => undefined);
if (!info) return false;
const owner = await readFile(join(path, "owner"), "utf8").catch(() => undefined);
return owner === undefined ? Date.now() - info.mtimeMs > 10_000 : !processAlive(Number(owner));
// writeFile creates the owner file before it writes the pid, so an empty one is a holder mid-write too.
const owner = await readFile(join(path, "owner"), "utf8").catch(() => "");
return owner === "" ? Date.now() - info.mtimeMs > 10_000 : !processAlive(Number(owner));
};
for (;;) {
try { await mkdir(mutex); await writeFile(join(mutex, "owner"), String(process.pid)); break; }
+80 -16
View File
@@ -17,6 +17,7 @@ import { type CaptureSetupSnapshot, credentialSavePromptDecline, describeSetupAc
import { silentProgress } from "../shared/progress.js";
import type { CaptureSetupReady } from "../config/explore-config.js";
import { READINESS_POLL_MS, defaultReadinessSeconds } from "./readiness.js";
import { KEYBOARD_POLL_MS, keyboardSettle, type KeyboardRead, type KeyboardSettle } from "./keyboard.js";
import { CAPTURE_OBSERVER_IDENTIFIER, captureObserverMarkers, parseCaptureDestination, captureDestinationMatches } from "./observer.js";
import { type CiAgentRuntimeOptions, runCiAgentNavigation, assertAgentCaptureContext } from "./agent/runtime.js";
import { createExplorationInputs } from "../config/exploration-inputs.js";
@@ -92,6 +93,11 @@ function snapshotNodes(raw: string): Record<string, unknown>[] {
return data.nodes;
}
/** Whether the controller says a keyboard shows: Argent does, and its Android tree has no input method window to show it. */
function keyboardVisibility(data: Record<string, unknown>): { keyboardVisible?: boolean } {
return typeof data.keyboardVisible === "boolean" ? { keyboardVisible: data.keyboardVisible } : {};
}
/** The visible element each required label matched, so an image assessor can find icon-only controls. */
export function verifiedCaptureLabels(nodes: readonly Record<string, unknown>[], requiredLabels: readonly string[],
/** The screen's portrait height, when the screenshot is that screen turned a quarter clockwise (an opened iPhone Duo). */
@@ -106,6 +112,24 @@ export function verifiedCaptureLabels(nodes: readonly Record<string, unknown>[],
}));
}
/**
* The full-screen logical space a screenshot shows: the one Application node at the origin whose aspect is the image's,
* at any pixel density. Failing that, the one whose aspect is the image's transpose (`transposed`): the image is then the
* screen turned a quarter. A square screen matches both ways and is read upright. `slack` is how many image pixels the
* root, scaled to the image, may miss it by; capture asks for an exact match.
*/
export function screenSpace(nodes: readonly Record<string, unknown>[], image: { width: number; height: number }, slack = 0): { width: number; height: number; transposed: boolean } | undefined {
const roots = (transposed: boolean) => nodes.filter(node => node.type === "Application" && isRecord(node.rect)
&& node.rect.x === 0 && node.rect.y === 0 && typeof node.rect.width === "number" && typeof node.rect.height === "number"
&& Number.isFinite(node.rect.width) && Number.isFinite(node.rect.height) && node.rect.width > 0 && node.rect.height > 0
&& Math.abs(transposed ? node.rect.width * image.width - node.rect.height * image.height : node.rect.width * image.height - node.rect.height * image.width) <= slack * node.rect.height);
const upright = roots(false), turned = upright.length ? [] : roots(true);
const matched = upright.length ? upright : turned;
if (matched.length !== 1) return undefined;
const rect = matched[0]!.rect as { width: number; height: number };
return { width: rect.width, height: rect.height, transposed: turned.length === 1 };
}
/**
* A swipe that closes a presented sheet, or undefined when the snapshot does not prove a sheet.
* The route observer is drawn at the focused scene's top-left corner. A sheet's scene starts below
@@ -377,7 +401,7 @@ export async function capturePlannedRoutes(options: {
lastSnapshot = raw;
const generation = typeof data.refsGeneration === "number" ? data.refsGeneration : undefined;
const decline = declined < 2 && generation ? credentialSavePromptDecline(nodes) : undefined;
if (!decline) return { nodes, truncated: false, ...(generation ? { refsGeneration: generation } : {}) };
if (!decline) return { nodes, truncated: false, ...(generation ? { refsGeneration: generation } : {}), ...keyboardVisibility(data) };
controller(["press", `@${String(decline.ref).replace(/^@/, "")}~s${generation}`, "--level", "full", "--json"]);
recordDeclinedPrompt("deterministic");
controller(["wait", "500", "--level", "full", "--json"]);
@@ -507,6 +531,35 @@ export async function capturePlannedRoutes(options: {
.sort());
/** The controller's own quiescence gate; a timeout is reported by the snapshot comparison below. */
const settle = () => { controller(["wait", "stable", "500", "8000", "--level", "full", "--json"], true); };
/**
* A screen that focused a text field shows the keyboard that field raises, as its users see it. The keyboard rises
* after the screen has settled, so the controller's quiescence gate can pass in between: in brentvatne/euxy#172
* (review run 39362642343) the rename sheet's base was captured with its cursor showing and no keyboard, its head with
* the keyboard up, and the 48% mismatch was the keyboard. The screen is read again until the keyboard is in the same
* place twice running (see keyboardSettle); a keyboard that never comes is not waited for long, and none is dismissed.
*/
const awaitKeyboard = (route: SourceRoute, first: KeyboardRead): readonly Record<string, unknown>[] => {
const settle = keyboardSettle(first);
if (!settle) return first.nodes;
const startedAt = clock();
let current = first, pollStartedAt = startedAt;
// A read counts for at least one poll interval, so a clock that stands still (a fake controller) still expires.
const elapsed = () => Math.max(clock() - startedAt, settle.reads * KEYBOARD_POLL_MS);
while (settle.more(elapsed())) {
(capture.sleep ?? pause)(KEYBOARD_POLL_MS - (clock() - pollStartedAt));
pollStartedAt = clock();
current = snapshot();
settle.read(current);
}
recordKeyboardWait(route.id, settle, elapsed());
return current.nodes;
};
const recordKeyboardWait = (route: string | undefined, settle: KeyboardSettle, elapsedMs: number) => {
const diagnostics = join(store.root, "capture-diagnostics");
mkdirSync(diagnostics, { recursive: true, mode: 0o700 });
appendFileSync(join(diagnostics, "keyboard.jsonl"), `${JSON.stringify({ ...(route ? { route } : { phase: "agent-observe" }), outcome: settle.outcome,
reads: settle.reads, elapsedMs, at: new Date(clock()).toISOString() })}\n`, { mode: 0o600 });
};
const attempted = new Set<string>();
const captureCurrent = async (route: SourceRoute, attempt: number, before?: readonly Record<string, unknown>[],
unattributed?: NonNullable<CaptureBundle["outcomes"][string]["observed"]> & { message: string }) => {
@@ -514,9 +567,11 @@ export async function capturePlannedRoutes(options: {
attempted.add(route.id);
const temporary = await mkdtemp(join(store.root, ".capture-"));
try {
const accepted = before ?? waitForDestination(route, () => resolveDeepLinkConfirmation());
if (unattributed) assertNoAlert({ route, nodes: accepted });
else landing(accepted, route, inventory);
const arrived = before ?? waitForDestination(route, () => resolveDeepLinkConfirmation());
if (unattributed) assertNoAlert({ route, nodes: arrived });
else landing(arrived, route, inventory);
const accepted = unattributed ? arrived : awaitKeyboard(route, { nodes: arrived });
if (accepted !== arrived) landing(accepted, route, inventory);
assertSafeScreenshot(accepted);
const screenshot = join(temporary, "screen.png");
await writeFile(screenshot, "", { flag: "wx", mode: 0o600 });
@@ -663,7 +718,7 @@ export async function capturePlannedRoutes(options: {
const generation = typeof data.refsGeneration === "number" ? data.refsGeneration : undefined;
const decline = declined < 2 && generation ? credentialSavePromptDecline(nodes) : undefined;
if (!decline) {
const snapshot: CaptureSetupSnapshot = { nodes, truncated: false, ...(generation ? { refsGeneration: generation } : {}) };
const snapshot: CaptureSetupSnapshot = { nodes, truncated: false, ...(generation ? { refsGeneration: generation } : {}), ...keyboardVisibility(data) };
return { snapshot, raw };
}
await command(["press", `@${String(decline.ref).replace(/^@/, "")}~s${generation}`, "--level", "full", "--json"], signal);
@@ -693,18 +748,11 @@ export async function capturePlannedRoutes(options: {
* then the transpose of the frame, and a tap is turned back (see `toLogical`).
*/
const bounds = (frame: Candidate) => {
const roots = (transposed: boolean) => frame.snapshot.nodes.filter(node => node.type === "Application" && isRecord(node.rect)
&& node.rect.x === 0 && node.rect.y === 0 && typeof node.rect.width === "number" && typeof node.rect.height === "number"
&& Number.isFinite(node.rect.width) && Number.isFinite(node.rect.height) && node.rect.width > 0 && node.rect.height > 0
&& (transposed ? node.rect.width * frame.width === node.rect.height * frame.height : node.rect.width * frame.height === node.rect.height * frame.width));
// A square screen matches both ways; it is read upright, as before rotated frames existed.
const upright = roots(false), turned = upright.length ? [] : roots(true);
const matched = upright.length ? upright : turned;
if (matched.length !== 1) throw captureError("AGENT_COORDINATE_BOUNDS_UNVERIFIED", "The screenshot does not identify one matching full-screen logical coordinate space.");
// The official iOS screenshot is full-screen. Its default is 1x, but a proven
// uniform pixel density must still map to XCTest points rather than raw pixels.
const rect = matched[0]!.rect as { width: number; height: number };
return { width: rect.width, height: rect.height, transposed: turned.length === 1 };
const space = screenSpace(frame.snapshot.nodes, frame);
if (!space) throw captureError("AGENT_COORDINATE_BOUNDS_UNVERIFIED", "The screenshot does not identify one matching full-screen logical coordinate space.");
return space;
};
/**
* A normalized point (x, y) in the screenshot, in the tree's logical points. Upright, it scales by the root. Turned:
@@ -769,16 +817,32 @@ export async function capturePlannedRoutes(options: {
}
}
};
/** The same wait for a focused field's keyboard as a route link's capture, before an observation's screenshot. */
const awaitKeyboardAsync = async (first: CaptureSetupSnapshot, signal: AbortSignal): Promise<CaptureSetupSnapshot> => {
const settle = keyboardSettle(first);
if (!settle) return first;
const startedAt = clock();
let current = first;
const elapsed = () => Math.max(clock() - startedAt, settle.reads * KEYBOARD_POLL_MS);
while (settle.more(elapsed())) {
await sleep(KEYBOARD_POLL_MS, undefined, { signal: liveSignal(signal) });
current = (await readSnapshot(signal)).snapshot;
settle.read(current);
}
recordKeyboardWait(undefined, settle, elapsed());
return current;
};
const observe = async (signal: AbortSignal, options?: { includeImage: boolean }) => {
await discard();
await appIdentity(signal);
const first = await readSnapshotThroughDeferral(signal);
const before = first.snapshot;
let before = first.snapshot;
if (first.retries > 0) {
await appendFile(join(dirname(tracePath), "accessibility-deferral.jsonl"), `${JSON.stringify({ retries: first.retries, waitedMs: first.deferredMs,
at: new Date(clock()).toISOString() })}\n`, { mode: 0o600 });
}
if (options?.includeImage === false) return { snapshot: before };
before = await awaitKeyboardAsync(before, signal);
try { assertSafeScreenshot(before.nodes); }
catch (error) {
if (error instanceof CiError && error.code === "CAPTURE_SENSITIVE_INPUT") return { snapshot: before, imageUnavailable: { code: error.code, message: error.message } };
+385 -80
View File
@@ -1,5 +1,5 @@
import { appearanceArgs, argentColorSchemes, bundleColorSchemes, mergeSchemeOutcomes, resolveColorSchemes, routeIdOfScreen, type AppearanceSetting, type ColorScheme } from "./appearance.js";
import { assertNoCaptureHealthIssues, captureHealthIssues, partitionCaptureHealthIssues } from "./health.js";
import { appearanceArgs, argentColorSchemes, bundleColorSchemes, mergeSchemeOutcomes, resolveColorSchemes, routeIdOfScreen, schemeScreenId, type AppearanceSetting, type ColorScheme } from "./appearance.js";
import { assertNoCaptureHealthIssues, capturedEvidenceInDoubt, captureHealthIssues, partitionCaptureHealthIssues } from "./health.js";
import { type RouteResetSetting } from "./route-reset.js";
import { defaultReadinessSeconds } from "./readiness.js";
import { captureLaunch, ffmpegRecordingTools, launchDirectory, type LaunchSetting, type RecordingTools } from "./launch-capture.js";
@@ -13,7 +13,7 @@ import { sourceRouteNames } from "../contracts/source-route.js";
import type { SourceCapturePlan } from "../source/capture-plan.js";
import { CONTRACT, type ExploreMap, type Platform, type RunStatus } from "../types.js";
import { ExploreStore, validateMap } from "../map/store.js";
import { CiError, serializeCiError, toCiError } from "../shared/ci-errors.js";
import { CiError, environmentSecrets, serializeCiError, toCiError } from "../shared/ci-errors.js";
import type { CaptureSetupAction, DevicePose } from "../config/explore-config.js";
import { type CiAgentRuntimeOptions } from "./agent/runtime.js";
import { EAS_CLI_VERSION, CONTROLLER_SESSION, isRecord, parseJson, baseEnvironment, nonEmpty } from "./common.js";
@@ -21,15 +21,18 @@ import { lstatIfExists, isWithin, temporaryCaptureDirectory, absoluteContractFil
import { loadAppIcon, loadCachedAppIcon, storeAppIcon } from "./app-icon.js";
import { defaultRunner, runAgentControllerCommand } from "./controller.js";
import { silentProgress, type Progress } from "../shared/progress.js";
import { assertCaptureEasProfile, assertSimulatorAvailable, fingerprintFrom, hasEasEnvironmentFetchFailure, nativeBuildFrom, existingBuild, repackProfileFrom, prepareExactSourceBuild, type RepackMode } from "./eas-build.js";
import { SIMULATOR_DOTENV_FILE_NAME, SIMULATOR_DOTENV_HEADER, simulatorSessionName, simulatorStartFailure, readSimulatorDotenvHandoff, assertSimulatorSessionIdentity, assertSimulatorStopped, collectRuntime, controllerVersion, simulatorOsVersion, type ActiveSimulatorRuntime, type SimulatorConnection } from "./simulator.js";
import { assertCaptureEasProfile, assertSimulatorAvailable, fingerprintFrom, hasEasEnvironmentFetchFailure, nativeBuildFrom, existingBuild, repackProfileFrom, prepareExactSourceBuild, HOSTED_REPACK_BUDGET_MS, type RepackMode, type RepackShare } from "./eas-build.js";
import { SIMULATOR_DOTENV_FILE_NAME, SIMULATOR_DOTENV_HEADER, endedOwnSimulatorSession, liveSimulatorSessionsNamed, simulatorSessionName, simulatorStartFailure, readSimulatorDotenvHandoff, assertSimulatorSessionIdentity, assertSimulatorStopped, collectRuntime, controllerVersion, simulatorOsVersion, type ActiveSimulatorRuntime, type SimulatorConnection } from "./simulator.js";
import { argentControllerInvocation } from "./controller-invocation.js";
import { startArgentBridgeRunner, type ArgentBridgeRunner } from "./argent-bridge-host.js";
import { sharedBuild } from "./shared-build.js";
import { LOCAL_EMULATOR_STATE_PREFIX, LOCAL_SIMULATOR_STATE_PREFIX, assertLocalEmulatorCompatible, createLocalSimulator, startLocalEmulator, waitForLocalEmulator, waitForLocalSimulator, deleteLocalEmulator,
import { LOCAL_EMULATOR_STATE_PREFIX, LOCAL_SIMULATOR_STATE_PREFIX, agentDeviceClaims, assertLocalEmulatorCompatible, createLocalSimulator, startLocalEmulator, waitForLocalEmulator, waitForLocalSimulator, deleteLocalEmulator,
deleteLocalSimulator, hostEmulatorAbi, installOnLocalEmulator, installOnLocalSimulator, approveLinkScheme, localPoseShown,
setLocalEmulatorAppearance, setLocalSimulatorAppearance, startLocalArgentServer, stopLocalAgentDevice, stopLocalArgentServer, type LocalArgentServer } from "./local-simulator.js";
pngSize, setLocalEmulatorAppearance, setLocalSimulatorAppearance, startLocalArgentServer, stopLocalAgentDevice, stopLocalArgentServer, type LocalArgentServer } from "./local-simulator.js";
import { ensureArgentToolServer } from "../shared/controller-cache.js";
import { openLocalCaptureRecord } from "./local-registry.js";
import { approveSessionUrlSchemes, simulatorControl, simulatorControlActions, type EasCommand } from "./simulator-control.js";
import { solidColour } from "../shared/png-evidence.js";
import { type UnattributedSetting } from "./unattributed.js";
import { appScheme, hasConcreteRoutePath, unavailableRoute, capturePlannedRoutes, deviceArgs } from "./route-capture.js";
@@ -68,8 +71,16 @@ export type CiMapOptions = {
repack?: RepackMode;
/** A folder the review's captures share builds through: targets of one platform, profile and commit build and repack once. */
buildShare?: string;
/** The matrix jobs of this GitHub run that repack the same app: they make one hosted repack per commit between them. */
repackShare?: RepackShare;
/** Which appearances to capture; `auto` (default) follows the app's `userInterfaceStyle`. */
appearance?: AppearanceSetting;
/**
* One session's share of the appearance passes when a side splits them across sessions: `capture` names the passes
* this run makes, out of the `all` it would make alone, so frames keep the ids of a single run (`<id>@dark`) and the
* exports merge. Absent: every resolved appearance.
*/
appearancePass?: { all: readonly ColorScheme[]; capture: readonly ColorScheme[] };
/** Whether to observe the app launch (splash screen) per appearance; `video` also keeps the recording. Defaults to `off`. */
launch?: LaunchSetting;
/** Decodes launch recordings; defaults to ffmpeg when installed. */
@@ -98,8 +109,18 @@ export type CiMapResult = {
};
function parseJsonOrUndefined(text: string): unknown {
try { return JSON.parse(text); } catch { return undefined; }
/** New EAS sessions a capture starts in place of one EAS failed to start (ERRORED while booting, or never named). */
const SESSION_START_RETRIES = 5;
/**
* The most time, from the first failed start, that replacement sessions may take, waits included. It keeps a review
* inside its CI job (90 minutes for a baseline) with a native build, a repack and both captures still to run.
*/
const SESSION_RETRY_BUDGET_MS = 30 * 60_000;
/** How long one `eas simulator:start` may take: eas-cli waits up to 15 minutes for a session, not counting the queue. */
const SESSION_START_TIMEOUT_MS = 20 * 60_000;
/** The wait before each of them: 10 s, doubling, at most 2 minutes. */
export function sessionRetryDelayMs(retry: number): number {
return Math.min(10_000 * 2 ** (retry - 1), 120_000);
}
/**
@@ -117,6 +138,17 @@ function sessionDurationLimit(error: unknown): number | undefined {
return match ? Math.floor(Number(match[1])) : undefined;
}
/**
* A start failure EAS may not repeat on the next try: a failed request, a dropped connection, a timeout, or no capacity
* right now. An account's quota, plan or permissions are not, and fail at once rather than after every retry.
*/
function transientSessionStartFailure(error: unknown): boolean {
if (error instanceof CiError && error.code === "CONTROLLER_REQUEST_TIMEOUT") return true;
const message = error instanceof Error ? error.message : String(error);
if (/not authori[sz]ed|unauthori[sz]ed|forbidden|permission|quota|billing|max duration/i.test(message)) return false;
return /GraphQL request failed|ECONNRESET|ETIMEDOUT|EAI_AGAIN|socket hang up|timed out|\b50[234]\b|capacity/i.test(message);
}
function positiveInteger(value: number, name: string): number {
if (!Number.isInteger(value) || value <= 0) {
throw new Error(`${name} must be a positive integer`);
@@ -195,7 +227,7 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
});
const routes = inventory.filter(route => selected.has(route.id));
const setup = revisionSetup(options.setup ?? [], plan, side);
const runner = options.runner ?? defaultRunner;
let runner = options.runner ?? defaultRunner;
const sourceEnvironment = options.environment ?? process.env;
const controller = options.controller ?? "agent-device";
const local = options.simulator === "local";
@@ -328,13 +360,21 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
+ (routes.length - missingRoutes.length ? `, ${routes.length - missingRoutes.length} reused from the baseline` : "")
+ (missingRoutes.length ? `, ${missingRoutes.length} to capture` : ""));
let captureFailure: unknown;
// Later appearance passes that failed after the primary one captured (see the pass loop).
const sessionFailures: NonNullable<CaptureBundle["sessionFailures"]> = [];
const finish = async (): Promise<CiMapResult> => {
if (!(await store.exists())) await store.initialize(map, false);
const incomplete = captureFailure !== undefined || selectedIds.some(id => outcomes[id]?.status !== "captured" || outcomes[id]?.landing !== "verified")
|| map.gaps.some(gap => gap.status === "open");
// A session making only secondary passes owns the `<id>@<scheme>` frames of those passes, not the primary ones, even
// when it could not switch to them and so captured none.
const ownFrameIds = options.appearancePass && !options.appearancePass.capture.includes(colorSchemes[0]!)
? selectedIds.flatMap(id => options.appearancePass!.capture.filter(scheme => scheme !== colorSchemes[0]).map(scheme => schemeScreenId(id, scheme)))
: selectedIds;
const incomplete = captureFailure !== undefined || ownFrameIds.some(id => outcomes[id]?.status !== "captured" || outcomes[id]?.landing !== "verified")
|| map.gaps.some(gap => gap.status === "open") || sessionFailures.length > 0;
map.run.status = incomplete || map.screens.length === 0 ? "partial" : "complete";
if (routes.length === 0) map.run.stopReason = "No source targets require capture.";
else if (captureFailure) map.run.stopReason = "Agent exploration did not finish; see capture diagnostics and bundle outcomes.";
else if (sessionFailures.length) map.run.stopReason = sessionFailures.map(failure => `The ${failure.scheme} pass failed (${failure.error.code}): ${failure.error.message.replace(/\.+$/, "")}.`).join(" ");
else if (incomplete) map.run.stopReason = "Some source targets could not be captured; see bundle outcomes.";
map.run.completedAt = new Date((options.now ?? Date.now)()).toISOString();
await store.write(map);
@@ -395,11 +435,14 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
...(Object.keys(launchRecords).length ? { launch: launchRecords } : {}),
...(options.agent ? { captureContexts: Object.fromEntries(selectedIds.filter(id => outcomes[id]?.status === "captured" && options.agent!.contexts[id])
.map(id => [id, options.agent!.contexts[id]!])), agentAcceptance: { policy: AGENT_ACCEPTANCE_POLICY,
frames: Object.fromEntries(Object.keys(images).map(id => [id, acceptedFrames[id]!])) } } : {}) };
frames: Object.fromEntries(Object.keys(images).map(id => [id, acceptedFrames[id]!])) } } : {}),
...(sessionFailures.length ? { sessionFailures } : {}) };
if (options.agent) assertAgentAcceptance(bundle, map, { required: true });
// The side that produces the evidence checks it the way the review will: an integrity defect fails here, naming
// its codes, instead of on the consuming side after both captures. Unavailable screens are not defects.
assertNoCaptureHealthIssues(partitionCaptureHealthIssues(captureHealthIssues(bundle, selectedIds)).integrity);
// A session capturing only a secondary appearance has no primary frames by design; the merged export is checked instead.
const partialPass = options.appearancePass !== undefined && !options.appearancePass.capture.includes(colorSchemes[0]!);
if (!partialPass) assertNoCaptureHealthIssues(partitionCaptureHealthIssues(captureHealthIssues(bundle, selectedIds)).integrity);
await writeFile(join(outputDirectory, "bundle.json"), `${JSON.stringify(bundle, null, 2)}\n`, { flag: "wx" });
if (metadata) await writeFile(join(outputDirectory, "execution.json"), `${JSON.stringify(metadata, null, 2)}\n`, { flag: "wx" });
assertCheckout();
@@ -415,17 +458,28 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
say(`reading EAS build profile ${options.easProfile}`);
const eas = (args: string[], timeoutMs = 5 * 60_000, cwd = project) => runner({ command: options.easCommand, args,
cwd, environment: easEnvironment, timeoutMs });
// `eas simulator:control` names the session by id, from the session's own project directory.
const sessionEas: EasCommand = (args, commandOptions) => runner({ command: options.easCommand, args, cwd: simulatorProject, environment: easEnvironment,
timeoutMs: commandOptions?.timeoutMs ?? 5 * 60_000, ...(commandOptions?.allowFailure ? { allowFailure: true } : {}) });
// A failed simulator:control action is sent again after a pause, which a test's clock replaces.
const controlOptions = { sleep: options.sleep };
// What the EAS session's host can do through simulator:control (iOS sessions on serve-sim 0.10.0 or later).
let controlActions = new Set<string>();
const configResult = eas(["config", "--platform", options.platform, "--profile", options.easProfile, "--json", "--non-interactive"]);
const profile = repackProfileFrom(configResult.stdout);
const configValue = parseJson(configResult.stdout, "eas config") as { appConfig: Record<string, unknown>; buildProfile: Record<string, unknown> };
scheme ??= appScheme(configValue.appConfig.scheme);
colorSchemes = resolveColorSchemes(options.appearance ?? "auto", configValue.appConfig, options.platform);
// Argent cannot switch the appearance itself; on a local device the host switches it (simctl, adb), so it is kept.
if (controller === "argent" && !local) {
const single = argentColorSchemes(colorSchemes, configValue.appConfig, options.platform);
if (single.skipped.length) say(`Argent cannot switch appearance yet (software-mansion/argent#1509); capturing ${single.schemes[0]} only, skipped ${single.skipped.join(", ")}`);
colorSchemes = single.schemes;
}
// Argent cannot switch the appearance itself. On a local device the host switches it (simctl, adb); on an EAS iOS
// session serve-sim does (appearance.set), which is known once the session starts; otherwise one appearance is captured.
let argentSchemes = controller === "argent" && !local ? argentColorSchemes(colorSchemes, configValue.appConfig, options.platform) : undefined;
const narrowToArgentScheme = () => {
if (!argentSchemes) return;
if (argentSchemes.skipped.length) say(`Argent cannot switch appearance on this session; capturing ${argentSchemes.schemes[0]} only, skipped ${argentSchemes.skipped.join(", ")}`);
colorSchemes = argentSchemes.schemes;
argentSchemes = undefined;
};
if (options.platform !== "ios") narrowToArgentScheme();
app = { name: profile.projectIdentity.name, ...(profile.projectIdentity.version ? { version: profile.projectIdentity.version } : {}),
...(profile.projectIdentity.buildNumber ? { buildNumber: profile.projectIdentity.buildNumber } : {}) };
projectId = profile.projectIdentity.projectId;
@@ -449,12 +503,15 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
let localOsVersion: string | undefined;
// Argent's tool-server on this machine, when Argent drives a local device.
let localArgent: LocalArgentServer | undefined;
let argentBridge: ArgentBridgeRunner | undefined;
const localDevice = () => localSimulator ?? localEmulator;
// A local device starts booting now and finishes while the app is built and repacked; the session waits for it.
let localEmulatorStart: Awaited<ReturnType<typeof startLocalEmulator>> | undefined;
const localName = `Explore ${options.device} (${side} ${options.gitCommit.slice(0, 8)})`;
// An EAS session's start, once begun (startSession): it resolves with the start command's failure, if any.
let sessionStart: Promise<unknown> | undefined;
// When the current session was asked for, so a session whose app took longer than its allowance can be replaced.
let sessionStartedAt: number | undefined;
const stopEarlyDevice = async () => {
// An EAS session started beside a repack that then failed: stop the exact session its handoff names.
if (sessionStart) {
@@ -468,6 +525,10 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
await rm(simulatorProject, { recursive: true, force: true });
await localRecord?.close();
};
const earlyCleanupFailure = (original: unknown, cleanup: unknown) => new CiError({ code: "SESSION_CLEANUP_FAILED", stage: "cleanup",
message: "A simulator Explore started before the app was ready could not be confirmed stopped.",
nextAction: "Stop the EAS Simulator session or local device this capture started (eas simulator:list), and verify its status.",
details: { originalFailure: serializeCiError(original, { secrets: [options.expoToken] }) }, cause: cleanup });
// What this side starts on this machine, so a later review cleans up after it if this process is stopped hard.
const localRecord = local ? await openLocalCaptureRecord(sourceEnvironment, simulatorProject) : undefined;
if (local) try {
@@ -475,13 +536,7 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
say(`starting a local ${options.device} emulator`);
localEmulatorStart = await startLocalEmulator({ runner, device: options.device, name: `explore-${side}-${options.gitCommit.slice(0, 8)}`, directory: simulatorProject,
cwd: simulatorProject, environment: restrictedEnvironment,
// agent-device refuses a serial an interrupted run still claims, even one it reports as stale, so that port is skipped.
...(controller === "agent-device" ? { claimed: (serial: string) => {
const status = runner({ command: options.agentDeviceCommand, args: ["device", "status", "--platform", "android", "--serial", serial, "--json"],
cwd: simulatorProject, environment: restrictedEnvironment, timeoutMs: 30_000, allowFailure: true });
const data = (parseJsonOrUndefined(status.stdout) as { data?: { claims?: unknown[]; hiddenStaleClaims?: number } } | undefined)?.data;
return (data?.claims?.length ?? 0) + (data?.hiddenStaleClaims ?? 0) > 0;
} } : {}),
...(controller === "agent-device" ? { claimed: (serial: string) => agentDeviceClaims({ runner, agentDevice: options.agentDeviceCommand, serial, cwd: simulatorProject, environment: restrictedEnvironment }) } : {}),
// Recorded as soon as it starts, so an interrupted review stops it too.
started: async serial => {
localEmulator = serial;
@@ -499,21 +554,26 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
} });
localOsVersion = created.osVersion;
}
} catch (error) { await stopEarlyDevice().catch(() => undefined); throw error; }
} catch (error) { await stopEarlyDevice().catch(cleanup => { throw earlyCleanupFailure(error, cleanup); }); throw error; }
// The session is named before its app exists: an EAS session starts while the app is repacked (startSession).
const sessionName = simulatorSessionName({ mapName: options.name, mapId: options.buildId, gitCommit: options.gitCommit, easBuildId: local ? "local" : "installed",
runIdentity: JSON.stringify({ mapDirectory, outputDirectory, run: sourceEnvironment.GITHUB_RUN_ID, attempt: sourceEnvironment.GITHUB_RUN_ATTEMPT }) });
// A session EAS failed to start is replaced once (see the session block), under a name of its own.
const sessionNameFor = (retry: number) => simulatorSessionName({ mapName: options.name, mapId: options.buildId, gitCommit: options.gitCommit, easBuildId: local ? "local" : "installed",
runIdentity: JSON.stringify({ mapDirectory, outputDirectory, run: sourceEnvironment.GITHUB_RUN_ID, attempt: sourceEnvironment.GITHUB_RUN_ATTEMPT, ...(retry ? { sessionRetry: retry } : {}) }) });
let sessionRetries = 0;
let sessionName = sessionNameFor(0);
/**
* Starts the EAS Simulator session in the background, empty, once the build to repack is known (not while a native
* build runs, which can take most of an hour of session time). The app is installed through the controller when both
* are ready. Resolves with the start command's failure, if any; the session block verifies what EAS handed over.
*/
const startSession = () => {
const startSession = (retry?: { timeoutMs: number }) => {
if (local || sessionStart) return;
say(`starting an EAS Simulator session (${options.device}) while the app is repacked`);
if (!retry) say(`starting an EAS Simulator session (${options.device}) while the app is repacked`);
sessionStartedAt = (options.now ?? Date.now)();
const request = (minutes: number) => ({ command: options.easCommand, args: ["simulator:start", "--name", sessionName, "--platform", options.platform, "--device", options.device,
"--type", controller, "--package-version", controllerVersion(controller), "--public-link",
"--max-duration-minutes", String(minutes), "--out-config-type", "dotenv", "--non-interactive"], cwd: simulatorProject, environment: easEnvironment, timeoutMs: 20 * 60_000 });
"--max-duration-minutes", String(minutes), "--out-config-type", "dotenv", "--non-interactive"], cwd: simulatorProject, environment: easEnvironment,
timeoutMs: retry?.timeoutMs ?? SESSION_START_TIMEOUT_MS });
// An injected runner (tests) answers synchronously; a real start runs beside the repack.
const start = async (minutes: number) => options.runner ? options.runner(request(minutes)) : runAgentControllerCommand(request(minutes), options.signal ?? new AbortController().signal);
sessionStart = writeFile(dotenvPath, SIMULATOR_DOTENV_HEADER, { flag: "wx", mode: 0o600 })
@@ -533,19 +593,25 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
const appDirectory = !options.buildShare ? await realpath(await mkdtemp(join(tmpdir(), "expo-explore-app-"))) : undefined;
let localApp: string | undefined;
type BuiltApp = { fingerprint: string; nativeEasBuildId: string | null; easBuildId: string | null; app?: string; reused: boolean; version?: string; buildNumber?: string };
const buildApp = async (keepApp: string): Promise<BuiltApp> => {
// What a hosted repack waits with and reports through: one deadline from the moment the session starts beside it, for
// the wait for another job's repack and every repack of its own. The repack is shared with this run's other matrix
// jobs only when it repacks a build EAS already had: a native build this job made is its own, so nobody else's key matches it.
const repackWaits = (shared: boolean) => ({ ...(shared ? { share: options.repackShare } : {}), progress: say, now: options.now, sleep: options.sleep,
signal: options.signal, deadline: (options.now ?? Date.now)() + HOSTED_REPACK_BUDGET_MS });
// `repacking` tells the captures sharing this build that their sessions can start too (sharedBuild).
const buildApp = async (keepApp: string, repacking?: () => Promise<void>): Promise<BuiltApp> => {
options.signal?.throwIfAborted();
say("computing the native fingerprint");
const fingerprintResult = eas(["fingerprint:generate", "--platform", options.platform, "--build-profile", options.easProfile, "--json", "--non-interactive"]);
const sourceFingerprint = fingerprintFrom(fingerprintResult.stdout);
const reusable = options.forceFreshBuild || hasEasEnvironmentFetchFailure(configResult.stderr) || hasEasEnvironmentFetchFailure(fingerprintResult.stderr)
? undefined : existingBuild(eas(["build:list", "--platform", options.platform, "--status", "finished", "--build-profile", options.easProfile,
"--fingerprint-hash", sourceFingerprint, ...(options.platform === "ios" ? ["--simulator"] : []), "--limit", "1", "--json", "--non-interactive"]).stdout);
"--fingerprint-hash", sourceFingerprint, ...(options.platform === "ios" ? ["--simulator"] : []), "--limit", "20", "--json", "--non-interactive"]).stdout);
if (reusable) {
say(`reusing EAS build ${reusable.id} (fingerprint ${sourceFingerprint.slice(0, 12)}); repacking it with this commit's JavaScript and the route observer`);
startSession();
startSession(); await repacking?.();
const repacked = await prepareExactSourceBuild({ platform: options.platform, runner, easCommand: options.easCommand, repackCommand: options.repackCommand,
project, nativeBuildId: reusable.id, fingerprint: sourceFingerprint, repackProfile: profile, easEnvironment, repackEnvironment: restrictedEnvironment, routes: inventory, repack: options.repack, easProfile: options.easProfile, keepApp });
project, nativeBuildId: reusable.id, fingerprint: sourceFingerprint, repackProfile: profile, easEnvironment, repackEnvironment: restrictedEnvironment, routes: inventory, repack: options.repack, easProfile: options.easProfile, keepApp, ...repackWaits(true) });
if (!repacked) throw new CiError({ code: "CAPTURE_OBSERVER_BUILD_UNSUPPORTED", stage: "instrumentation",
message: "The build environment cannot be reproduced for automatic capture instrumentation.",
nextAction: "Use a simulator build profile whose required bundle environment can be pulled for local repacking; an uninstrumented build cannot verify captures." });
@@ -557,9 +623,9 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
: `no finished simulator build has fingerprint ${sourceFingerprint.slice(0, 12)}; building on EAS and waiting for it to finish`);
const built = nativeBuildFrom(eas(["build", "--platform", options.platform, "--profile", options.easProfile, "--wait", "--json", "--non-interactive"], 45 * 60_000).stdout);
say(`EAS build ${built.id} finished; repacking it with the route observer`);
startSession();
startSession(); await repacking?.();
const repacked = await prepareExactSourceBuild({ platform: options.platform, runner, easCommand: options.easCommand, repackCommand: options.repackCommand,
project, nativeBuildId: built.id, fingerprint: built.fingerprint, repackProfile: profile, easEnvironment, repackEnvironment: restrictedEnvironment, routes: inventory, repack: options.repack, easProfile: options.easProfile, keepApp });
project, nativeBuildId: built.id, fingerprint: built.fingerprint, repackProfile: profile, easEnvironment, repackEnvironment: restrictedEnvironment, routes: inventory, repack: options.repack, easProfile: options.easProfile, keepApp, ...repackWaits(false) });
if (!repacked) throw new CiError({ code: "CAPTURE_OBSERVER_BUILD_UNSUPPORTED", stage: "instrumentation",
message: "The fresh native build cannot be instrumented with the required route observer.",
nextAction: "Use a simulator profile with a reproducible bundle environment so both revisions can be repacked with the same observer." });
@@ -573,17 +639,36 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
? await sharedBuild({ directory: options.buildShare, appName, ...(options.signal ? { signal: options.signal } : {}), build: buildApp, owner: options.outputDirectory,
key: { platform: options.platform, easProfile: options.easProfile, bundleId: options.bundleId, gitCommit: options.gitCommit, repack: options.repack ?? "eas", local,
routes: inventory.map(route => route.id) },
waiting: () => say("using the build another target of this review is preparing for this commit") })
waiting: () => say("using the build another capture of this run is preparing for this commit"),
// The session is named before its app exists, so a waiter's session boots while the other capture repacks.
repacking: () => startSession() })
: await buildApp(join(appDirectory!, appName));
// A capture that used another target's build starts its session now.
// A capture that used another capture's build has its session started above; this starts the rest.
startSession();
// A session started beside an app that took longer to prepare than its allowance has less than the capture's minutes
// left, so it is replaced. One that named no session is left to the session block below, which finds out why.
const clock = options.now ?? Date.now;
if (!local && sessionStart && sessionStartedAt !== undefined && clock() - sessionStartedAt > SESSION_PREPARATION_MINUTES * 60_000) {
await sessionStart;
const handoff = await readSimulatorDotenvHandoff(dotenvPath).catch(() => undefined);
if (handoff?.sessionId) {
say(`the app took ${Math.round((clock() - sessionStartedAt) / 60_000)} minutes to prepare, more than the ${SESSION_PREPARATION_MINUTES} session ${handoff.sessionId} was started with for it; replacing it so the capture keeps its ${options.maxDurationMinutes} minutes`);
assertSimulatorStopped(eas(["simulator:stop", "--id", handoff.sessionId, "--json", "--non-interactive"], 2 * 60_000, simulatorProject).stdout, handoff.sessionId);
await rm(dotenvPath, { force: true });
sessionRetries += 1;
sessionName = sessionNameFor(sessionRetries);
sessionStart = undefined;
startSession({ timeoutMs: SESSION_START_TIMEOUT_MS });
}
}
fingerprint = built.fingerprint; nativeEasBuildId = built.nativeEasBuildId; easBuildId = built.easBuildId; localApp = built.app; reusedEasBuild = built.reused;
if (!app.version && built.version) app.version = built.version;
if (!app.buildNumber && built.buildNumber) app.buildNumber = built.buildNumber;
options.signal?.throwIfAborted();
} catch (error) {
// A build or repack that fails leaves nothing running: a device started for it goes, with the session's files and the app.
await stopEarlyDevice().catch(() => undefined);
// One that does not go is the failure to report: a session left running is never hidden behind the build's error.
await stopEarlyDevice().catch(cleanup => { throw earlyCleanupFailure(error, cleanup); });
if (appDirectory) await rm(appDirectory, { recursive: true, force: true });
throw error;
}
@@ -607,8 +692,79 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
throw new Error("The local device did not start before capture");
} else {
if (!localApp || !sessionStart) throw new Error("An EAS Simulator session needs the instrumented app on this machine");
const startError = await sessionStart;
const handoff = await readSimulatorDotenvHandoff(dotenvPath);
let startError = await sessionStart;
let handoff = await readSimulatorDotenvHandoff(dotenvPath);
// EAS sometimes fails to start a session: it errors while booting, or the start request fails before a session is
// named. After every failed start, whatever it may have left running is accounted for; then, when the failure is
// provably EAS's, a new session replaces it, up to SESSION_START_RETRIES times within SESSION_RETRY_BUDGET_MS, with
// a growing wait. A session Explore cannot prove is its own is never replaced or stopped: that stays a handoff
// failure below, as does a lookup that fails.
const clock = options.now ?? Date.now;
const endedSessions: string[] = [];
let ended: { sessionId: string; status: string } | undefined;
let retryDeadline: number | undefined;
for (;;) {
const easRead = (args: string[]) => {
try { return runner({ command: options.easCommand, args, cwd: project, environment: easEnvironment, timeoutMs: 2 * 60_000, allowFailure: true }).stdout; }
catch { return ""; }
};
let reason: string | undefined;
ended = undefined;
if (handoff.sessionId) {
const status = endedOwnSimulatorSession(easRead(["simulator:get", "--id", handoff.sessionId, "--json", "--non-interactive"]),
{ sessionId: handoff.sessionId, sessionName, project: profile.projectIdentity, type: controller, platform: options.platform });
if (status) {
ended = { sessionId: handoff.sessionId, status };
endedSessions.push(handoff.sessionId);
reason = `EAS ended simulator session ${handoff.sessionId} (${status}) before it was ready`;
}
} else if (startError && transientSessionStartFailure(startError)) {
// The failed request may still have created a session: any by this exact name is stopped, so none is left running.
const live = liveSimulatorSessionsNamed(easRead(["simulator:list", "--name", sessionName, "--status", "new", "--status", "queued", "--status", "starting",
"--status", "in-progress", "--limit", "100", "--json", "--non-interactive"]), sessionName, profile.projectIdentity);
if (live) {
// No session file names these sessions, so one that does not stop is reported as left running, never as a
// capture failure a split side could keep as a partial report.
for (const id of live) {
try { assertSimulatorStopped(eas(["simulator:stop", "--id", id, "--json", "--non-interactive"], 2 * 60_000).stdout, id); }
catch (error) { throw new CiError({ code: "SESSION_CLEANUP_FAILED", stage: "cleanup",
message: `EAS Simulator session ${id}, started by a request that failed, could not be confirmed stopped.`,
nextAction: `Stop EAS Simulator session ${id} and verify its status.`, details: { sessionId: id }, cause: error }); }
}
reason = `EAS could not start the simulator session (${simulatorStartFailure(startError, [options.expoToken]).message})`;
}
}
if (!reason || sessionRetries >= SESSION_START_RETRIES) break;
retryDeadline ??= clock() + SESSION_RETRY_BUDGET_MS;
const delayMs = sessionRetryDelayMs(sessionRetries + 1);
// A new start needs at least a minute of the budget after its wait; one started later could not finish in it.
const startTimeoutMs = Math.min(SESSION_START_TIMEOUT_MS, retryDeadline - clock() - delayMs);
if (startTimeoutMs < 60_000) break;
sessionRetries += 1;
say(`${reason}; starting a new EAS Simulator session in ${Math.round(delayMs / 1000)} s (${sessionRetries}/${SESSION_START_RETRIES})`);
options.signal?.throwIfAborted();
// Asynchronous and cancellable, so targets capturing beside this one keep running and Ctrl+C stops the wait.
if (options.sleep) options.sleep(delayMs);
else await new Promise<void>((resolve, reject) => {
const timer = setTimeout(() => { options.signal?.removeEventListener("abort", abort); resolve(); }, delayMs);
const abort = () => { clearTimeout(timer); reject(options.signal!.reason); };
options.signal?.addEventListener("abort", abort, { once: true });
});
options.signal?.throwIfAborted();
await rm(dotenvPath, { force: true });
sessionName = sessionNameFor(sessionRetries);
sessionStart = undefined;
startSession({ timeoutMs: startTimeoutMs });
startError = await sessionStart;
handoff = await readSimulatorDotenvHandoff(dotenvPath);
}
// Every session EAS ended is its own failure to start, never a question of ownership.
if (ended) throw new CiError({ code: "SIMULATOR_START_FAILED", stage: "session",
message: `EAS could not start a simulator session: ${endedSessions.length === 1 ? `session ${ended.sessionId} ended (${ended.status}) before it was ready`
: `${endedSessions.length} sessions ended before they were ready (the last, ${ended.sessionId}, ${ended.status})`}.`,
nextAction: "Rerun Explore. If sessions keep failing to start, check the EAS Simulator status and the sessions' logs on expo.dev.",
retryable: true, details: { attempts: sessionRetries + 1, endedSessions, ...(startError ? { startFailure: simulatorStartFailure(startError, [options.expoToken]) } : {}) },
...(startError ? { cause: startError } : {}) });
const startSecrets = [options.expoToken, ...(handoff.connection ? [handoff.connection.daemonAuthToken] : [])];
const startFailure = startError ? simulatorStartFailure(startError, startSecrets) : undefined;
if (handoff.sessionId) {
@@ -638,6 +794,28 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
if (!handoff.connection) throw new Error("Simulator handoff is missing its controller connection");
connection = handoff.connection;
await rm(dotenvPath, { force: true });
if (options.platform === "ios") controlActions = simulatorControlActions(sessionEas, sessionId, controlOptions);
if (!controlActions.has("appearance.set")) narrowToArgentScheme();
// A side split over sessions gives this one only some of its appearances (captureSideInSessions). One that cannot
// switch to any of them has nothing to capture, so it stops here, before the app is installed or the device folded:
// in brentvatne/euxy's main baseline run 39487618529, the dark session folded the device for frames it would never
// take, and its failed fold failed the target that the light session had captured.
const pass = options.appearancePass;
if (pass && !colorSchemes.some(scheme => pass.capture.includes(scheme))) {
const skipped = pass.capture.join(", ");
// The export merge needs the primary appearance's frames; a session that cannot take them fails the side.
if (pass.capture.includes(pass.all[0]!)) throw new CiError({ code: "SESSION_APPEARANCE_UNAVAILABLE", stage: "session", retryable: true,
message: `Simulator session ${sessionId} cannot switch to ${skipped}, the primary appearance it was given to capture.`,
nextAction: "Rerun the capture on a new session, or set capture.appearance to the appearance this controller renders." });
say(`session ${sessionId} cannot switch to ${skipped}, the appearance it was given; it stops without capturing, and the report has no ${skipped} frames`);
const error = { code: "SESSION_APPEARANCE_UNAVAILABLE", stage: "session", retryable: true,
message: `Simulator session ${sessionId} could not switch to ${skipped}.`, nextAction: "Rerun the capture on a new session." };
for (const id of selectedIds) for (const scheme of pass.capture) {
outcomes[schemeScreenId(id, scheme)] = { status: "failed", attempts: 0, landing: "unverified", reason: error.message, error };
}
// The finally block below stops the session after the export.
return await finish();
}
}
const controllerConfig = join(simulatorProject, "agent-device.json");
await writeFile(controllerConfig, "{}\n", { flag: "wx", mode: 0o400 });
@@ -660,19 +838,34 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
AGENT_DEVICE_SESSION: CONTROLLER_SESSION, AGENT_DEVICE_SESSION_LOCK: "reject", AGENT_DEVICE_PLATFORM: options.platform };
// Every controller call below goes to the session's controller; for Argent that is the bridge.
const controllerCommand = argent?.command ?? options.agentDeviceCommand;
// The capture's commands keep one bridge on a worker thread (EXPLORE_ARGENT_BRIDGE=process: one process per command, as before).
if (argent && !options.runner && sourceEnvironment.EXPLORE_ARGENT_BRIDGE !== "process") {
argentBridge = startArgentBridgeRunner({ command: argent.command, environment: argent.environment, fallback: defaultRunner, progress: say });
runner = argentBridge.runner;
}
if (!local && localApp) {
// The session started empty while the app was repacked; the controller installs it now.
say(`installing the app on session ${sessionId}`);
runner({ command: controllerCommand, args: ["install", localApp, "--session", CONTROLLER_SESSION, ...deviceArgs({ platform: options.platform, deviceModel: options.device }),
"--platform", options.platform, "--json"], cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 15 * 60_000 });
// Before any link opens, as a local capture does with simctl: the app's own links then open without the prompt.
if (sessionId && scheme) {
const approved = approveSessionUrlSchemes(sessionEas, sessionId, controlActions, options.bundleId, [scheme], controlOptions);
if (approved.length) say(`approved ${approved.map(item => `${item}://`).join(", ")} links on session ${sessionId}`);
}
}
if (options.devicePose) {
// A fold relights the other panel and resizes the window. It comes before the app is first opened or read, so the
// controller's accessibility reader starts on the lit panel: Argent's, started on the outer panel, can stay there.
say(`folding ${options.device} ${options.devicePose}`);
const target = deviceArgs({ platform: options.platform, deviceModel: options.device, ...(localDevice() ? { provider: "local-simulator" as const, controllerDeviceId: localDevice()! } : {}) });
// agent-device reads the hinge back after folding and fails to (locally and on EAS); its display is checked instead.
const easAgentDevice = !!sessionId && controller === "agent-device";
// Argent's bridge sends a fold again, twice, when the simulator-server it needs did not start within Argent's 30 s,
// so a fold can take three of those waits and the panel checks after it: 4 minutes covers that.
const fold = () => runner({ command: controllerCommand, args: ["fold", options.devicePose!, "--session", CONTROLLER_SESSION, ...target, "--platform", options.platform, "--json"],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 2 * 60_000, ...(localSimulator ? { allowFailure: true } : {}) });
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 4 * 60_000, ...(localSimulator || easAgentDevice ? { allowFailure: true } : {}) });
const pause = options.sleep ?? (milliseconds => { Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, milliseconds); });
if (localSimulator && controller === "argent") {
// Argent folds without a session and checks the fold against its own reader; this machine also checks the display.
fold();
@@ -690,6 +883,62 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
...(options.sleep ? { sleep: options.sleep } : {}) })) throw new CiError({ code: "CONTROLLER_FOLD_UNVERIFIED", stage: "session",
message: `${options.device} did not show the ${options.devicePose} display after folding: ${(folded.stderr || folded.stdout).trim().split("\n").at(-1) ?? ""}`.trim(),
nextAction: "Rerun; if it persists, check that this Xcode's simulator supports folding the device." });
} else if (easAgentDevice && sessionId) {
// On EAS, agent-device also folds only inside a session, so the app opens first. The first fold of a fresh session
// can leave the panel it lights black, as it does for Argent: the controller then reads the old panel and a
// relaunch draws nothing. The display is the check, through the controller's own screenshot: in the pose's shape
// and drawn. If it is not, SpringBoard is restarted, the app reopened and the device folded again, once.
// The open pose's shape is taken to be landscape. That is unverified for agent-device on EAS: local agent-device
// runs returned the opened inner panel on its portrait axes (2007×2853), as explore:binary turns it from; the
// landscape shots seen on EAS (2853×2007) were Argent's, taken with its LandscapeRight rotation.
const open = (relaunch: boolean) => runner({ command: controllerCommand, args: ["open", options.bundleId, ...(relaunch ? ["--relaunch"] : []), "--session", CONTROLLER_SESSION,
...target, "--platform", options.platform, "--json"], cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 15 * 60_000 });
const shown = async () => {
if (options.devicePose !== "open" && options.devicePose !== "closed") return true;
const file = join(simulatorProject, "pose.png");
// The display relights and resizes over a moment after the hinge moves.
for (let attempt = 0; attempt < 10; attempt += 1) {
if (attempt) pause(1_000);
runner({ command: controllerCommand, args: ["screenshot", file, "--session", CONTROLLER_SESSION, ...target, "--platform", options.platform, "--json"],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 2 * 60_000, allowFailure: true });
const png = await readFile(file).catch(() => Buffer.alloc(0));
await rm(file, { force: true });
const size = pngSize(png);
if (size && (options.devicePose === "open" ? size.width > size.height : size.height > size.width) && !solidColour(png)) return true;
}
return false;
};
open(false);
fold();
if (!await shown()) {
if (!controlActions.has("springboard.restart")) throw new CiError({ code: "CONTROLLER_FOLD_UNVERIFIED", stage: "session",
message: `${options.device} did not show the ${options.devicePose} display after folding.`, nextAction: "Rerun the capture on a new session." });
say(`${options.device} did not draw the ${options.devicePose} display after the fold; restarting SpringBoard and folding ${options.devicePose} again`);
simulatorControl(sessionEas, sessionId, "springboard.restart", [], controlOptions);
pause(8_000);
open(true);
fold();
if (!await shown()) throw new CiError({ code: "CONTROLLER_READER_STUCK", stage: "session",
message: `${options.device} did not draw the ${options.devicePose} display after folding, nor after SpringBoard was restarted and it was folded again.`,
nextAction: "Rerun the capture; if it keeps happening, report it to Explore with the session's link." });
}
} else if (controller === "argent" && sessionId && controlActions.has("springboard.restart")) {
// The bridge has already reset Argent's device helpers once when its reader stayed on the dark panel. On EAS the
// tool-server cannot be restarted, so the next step is SpringBoard, then the same fold and check again.
try { fold(); }
catch (error) {
if (toCiError(error).code !== "CONTROLLER_FOLD_UNVERIFIED") throw error;
say(`Argent kept reading the other panel after the fold; restarting SpringBoard and folding ${options.devicePose} again`);
simulatorControl(sessionEas, sessionId, "springboard.restart", [], controlOptions);
pause(8_000);
try { fold(); }
catch (retry) {
if (toCiError(retry).code !== "CONTROLLER_FOLD_UNVERIFIED") throw retry;
throw new CiError({ code: "CONTROLLER_READER_STUCK", stage: "session",
message: `Argent's accessibility reader kept reading the panel the fold turned off, after its device helpers were reset and SpringBoard was restarted.`,
nextAction: "Rerun the capture; if it keeps happening, report it to Explore: this session's Argent tool-server needs a restart that EAS cannot do yet.", cause: retry });
}
}
} else fold();
}
runtimeStage = "runtime";
@@ -700,9 +949,28 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
// (APP_NOT_READY). On this machine the capture runs that reader, so it recovers in two steps, each followed by one more
// open: a fresh tool-server, whose reader starts on the lit panel; then, as XcodeBuildMCP#537 found for this panel,
// SpringBoard restarted as well. The bridge's cached view of the device goes with the old server.
// On EAS the tool-server is the session's and must never be stopped, so the recovery there resets Argent's device
// helpers (reset-reader), then restarts SpringBoard as well, and then gives up with a reason rather than retrying.
const easArgentRecovery = !local && controller === "argent" && !!sessionId && options.platform === "ios";
for (let recovery = 0; ; recovery += 1) try { runtime = collect(); break; }
catch (error) {
if (!localArgent || recovery === 2 || toCiError(error).code !== "APP_NOT_READY") throw error;
if (toCiError(error).code !== "APP_NOT_READY") throw error;
if (easArgentRecovery && sessionId) {
if (recovery === 2 || recovery === 1 && !controlActions.has("springboard.restart")) throw new CiError({ code: "CONTROLLER_READER_STUCK", stage: "session",
message: "Argent's accessibility reader kept reading the panel the fold turned off, after its device helpers were reset"
+ `${recovery === 2 ? " and SpringBoard was restarted" : ""}.`,
nextAction: "Rerun the capture; if it keeps happening, report it to Explore: this session's Argent tool-server needs a restart that EAS cannot do yet.", cause: error });
if (recovery === 1) {
say("Argent still did not read the app; restarting SpringBoard, then resetting Argent's device helpers and opening the app again");
simulatorControl(sessionEas, sessionId, "springboard.restart", [], controlOptions);
// SpringBoard takes a moment to come back, and sends the app to the background; the next open brings it back.
(options.sleep ?? (milliseconds => { Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, milliseconds); }))(8_000);
} else say("Argent did not read the app after starting it; resetting Argent's device helpers and opening the app again");
runner({ command: controllerCommand, args: ["reset-reader", "--session", CONTROLLER_SESSION, "--platform", options.platform, "--json"],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 2 * 60_000 });
continue;
}
if (!localArgent || recovery === 2) throw error;
say(recovery === 0 ? "Argent did not read the app after starting it; restarting Argent's tool-server and opening the app again"
: "Argent still did not read the app; restarting SpringBoard and Argent's tool-server and opening the app again");
stopLocalArgentServer(localArgent);
@@ -738,45 +1006,79 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
// Every pass sets the appearance explicitly, so a capture does not inherit whatever the simulator last used.
const passes: CaptureBundle["outcomes"][] = [];
const recordingTools = launchSetting === "off" ? undefined : options.recordingTools === undefined ? ffmpegRecordingTools() : options.recordingTools ?? undefined;
const capturedSchemes = options.appearancePass ? colorSchemes.filter(scheme => options.appearancePass!.capture.includes(scheme)) : colorSchemes;
const activeRuntime = runtime;
for (const [index, colorScheme] of colorSchemes.entries()) {
// Argent cannot switch the appearance yet (software-mansion/argent#1509): on EAS its single pass runs as the device
// starts, and on this machine the host switches it.
if (controller !== "argent") runner({ command: controllerCommand, args: [...appearanceArgs(colorScheme), "--session", CONTROLLER_SESSION, ...deviceArgs(runtime), "--platform", options.platform],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 60_000 });
else if (localSimulator) setLocalSimulatorAppearance({ runner, udid: localSimulator, scheme: colorScheme, cwd: simulatorProject, environment: restrictedEnvironment });
else if (localEmulator) setLocalEmulatorAppearance({ runner, serial: localEmulator, scheme: colorScheme, cwd: simulatorProject, environment: restrictedEnvironment });
// A scheme change recreates the Android activity in place, and the sheets it re-presents lose their inset; a
// relaunch starts the app in the new scheme, the same way the first pass did.
if (launchSetting !== "off") {
// Observing the launch relaunches the app before setup runs, so no UI-created state is lost.
say(`observing the app launch${colorSchemes.length > 1 ? ` (${colorScheme})` : ""}`);
try {
// A restored baseline's launch evidence is replaced only by a fresh observation.
const record = await captureLaunch({ bundleId: options.bundleId, keepVideo: launchSetting === "video", platform: options.platform,
directory: join(launchWorkDirectory, launchDirectory(colorScheme)), relativeDirectory: launchDirectory(colorScheme), ...(options.now ? { now: options.now } : {}),
...(recordingTools ? { recordingTools } : {}),
controller: (args, controllerOptions) => runner({ command: controllerCommand, args: [...args, "--session", CONTROLLER_SESSION, ...deviceArgs(runtime!), "--platform", options.platform],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: controllerOptions?.timeoutMs ?? 60_000, ...(controllerOptions?.allowFailure ? { allowFailure: true } : {}) }) });
launch[colorScheme] = { root: launchWorkDirectory, record };
} catch (error) {
// The launch is supplementary evidence: a failure to observe it leaves the route capture intact.
say(`the app launch could not be observed (${(error as Error).message}); the report has no launch evidence for ${colorScheme}`);
await rm(join(launchWorkDirectory, launchDirectory(colorScheme)), { recursive: true, force: true });
}
} else if (index > 0) runner({ command: controllerCommand, args: ["open", options.bundleId, "--relaunch", "--session", CONTROLLER_SESSION,
...deviceArgs(runtime), "--platform", options.platform, "--level", "full", "--json"], cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 60_000 });
runtime.colorScheme = colorScheme;
if (colorSchemes.length > 1) say(`appearance ${colorScheme} (${index + 1}/${colorSchemes.length})`);
passes.push(await capturePlannedRoutes({ capture: { ...options, agentDeviceCommand: controllerCommand, setup, progress: say }, routes: missingRoutes, inventory, scheme, appName: app.name, runtime,
connection, environment: controllerEnvironment, store, map, acceptedFrames, outcomes: index === 0 ? outcomes : {},
...(colorSchemes.length > 1 ? { appearance: { scheme: colorScheme, primary: index === 0 } } : {}) }));
// A pass another session makes leaves its slot empty, so the passes stay aligned with the appearances they merge by.
if (!capturedSchemes.includes(colorScheme)) { passes.push({}); continue; }
const passOutcomes: CaptureBundle["outcomes"] = index === 0 ? outcomes : {};
const capturePass = async () => {
// Argent cannot switch the appearance yet (software-mansion/argent#1509): the host does, simctl or adb on this
// machine and serve-sim on an EAS iOS session; elsewhere its single pass runs as the device starts.
if (controller !== "argent") runner({ command: controllerCommand, args: [...appearanceArgs(colorScheme), "--session", CONTROLLER_SESSION, ...deviceArgs(activeRuntime), "--platform", options.platform],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 60_000 });
else if (sessionId && controlActions.has("appearance.set")) simulatorControl(sessionEas, sessionId, "appearance.set", [colorScheme], controlOptions);
else if (localSimulator) setLocalSimulatorAppearance({ runner, udid: localSimulator, scheme: colorScheme, cwd: simulatorProject, environment: restrictedEnvironment });
else if (localEmulator) setLocalEmulatorAppearance({ runner, serial: localEmulator, scheme: colorScheme, cwd: simulatorProject, environment: restrictedEnvironment });
// A scheme change recreates the Android activity in place, and the sheets it re-presents lose their inset; a
// relaunch starts the app in the new scheme, the same way the first pass did.
if (launchSetting !== "off") {
// Observing the launch relaunches the app before setup runs, so no UI-created state is lost.
say(`observing the app launch${colorSchemes.length > 1 ? ` (${colorScheme})` : ""}`);
try {
// A restored baseline's launch evidence is replaced only by a fresh observation.
const record = await captureLaunch({ bundleId: options.bundleId, keepVideo: launchSetting === "video", platform: options.platform,
directory: join(launchWorkDirectory, launchDirectory(colorScheme)), relativeDirectory: launchDirectory(colorScheme), ...(options.now ? { now: options.now } : {}),
...(recordingTools ? { recordingTools } : {}),
controller: (args, controllerOptions) => runner({ command: controllerCommand, args: [...args, "--session", CONTROLLER_SESSION, ...deviceArgs(activeRuntime), "--platform", options.platform],
cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: controllerOptions?.timeoutMs ?? 60_000, ...(controllerOptions?.allowFailure ? { allowFailure: true } : {}) }) });
launch[colorScheme] = { root: launchWorkDirectory, record };
} catch (error) {
// The launch is supplementary evidence: a failure to observe it leaves the route capture intact.
say(`the app launch could not be observed (${(error as Error).message}); the report has no launch evidence for ${colorScheme}`);
await rm(join(launchWorkDirectory, launchDirectory(colorScheme)), { recursive: true, force: true });
}
} else if (index > 0) runner({ command: controllerCommand, args: ["open", options.bundleId, "--relaunch", "--session", CONTROLLER_SESSION,
...deviceArgs(activeRuntime), "--platform", options.platform, "--level", "full", "--json"], cwd: simulatorProject, environment: controllerEnvironment, timeoutMs: 60_000 });
activeRuntime.colorScheme = colorScheme;
if (colorSchemes.length > 1) say(`appearance ${colorScheme} (${index + 1}/${colorSchemes.length})`);
return await capturePlannedRoutes({ capture: { ...options, ...(argentBridge && !options.agent ? { runner: argentBridge.runner } : {}), agentDeviceCommand: controllerCommand, setup, progress: say }, routes: missingRoutes, inventory, scheme, appName: app.name, runtime: activeRuntime,
connection, environment: controllerEnvironment, store, map, acceptedFrames, outcomes: passOutcomes,
...(colorSchemes.length > 1 ? { appearance: { scheme: colorScheme, primary: index === 0 } } : {}) });
};
// A later pass of a session whose primary pass captured fails only its own appearance, as a secondary session does
// when the passes are split over sessions (captureSideInSessions): in brentvatne/euxy's main baseline run
// 39506374824 a failed dark session failed a target whose light session had captured 16 of 16 routes.
if (index === 0 || !capturedSchemes.includes(colorSchemes[0]!) || options.agent) { passes.push(await capturePass()); continue; }
try { passes.push(await capturePass()); }
catch (error) {
const failure = toCiError(error, { code: `CAPTURE_${runtimeStage.toUpperCase()}_FAILED`, stage: runtimeStage,
nextAction: "Inspect the capture diagnostic, correct the failing prerequisite, and rerun Explore." });
if (options.signal?.aborted || capturedEvidenceInDoubt(failure.code)) throw failure;
const { code, stage, message, nextAction, retryable } = serializeCiError(failure, { secrets: [options.expoToken, ...environmentSecrets(sourceEnvironment)] });
say(`the ${colorScheme} pass failed (${code}: ${message}); keeping the ${colorSchemes[0]} captures, and the report marks every ${colorScheme} frame it did not take failed`);
sessionFailures.push({ scheme: colorScheme, error: { code, stage, message, nextAction, retryable } });
// The routes the pass finished keep what it recorded.
passes.push(passOutcomes);
}
}
runtime.colorScheme = colorSchemes[0]!;
Object.assign(outcomes, mergeSchemeOutcomes(passes, colorSchemes));
// A session without the primary pass keeps each outcome as its pass recorded it, under `<id>@<scheme>`: the export
// merge recombines them with the primary pass's (mergeSchemeOutcomes), which only then can tell an unattributed
// frame beside a primary one from a lost route.
if (options.appearancePass && !capturedSchemes.includes(colorSchemes[0]!)) {
for (const [index, pass] of passes.entries()) for (const [id, outcome] of Object.entries(pass)) outcomes[schemeScreenId(id, colorSchemes[index])] = outcome;
} else Object.assign(outcomes, mergeSchemeOutcomes(passes, colorSchemes));
// A failed pass records its failure under each route it did not finish, as the export merge does for a failed session.
for (const { scheme: failedScheme, error } of sessionFailures) {
const pass = passes[colorSchemes.indexOf(failedScheme)] ?? {};
for (const id of selectedIds) if (!pass[id]) outcomes[schemeScreenId(id, failedScheme)] = { status: "failed", attempts: 0, landing: "unverified", reason: error.message, error };
}
} catch (error) { primaryError = toCiError(error, { code: `CAPTURE_${runtimeStage.toUpperCase()}_FAILED`, stage: runtimeStage,
nextAction: "Inspect the capture diagnostic, correct the failing prerequisite, and rerun Explore." }); }
finally {
if (appDirectory) await rm(appDirectory, { recursive: true, force: true });
argentBridge?.close();
if (localArgent) stopLocalArgentServer(localArgent);
if (localEmulator) {
try {
@@ -840,7 +1142,10 @@ async function executeCiMap(options: CiMapOptions, setStage: (stage: string) =>
throw primaryError;
}
setStage("export");
const captured = selectedIds.filter(id => outcomes[id]?.status === "captured").length;
say(`${captured} of ${selectedIds.length} ${selectedIds.length === 1 ? "route" : "routes"} captured`);
const ownSchemes = options.appearancePass && !options.appearancePass.capture.includes(colorSchemes[0]!)
? colorSchemes.filter(scheme => options.appearancePass!.capture.includes(scheme)) : undefined;
const captured = selectedIds.filter(id => ownSchemes
? ownSchemes.every(scheme => outcomes[schemeScreenId(id, scheme)]?.status === "captured") : outcomes[id]?.status === "captured").length;
say(`${captured} of ${selectedIds.length} ${selectedIds.length === 1 ? "route" : "routes"} captured${ownSchemes ? ` in ${ownSchemes.join(", ")}` : ""}`);
return finish();
}
+17 -5
View File
@@ -6,6 +6,8 @@ export type CaptureSetupSnapshot = {
nodes: readonly Record<string, unknown>[];
refsGeneration?: number;
truncated?: boolean;
/** Whether the controller says a software keyboard shows, when it says (Argent does). */
keyboardVisible?: boolean;
};
type Awaitable<T> = T | Promise<T>;
@@ -98,13 +100,24 @@ function failure(code: string, message: string, details?: Record<string, unknown
...(details ? { details } : {}) });
}
export const SETUP_SELECTOR_MAX = 256;
/** A label or identifier the executor can select by; a replay plan's selectors are checked against it before any device starts. */
export function isSetupSelector(value: unknown): value is string {
return typeof value === "string" && value.trim().length > 0 && value.length <= SETUP_SELECTOR_MAX && !/[\u0000-\u001f\u007f]/.test(value);
}
/** Text the executor can fill or type: one line of at most 4,096 characters. */
export function isSetupInput(value: unknown): value is string {
return typeof value === "string" && value.length <= 4_096 && !/[\u0000\r\n]/.test(value);
}
function selectorValue(selector: CaptureSetupSelector): { key: "label" | "identifier"; value: string } {
const hasLabel = typeof selector.label === "string";
const hasIdentifier = typeof selector.identifier === "string";
const key = hasLabel ? "label" : "identifier";
const value = selector[key];
if (hasLabel === hasIdentifier || typeof value !== "string" || !value.trim()
|| value.length > 256 || /[\u0000-\u001f\u007f]/.test(value)) {
if (hasLabel === hasIdentifier || !isSetupSelector(value)) {
throw failure("SETUP_SELECTOR_INVALID", "A UI selector must contain exactly one nonempty label or identifier.");
}
return { key, value };
@@ -274,8 +287,7 @@ export async function executeCaptureSetup(actions: readonly CaptureSetupAction[]
case "tap":
case "fill": {
selectorValue(action);
if (action.action === "fill" && (typeof action.text !== "string" || action.text.length > 4_096
|| /[\u0000\r\n]/.test(action.text))) {
if (action.action === "fill" && !isSetupInput(action.text)) {
throw failure("SETUP_INPUT_INVALID", "A UI fill input must be a single line of at most 4,096 characters.");
}
let resolved = false;
@@ -330,7 +342,7 @@ export async function executeCaptureSetup(actions: readonly CaptureSetupAction[]
break;
}
case "type": {
if (typeof action.text !== "string" || !action.text || action.text.length > 4_096 || /[\u0000\r\n]/.test(action.text)) {
if (!isSetupInput(action.text) || !action.text) {
throw failure("SETUP_INPUT_INVALID", "A UI type input must be a nonempty single line of at most 4,096 characters.");
}
let ready = false;
+8 -4
View File
@@ -19,21 +19,23 @@ const POLL_MS = 2_000;
/**
* Runs `build` once per key across every capture sharing `directory`; the app it leaves goes in the build's folder
* (`keepApp`), which outlives each capture. A waiter is told so it can say whose build it uses.
* (`keepApp`), which outlives each capture. A waiter is told so it can say whose build it uses, and told again
* (`repacking`) once the builder calls its `repacking` (the native build is known and is being repacked), or the
* build is done: a waiter starts its session then, not while a native build may run for most of an hour.
*/
export async function sharedBuild<T extends object>(options: { directory: string; key: SharedBuildKey; appName: string; signal?: AbortSignal;
build: (keepApp: string) => Promise<T>; waiting?: () => void;
build: (keepApp: string, repacking: () => Promise<void>) => Promise<T>; waiting?: () => void; repacking?: () => void;
/** Who builds it (the capture's output directory), so a review can fail its waiters if that capture dies (abandonSharedBuilds). */
owner?: string }): Promise<T> {
const folder = join(options.directory, sharedBuildId(options.key));
const result = join(folder, "result.json"), failure = join(folder, "failure.json");
const result = join(folder, "result.json"), failure = join(folder, "failure.json"), repack = join(folder, "repacking");
await mkdir(options.directory, { recursive: true });
let owner = true;
try { await mkdir(folder); } catch (error) { if ((error as NodeJS.ErrnoException).code !== "EEXIST") throw error; owner = false; }
if (owner) {
if (options.owner) await writeFile(join(folder, "owner"), options.owner);
try {
const built = await options.build(join(folder, options.appName));
const built = await options.build(join(folder, options.appName), () => writeFile(repack, "").catch(() => undefined));
await writeFile(`${result}.tmp`, JSON.stringify(built));
await rename(`${result}.tmp`, result);
return built;
@@ -44,9 +46,11 @@ export async function sharedBuild<T extends object>(options: { directory: string
}
}
options.waiting?.();
let repackSeen = false;
for (;;) {
options.signal?.throwIfAborted();
const done = await readFile(result, "utf8").catch(() => undefined);
if (!repackSeen && (done !== undefined || existsSync(repack))) { repackSeen = true; options.repacking?.(); }
if (done !== undefined) return JSON.parse(done) as T;
const failed = await readFile(failure, "utf8").catch(() => undefined);
if (failed !== undefined) {
+111
View File
@@ -0,0 +1,111 @@
import { CiError } from "../shared/ci-errors.js";
import { isRecord } from "./common.js";
/**
* Runs the pinned EAS CLI with these arguments; throws when it exits non-zero unless `allowFailure` is set. A runner that
* knows the exit status reports it as `status`.
*/
export type EasCommand = (args: string[], options?: { timeoutMs?: number; allowFailure?: boolean }) => { stdout: string; stderr?: string; status?: number | null };
/**
* Pauses before each resend of a failed action. Every action Explore runs is idempotent, and the session's host can fail
* one once: in brentvatne/euxy's baseline run 39348933440, `app.approveUrlScheme` failed in its `simctl spawn … defaults
* write`, which failed the dark-appearance session and with it the target's whole baseline.
*/
const SIMULATOR_CONTROL_RETRY_DELAYS_MS = [3_000, 6_000] as const;
/**
* What eas-cli 25.2.0 says when the session's serve-sim predates the action list (it refuses `actions.describe`, or
* answers it without a catalog), and what an EAS CLI without `simulator:control` says. Those sessions offer no actions,
* and asking again would not change that.
*/
const NO_ACTION_LIST = /serve-sim version that does not support this action|did not return its list of actions|command simulator:control not found/;
/**
* The host actions an EAS Simulator iOS session offers through `eas simulator:control` (serve-sim's actions, 0.10.0 and
* later for the ones Explore uses): an empty set when the session's serve-sim or the EAS CLI predates the list, so a
* caller falls back to what it did before them. The session is always named by id, never taken from a
* `.env.eas-simulator` beside it.
*
* These are a fallback: what the session's controller can do itself (agent-device or Argent: fold, appearance, a
* reader reset) goes through the controller. Explore uses an action here only for what neither controller offers —
* approving a link scheme, restarting SpringBoard, and Argent's appearance (software-mansion/argent#1509).
*
* eas-cli prints the list as JSON on stdout only when it succeeds; a failure prints its reason on stderr. A list that
* fails is asked for again after a pause, and one that still fails throws a retryable error rather than reading as "no
* actions": in brentvatne/euxy's main baseline run 39487618529, one failed list made the dark-appearance session look as
* if it offered none, so it dropped dark, approved no link scheme, and skipped the SpringBoard restart its fold needed,
* which failed the target. A session whose host cannot answer three times in a row would fail those actions too, and
* capturing on without them publishes a report that silently lacks an appearance; a new session is the way out.
*/
export function simulatorControlActions(eas: EasCommand, sessionId: string, options: SimulatorControlOptions = {}): Set<string> {
const sleep = options.sleep ?? (milliseconds => { Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, milliseconds); });
const args = ["simulator:control", "--list", "--id", sessionId, "--json", "--non-interactive"];
for (let retry = 0; ; retry += 1) {
let failure: string;
try {
const listed = eas(args, { timeoutMs: 2 * 60_000, allowFailure: true });
const actions = (() => { try { return JSON.parse(listed.stdout) as unknown; } catch { return undefined; } })();
if ((listed.status ?? 0) === 0 && isRecord(actions)) return new Set(Object.keys(actions));
const output = `${listed.stderr ?? ""}\n${listed.stdout}`;
if (NO_ACTION_LIST.test(output)) return new Set();
// eas-cli ends a failure with "simulator:control command failed."; the line before it says why.
failure = output.split("\n").map(line => line.trim()).filter(line => line && !/command failed\.?$/.test(line)).at(-1)?.slice(0, 300) || "it printed no list";
} catch (error) {
if (NO_ACTION_LIST.test(error instanceof Error ? error.message : String(error))) return new Set();
failure = firstLine(error);
}
const delayMs = SIMULATOR_CONTROL_RETRY_DELAYS_MS[retry];
if (delayMs === undefined) throw new CiError({ code: "SIMULATOR_CONTROL_UNAVAILABLE", stage: "session", retryable: true,
message: `EAS could not list the simulator:control actions of session ${sessionId} after ${retry + 1} attempts: ${failure}`,
nextAction: "Rerun the capture on a new session. Without the list, Explore cannot tell whether the session can switch appearance, approve the app's links, or restart SpringBoard.",
details: { sessionId, attempts: retry + 1 } });
// A list that succeeds on a later attempt would otherwise leave no trace of the failure in the job log.
process.stderr.write(`simulator:control --list failed for session ${sessionId} (${failure}); asking again in ${delayMs / 1000} s\n`);
sleep(delayMs);
}
}
/** Lets a test replace the pause between resends of a failed action. */
export type SimulatorControlOptions = { sleep?: ((milliseconds: number) => void) | undefined };
/**
* Runs one `eas simulator:control` action on the session; eas-cli fills in the device. A failed action is sent again
* after a pause; one that still fails throws with EAS's reason.
*/
export function simulatorControl(eas: EasCommand, sessionId: string, action: string, values: readonly string[] = [], options: SimulatorControlOptions = {}): void {
const sleep = options.sleep ?? (milliseconds => { Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0, milliseconds); });
// Values follow `--` so one that starts with a dash is never read as a flag.
const args = ["simulator:control", action, "--id", sessionId, "--json", "--non-interactive", "--", ...values];
for (let retry = 0; ; retry += 1) {
try { eas(args, { timeoutMs: 2 * 60_000 }); return; }
catch (error) {
const delayMs = SIMULATOR_CONTROL_RETRY_DELAYS_MS[retry];
if (delayMs === undefined) throw error;
// A resend that succeeds would otherwise leave no trace of the failure in the job log.
process.stderr.write(`simulator:control ${action} failed (${firstLine(error)}); sending it again in ${delayMs / 1000} s\n`);
sleep(delayMs);
}
}
}
/**
* Lets the app open its own links without iOS asking "Open in “App”?" first, like local captures do with simctl. A
* session without the action keeps the controllers' prompt handling; http and https never prompt and are skipped.
*/
export function approveSessionUrlSchemes(eas: EasCommand, sessionId: string, actions: ReadonlySet<string>, bundleId: string, schemes: readonly string[],
options: SimulatorControlOptions = {}): string[] {
if (!actions.has("app.approveUrlScheme")) return [];
const approved: string[] = [];
for (const scheme of new Set(schemes.map(value => value.toLowerCase()))) {
if (scheme === "http" || scheme === "https" || !/^[a-z][a-z0-9+.-]*$/.test(scheme)) continue;
simulatorControl(eas, sessionId, "app.approveUrlScheme", [bundleId, scheme], options);
approved.push(scheme);
}
return approved;
}
/** The first line of a failure, for a log line. */
function firstLine(error: unknown): string {
return (error instanceof Error ? error.message : String(error)).split("\n", 1)[0]!.slice(0, 300);
}
+41 -11
View File
@@ -237,17 +237,7 @@ export function assertSimulatorSessionIdentity(
) {
throw new Error("eas simulator:get did not prove the exact expected simulator session");
}
// eas-cli 22.4.0 omits the `app` record; its `deviceRunSessionUrl` still binds the session
// to one exact owner and slug, so accept either proof but never neither.
const expectedSessionUrl = `https://expo.dev/accounts/${expectedProject.owner}`
+ `/projects/${expectedProject.slug}/simulator-sessions/${expectedSessionId}`;
const projectProven = isRecord(value.app)
? value.app.id === expectedProject.projectId
&& value.app.slug === expectedProject.slug
&& isRecord(value.app.ownerAccount)
&& value.app.ownerAccount.name === expectedProject.owner
: value.deviceRunSessionUrl === expectedSessionUrl;
if (!projectProven) {
if (!sessionProjectProven(value, expectedSessionId, expectedProject)) {
throw new Error("eas simulator:get did not prove the exact expected simulator session");
}
if (expectedConnection && !remoteConfigMatches(value.remoteConfig, expectedConnection)) {
@@ -255,6 +245,46 @@ export function assertSimulatorSessionIdentity(
}
}
/**
* eas-cli 22.4.0 omits the `app` record; its `deviceRunSessionUrl` still binds the session to one exact owner and
* slug, so either proof is accepted but never neither.
*/
function sessionProjectProven(value: Record<string, unknown>, sessionId: string, project: RepackProfile["projectIdentity"]): boolean {
const expectedSessionUrl = `https://expo.dev/accounts/${project.owner}/projects/${project.slug}/simulator-sessions/${sessionId}`;
return isRecord(value.app)
? value.app.id === project.projectId && value.app.slug === project.slug && isRecord(value.app.ownerAccount) && value.app.ownerAccount.name === project.owner
: value.deviceRunSessionUrl === expectedSessionUrl;
}
/**
* The status (ERRORED or STOPPED) of a session `eas simulator:get` shows EAS ended before Explore could use it, when it
* is provably this capture's own: the exact id, name, type, platform and project. Only such a session is replaced by a
* new one; anything less certain (undefined here) stays a handoff failure.
*/
export function endedOwnSimulatorSession(stdout: string, expected: { sessionId: string; sessionName: string; project: RepackProfile["projectIdentity"];
type: "agent-device" | "argent"; platform: Platform }): "ERRORED" | "STOPPED" | undefined {
let value: unknown;
try { value = JSON.parse(stdout); } catch { return undefined; }
return isRecord(value) && value.id === expected.sessionId && value.name === expected.sessionName && value.type === expected.type
&& value.platform === expected.platform.toUpperCase() && (value.status === "ERRORED" || value.status === "STOPPED")
&& sessionProjectProven(value, expected.sessionId, expected.project) ? value.status : undefined;
}
/**
* This project's sessions with exactly this name that EAS may still be starting or running, from
* `eas simulator:list --json`: their ids when every one is provably this project's, or undefined when the listing
* cannot be read or a session by that name cannot be proven Explore's, so a caller never assumes none exist.
*/
export function liveSimulatorSessionsNamed(stdout: string, sessionName: string, project: RepackProfile["projectIdentity"]): string[] | undefined {
let value: unknown;
try { value = JSON.parse(stdout); } catch { return undefined; }
if (!isRecord(value) || !Array.isArray(value.sessions)) return undefined;
const live = new Set(["NEW", "QUEUED", "STARTING", "IN_PROGRESS"]);
const named = value.sessions.filter((session): session is Record<string, unknown> => isRecord(session) && session.name === sessionName && live.has(String(session.status)));
if (named.some(session => typeof session.id !== "string" || !/^[A-Za-z0-9._-]{1,256}$/.test(session.id) || !sessionProjectProven(session, session.id, project))) return undefined;
return named.map(session => session.id as string);
}
/** The session's published controller endpoint is the one Explore holds: agent-device's daemon, or Argent's tool-server. */
function remoteConfigMatches(remoteConfig: unknown, connection: SimulatorConnection): boolean {
if (!isRecord(remoteConfig)) return false;
+44 -7
View File
@@ -1,4 +1,4 @@
import { routeIdOfScreen } from "../capture/appearance.js";
import { bundleColorSchemes, routeIdOfScreen, schemeScreenId } from "../capture/appearance.js";
import { createHash } from "node:crypto";
import { cp, lstat, mkdtemp, readFile, readdir, realpath, rm, writeFile } from "node:fs/promises";
import { isAbsolute, join, relative, resolve, sep } from "node:path";
@@ -15,7 +15,10 @@ import { ARGENT_VERSION } from "../shared/argent-version.js";
const METADATA = "cache-metadata.json";
const CONTRACT = "explore.capture-cache";
const MAX_BYTES = 250 * 1024 * 1024;
/** The most evidence a capture cache holds; a downloaded baseline is refused past it before it is unpacked. */
export const CAPTURE_CACHE_MAX_BYTES = 250 * 1024 * 1024;
/** The most files a capture cache holds, besides its metadata. */
export const CAPTURE_CACHE_MAX_FILES = 10_000;
type Recipe = { inventory: unknown; profile: string; config: string; toolsCommit: string };
type FileRecord = { path: string; bytes: number; sha256: string };
type CacheMetadata = { contract: typeof CONTRACT; recipe: string; commit: string; files: FileRecord[] };
@@ -28,6 +31,9 @@ const recipeKey = ({ inventory, profile, config, toolsCommit }: Recipe) => {
/** Only settings that can change captured runtime evidence belong in the recipe. */
export function captureRecipeConfig(config: ExploreConfig, instructionsSha256?: string): string {
// How many sessions capture the evidence (capture.maxSessions), and how long a review waits for a baseline, do not change it.
const { maxSessions: _maxSessions, baselineWaitMinutes: _baselineWaitMinutes, ...capture } = config.capture;
void _maxSessions; void _baselineWaitMinutes;
return canonicalJson({
observer: CAPTURE_OBSERVER_HASH,
mode: config.mode,
@@ -36,7 +42,7 @@ export function captureRecipeConfig(config: ExploreConfig, instructionsSha256?:
...(config.app.scheme === undefined ? {} : { scheme: config.app.scheme }) },
device: config.device, parameters: config.parameters, setup: config.setup ?? [],
// The repack as it resolves, so a config that leaves it to the default keeps the recipe it had when the default was written out.
capture: { ...config.capture, repack: captureRepack(config) },
capture: { ...capture, repack: captureRepack(config) },
// agent-device recipes stay as they were; an Argent capture is keyed to the Argent release that drove it.
...(config.capture.controller === "argent" ? { controllerVersion: ARGENT_VERSION } : {}),
});
@@ -48,7 +54,8 @@ function requiresAgentAcceptance(recipe: Recipe, bundle: CaptureBundle): boolean
catch { return false; }
}
async function validateEvidence(directory: string, bundle: CaptureBundle): Promise<void> {
/** Checks an export the way a cache entry is checked before it is written or restored: its provenance, profile and frames. */
export async function validateEvidence(directory: string, bundle: CaptureBundle): Promise<void> {
const map = await readExportedMap(directory, { checkArtifacts: true });
const execution = parseRuntimeExecutionContext(JSON.parse(await readFile(join(directory, "execution.json"), "utf8")));
if (!bundle.runtimeProfile || !/^[a-f0-9]{40}$/.test(bundle.gitCommit)
@@ -101,12 +108,37 @@ async function fileInventory(directory: string, path = ""): Promise<FileRecord[]
else if (info.isFile()) files.push({ path: child, bytes: info.size, sha256: hash(await readFile(join(directory, child))) });
else throw new Error("Capture cache contains an unsupported file");
}
if (files.length > 10_000 || files.reduce((sum, file) => sum + file.bytes, 0) > MAX_BYTES) {
if (files.length > CAPTURE_CACHE_MAX_FILES || files.reduce((sum, file) => sum + file.bytes, 0) > CAPTURE_CACHE_MAX_BYTES) {
throw new Error("Capture cache exceeds the evidence size limit");
}
return files;
}
/**
* The routes a review can reuse from a capture: a verified frame in every appearance it renders. A baseline whose dark
* session failed (brentvatne/euxy's main baseline run 39506374824) holds its light frames and a failed `<route>@dark`;
* reusing the light frame alone would leave the review's base with no dark frame, and dark would read as unavailable
* because of the base. Such a route is captured by the review's base instead, in every appearance.
*/
export function reusableSurfaceIds(bundle: CaptureBundle): string[] {
const secondarySchemes = bundleColorSchemes(bundle).slice(1);
return bundle.selectedSurfaceIds.filter(id => hasVerifiedCapture(bundle, id)
&& secondarySchemes.every(scheme => bundle.images[schemeScreenId(id, scheme)] && bundle.outcomes[schemeScreenId(id, scheme)] === undefined));
}
/**
* Why a capture of `commit` in `sourceDirectory` is not offered for reuse, when it is there and captured some routes:
* none has every appearance (its dark session failed). Undefined otherwise, the cache's usual misses included.
*/
export async function incompleteCaptureReason(sourceDirectory: string, commit: string): Promise<string | undefined> {
try {
const bundle = await readCaptureBundle(sourceDirectory);
if (bundle.gitCommit !== commit || reusableSurfaceIds(bundle).length || !bundle.selectedSurfaceIds.some(id => hasVerifiedCapture(bundle, id))) return undefined;
const failed = (bundle.sessionFailures ?? []).map(failure => failure.scheme);
return failed.length ? `its ${failed.join(" and ")} session failed, so no route has every appearance` : "no route was captured in every appearance";
} catch { return undefined; }
}
/** Cache full exported evidence, including accessibility maps and all referenced screenshots. */
export async function writeCaptureArtifactCache(directory: string, recipe: Recipe): Promise<boolean> {
try {
@@ -116,6 +148,9 @@ export async function writeCaptureArtifactCache(directory: string, recipe: Recip
if (!bundle.runtimeProfile) return false;
assertAgentAcceptance(bundle, undefined, { required: requiresAgentAcceptance(recipe, bundle) });
await validateEvidence(directory, bundle);
// A capture with no route a review could reuse is not offered: a review that found it would wait for it, and capture
// its sides in turn instead of together, only to capture every route of the base itself.
if (!reusableSurfaceIds(bundle).length) return false;
const metadata: CacheMetadata = {
contract: CONTRACT, recipe: recipeKey(recipe), commit: bundle.gitCommit, files: await fileInventory(directory),
};
@@ -175,7 +210,8 @@ export async function restoreCaptureArtifactCache(options: Recipe & {
stage = await mkdtemp(`${options.directory}.cache-`);
const stagedEvidence = join(stage, "evidence");
await cp(source, stagedEvidence, { recursive: true, force: false, errorOnExist: true });
const available = options.selectedIds.filter(id => hasVerifiedCapture(bundle, id)
const reusable = new Set(reusableSurfaceIds(bundle));
const available = options.selectedIds.filter(id => reusable.has(id)
&& (!options.captureContexts || options.captureContexts[id] && bundle.captureContexts?.[id]
&& canonicalJson(options.captureContexts[id]) === canonicalJson(bundle.captureContexts[id])));
if (!available.length) return false;
@@ -186,7 +222,8 @@ export async function restoreCaptureArtifactCache(options: Recipe & {
const captureContexts = bundle.captureContexts ? Object.fromEntries(available.map(id => [id, bundle.captureContexts![id]])) : undefined;
const agentAcceptance = bundle.agentAcceptance ? { policy: bundle.agentAcceptance.policy,
frames: Object.fromEntries(available.map(id => [id, bundle.agentAcceptance!.frames[id]!])) } : undefined;
await writeFile(join(stagedEvidence, "bundle.json"), JSON.stringify({ ...bundle, side: "base", selectedSurfaceIds: available, images, outcomes,
// Only complete routes are restored, so a failed pass of the baseline's is no part of the restored bundle.
await writeFile(join(stagedEvidence, "bundle.json"), JSON.stringify({ ...bundle, side: "base", selectedSurfaceIds: available, images, outcomes, sessionFailures: undefined,
...(captureContexts ? { captureContexts } : {}), ...(agentAcceptance ? { agentAcceptance } : {}) }));
await cp(stagedEvidence, options.directory, { recursive: true, force: false, errorOnExist: true });
return true;
+203
View File
@@ -0,0 +1,203 @@
import { openAsBlob } from "node:fs";
import { mkdir, mkdtemp, open, rm, writeFile } from "node:fs/promises";
import { dirname, join, relative, resolve } from "node:path";
import { setTimeout as delay } from "node:timers/promises";
import { BlobReader, Uint8ArrayWriter, ZipReader, configure } from "@zip.js/zip.js";
import { CiError } from "../shared/ci-errors.js";
import { silentProgress, type Progress } from "../shared/progress.js";
import { retryGitHubLookup } from "../report/github-lookup.js";
import { safeRelativePath } from "../binary/archive.js";
import { CAPTURE_CACHE_MAX_BYTES, CAPTURE_CACHE_MAX_FILES } from "./artifact-cache.js";
import { baselineLookup, githubContext, trustedBaselineRun, type GithubArtifact, type GithubRun } from "./workflow-steps.js";
/**
* A review that waits for main's baseline of its base commit instead of capturing the base itself. Right after a
* merge, the next pull request's review starts while the baseline workflow is still capturing the new base: in
* brentvatne/euxy's review run 39348956443 it had started 15 s before, so every target captured its base too, about
* 10–14 minutes per target instead of 6, on twice the EAS Simulator sessions. A baseline job takes 6–8 minutes and
* head capture 5–8, so `runCiReview` starts head at once and waits here only where base would start.
*/
/** `capture.baselineWaitMinutes` when the config sets none: two baseline jobs' worth, so one queued behind another still counts. */
export const DEFAULT_BASELINE_WAIT_MINUTES = 15;
const POLL_MS = 20_000;
/** GitHub's Actions API lists an artifact a little after its upload (#47); a finished job gets this long for its artifact to appear. */
const ARTIFACT_LISTING_GRACE_MS = 60_000;
/** A zip of the largest capture cache, stored rather than deflated, plus its headers. */
const MAX_ARTIFACT_ZIP_BYTES = CAPTURE_CACHE_MAX_BYTES + 16 * 1024 * 1024;
type GithubJob = { name: string; status?: string; conclusion?: string | null };
export type BaselineWaitOptions = {
/** The baseline run `ci:find-baseline` named; it is checked again here, by the same trust rules. */
run: number; sha: string; target: string;
/** The longest the review waits, from its first poll (`capture.baselineWaitMinutes`). */
timeoutMs: number; pollMs?: number;
environment?: NodeJS.ProcessEnv; transport?: typeof fetch; progress?: Progress; signal?: AbortSignal;
now?: () => number; sleep?: (milliseconds: number, signal?: AbortSignal) => Promise<void>;
};
/** A failed GitHub read that may succeed later: no answer or a 5xx. A bad token or a missing permission is not. */
const transientGitHubFailure = (error: unknown) => error instanceof CiError && error.retryable;
const describe = (error: unknown) => error instanceof CiError ? `${error.code}: ${error.message}` : error instanceof Error ? error.message : String(error);
const statusText = (status: string | undefined) => (status ?? "queued").replaceAll("_", " ");
/**
* Polls one baseline run until this target's baseline artifact is listed, and answers with it. The run must pass the
* trust check a finished baseline passes. Each target's job uploads its own artifact when it finishes, so a target is
* reused even when another target's job fails (#48); this target's job failing or being cancelled ends the wait, as do
* the time limit, a missing permission and GitHub failures that outlast their retries. The reason says which.
*/
export async function waitForBaselineArtifact(options: BaselineWaitOptions): Promise<{ artifact: { id: number; run: number } } | { reason: string }> {
const now = options.now ?? Date.now;
const sleep = options.sleep ?? ((milliseconds: number, signal?: AbortSignal) => delay(milliseconds, undefined, signal ? { signal } : {}));
const progress = options.progress ?? silentProgress;
const started = now();
const short = options.sha.slice(0, 8);
try {
const lookup = await baselineLookup({ sha: options.sha, target: options.target, ...(options.environment ? { environment: options.environment } : {}),
...(options.transport ? { transport: options.transport } : {}),
retry: read => retryGitHubLookup(read, transientGitHubFailure, milliseconds => sleep(milliseconds, options.signal)) });
if (!lookup) return { reason: "the repository has no baseline workflow" };
const { context, get, target } = lookup;
// The matrix names each baseline job after its target; an older single-platform workflow's one job is `baseline`.
const jobNames = [`baseline (${target})`, ...(target === "ios" ? ["baseline"] : [])];
let finishedAt: number | undefined;
let logged = "";
for (;;) {
const run = await get<GithubRun>(`repos/${context.repository}/actions/runs/${options.run}`);
if (!trustedBaselineRun(run, lookup)) return { reason: `run ${options.run} is not a baseline run of ${short} on ${lookup.defaultBranch}` };
for (const name of lookup.names) {
const listed = await get<{ artifacts?: GithubArtifact[] }>(`repos/${context.repository}/actions/runs/${run.id}/artifacts?name=${encodeURIComponent(name)}&per_page=100`);
const artifact = (listed?.artifacts ?? []).filter(candidate => !candidate.expired && (candidate.name ?? name) === name
&& Number.isSafeInteger(candidate.id) && candidate.id > 0).sort((a, b) => b.id - a.id)[0];
if (artifact) return { artifact: { id: artifact.id, run: run.id } };
}
const jobs = await get<{ jobs?: GithubJob[] }>(`repos/${context.repository}/actions/runs/${run.id}/jobs?per_page=100`);
const job = jobs?.jobs?.find(candidate => jobNames.includes(candidate.name));
if (job?.status === "completed" && job.conclusion !== "success") {
return { reason: `its ${target} job ${job.conclusion === "cancelled" ? "was cancelled" : job.conclusion === "failure" ? "failed" : `ended (${job.conclusion ?? "no conclusion"})`}` };
}
if (job?.status === "completed" || run.status === "completed") {
finishedAt ??= now();
if (now() - finishedAt >= ARTIFACT_LISTING_GRACE_MS) return { reason: job ? `its ${target} job finished, but GitHub lists no ${lookup.names[0]} artifact`
: `the run ended (${run.conclusion ?? "no conclusion"}) without a ${target} baseline job` };
}
const waited = now() - started;
const state = job ? statusText(job.status) : run.status === "completed" ? "finished" : "not started";
if (waited >= options.timeoutMs) return { reason: `its ${target} job is still ${state} after ${Math.round(waited / 60_000)} min` };
// A line when the state changes and once a minute, not on every poll.
const line = `waiting for ${lookup.defaultBranch}'s baseline run ${run.id} of ${short} (${target}: ${state}, ${Math.floor(waited / 60_000)} min)`;
if (line !== logged) { progress(line); logged = line; }
await sleep(Math.min(options.pollMs ?? POLL_MS, options.timeoutMs - waited), options.signal);
}
} catch (error) {
return { reason: options.signal?.aborted ? "the review stopped waiting" : describe(error) };
}
}
/**
* Downloads a baseline artifact and unpacks it into `directory`, which must not exist yet: what
* actions/download-artifact would have put in EXPLORE_CAPTURE_CACHE. The token goes to GitHub only, never to the
* storage URL GitHub redirects to. Paths are confined to the directory and the cache's size limits apply before
* anything is written; restoreCaptureArtifactCache then checks the evidence as it checks a downloaded one.
*/
export async function downloadBaselineArtifact(options: { artifact: number; directory: string; environment?: NodeJS.ProcessEnv; transport?: typeof fetch;
signal?: AbortSignal; sleep?: (milliseconds: number, signal?: AbortSignal) => Promise<void> }): Promise<void> {
const environment = options.environment ?? process.env; const transport = options.transport ?? fetch;
const context = githubContext(environment);
const sleep = options.sleep ?? ((milliseconds: number, signal?: AbortSignal) => delay(milliseconds, undefined, signal ? { signal } : {}));
const signal = (milliseconds: number) => options.signal ? AbortSignal.any([options.signal, AbortSignal.timeout(milliseconds)]) : AbortSignal.timeout(milliseconds);
const path = `repos/${context.repository}/actions/artifacts/${options.artifact}/zip`;
// A 404 just after the artifact was listed is the Actions API catching up (#47), as a 5xx or no answer may be.
const located = await retryGitHubLookup(async () => {
let response: Response;
try {
response = await transport(`https://api.github.com/${path}`, { redirect: "manual", signal: signal(60_000),
headers: { authorization: `Bearer ${context.token}`, accept: "application/vnd.github+json", "x-github-api-version": "2022-11-28" } });
} catch (error) {
throw new CiError({ code: "GITHUB_REQUEST_FAILED", stage: "capture-base", message: `GitHub could not be reached (${path}).`,
nextAction: "Check GitHub availability.", retryable: true, cause: error });
}
if (response.status >= 300 && response.status < 400) {
await response.body?.cancel().catch(() => {});
return { location: response.headers.get("location") ?? "" };
}
if (response.ok) return { response };
await response.body?.cancel().catch(() => {});
throw new CiError({ code: response.status === 403 ? "GITHUB_PERMISSION_DENIED" : "GITHUB_REQUEST_FAILED", stage: "capture-base",
message: `GitHub answered HTTP ${response.status} for ${path}.`, retryable: response.status === 404 || response.status >= 500,
nextAction: response.status === 403 ? "Grant the capture job actions: read." : "Rerun the workflow; check GitHub's status if it keeps failing." });
}, transientGitHubFailure, milliseconds => sleep(milliseconds, options.signal));
let response = located.response;
if (!response) {
const location = URL.canParse(located.location ?? "") ? new URL(located.location!) : undefined;
if (location?.protocol !== "https:") throw new Error("GitHub redirected the artifact download to no HTTPS address");
response = await transport(location.href, { signal: signal(10 * 60_000) });
if (!response.ok || !response.body) throw new Error(`The artifact download answered HTTP ${response.status}`);
}
await mkdir(dirname(resolve(options.directory)), { recursive: true });
const stage = await mkdtemp(`${resolve(options.directory)}.download-`);
try {
const archive = join(stage, "artifact.zip");
const file = await open(archive, "wx", 0o600);
try {
let bytes = 0;
for await (const chunk of response.body as unknown as AsyncIterable<Uint8Array>) {
bytes += chunk.byteLength;
if (bytes > MAX_ARTIFACT_ZIP_BYTES) throw new Error("The baseline artifact is larger than a capture cache can be");
await file.write(chunk);
}
} finally { await file.close(); }
await unpackArtifact(archive, options.directory);
} finally {
await rm(stage, { recursive: true, force: true });
}
}
async function unpackArtifact(archive: string, directory: string): Promise<void> {
configure({ useWebWorkers: false });
const zip = new ZipReader(new BlobReader(await openAsBlob(archive)), { useWebWorkers: false });
try {
const entries = await zip.getEntries();
const files = entries.filter(entry => !entry.directory);
// The cache's own metadata file is the one file beyond its evidence.
if (files.length > CAPTURE_CACHE_MAX_FILES + 1 || files.reduce((sum, entry) => sum + entry.uncompressedSize, 0) > CAPTURE_CACHE_MAX_BYTES + 2 * 1024 * 1024) {
throw new Error("The baseline artifact exceeds the capture cache's limits");
}
const root = resolve(directory);
await mkdir(root);
for (const entry of files) {
const path = safeRelativePath(entry.filename);
const target = path ? resolve(root, path) : undefined;
if (!target || relative(root, target).startsWith("..") || !entry.getData) throw new Error(`The baseline artifact holds an unsafe path: ${entry.filename}`);
await mkdir(dirname(target), { recursive: true });
await writeFile(target, await entry.getData(new Uint8ArrayWriter(), { checkSignature: true, useWebWorkers: false }), { flag: "wx", mode: 0o600 });
}
} finally {
await zip.close();
}
}
/**
* The wait as `capture:review` runs it: the downloaded baseline's directory, or undefined, and then base captures as it
* would have. It never fails the review; each outcome is one progress line.
*/
export async function awaitPendingBaseline(options: BaselineWaitOptions & { directory: string }): Promise<string | undefined> {
const progress = options.progress ?? silentProgress;
const short = options.sha.slice(0, 8);
const outcome = await waitForBaselineArtifact(options);
if ("reason" in outcome) {
progress(`base: stopped waiting for the baseline of ${short} (${outcome.reason}); capturing the base too`);
return undefined;
}
try {
await downloadBaselineArtifact({ ...options, artifact: outcome.artifact.id });
progress(`base: downloaded the baseline of ${short} that run ${outcome.artifact.run} captured`);
return options.directory;
} catch (error) {
await rm(options.directory, { recursive: true, force: true }).catch(() => {});
progress(`base: the baseline of ${short} from run ${outcome.artifact.run} could not be downloaded (${options.signal?.aborted ? "the review stopped waiting" : describe(error)}); capturing the base too`);
return undefined;
}
}
+2 -2
View File
@@ -1,6 +1,6 @@
import { Worker, isMainThread, parentPort, workerData } from "node:worker_threads";
import { runCiMap, type CiMapOptions, type CiMapResult } from "../capture/run.js";
import { CiError, serializeCiError, type CiErrorData } from "../shared/ci-errors.js";
import { CiError, environmentSecrets, serializeCiError, type CiErrorData } from "../shared/ci-errors.js";
import type { Progress } from "../shared/progress.js";
type WorkerMessage = { type: "progress"; message: string } | { type: "done"; result: CiMapResult } | { type: "error"; error: CiErrorData };
@@ -41,5 +41,5 @@ if (!isMainThread && parentPort) {
port.unref();
runCiMap({ ...options, signal: controller.signal, progress: message => port.postMessage({ type: "progress", message } satisfies WorkerMessage) })
.then(result => port.postMessage({ type: "done", result } satisfies WorkerMessage))
.catch((error: unknown) => port.postMessage({ type: "error", error: serializeCiError(error, { secrets: [options.expoToken] }) } satisfies WorkerMessage));
.catch((error: unknown) => port.postMessage({ type: "error", error: serializeCiError(error, { secrets: [options.expoToken, ...environmentSecrets(options.environment ?? {})] }) } satisfies WorkerMessage));
}
+15 -7
View File
@@ -474,7 +474,8 @@ export async function runDoctor(options: DoctorOptions): Promise<DoctorReport> {
checks.push(config.mode === "agent"
? { step: 3, id: "mode", status: "pass", message: `Agent mode with ${config.agent!.provider} ${config.agent!.model}. Capture navigation uses the model and is billed to your ${config.agent!.provider} account.` }
: { step: 3, id: "mode", status: "pass", message: "Deterministic mode: capture replays config.yaml without a model, so it needs no API key and has no AI charges." });
if (runTargets(config, { controller: options.controller }).some(target => target.capture.controller === "argent")) checks.push(argentCheck(environment));
const argentTargets = runTargets(config, { controller: options.controller }).filter(target => target.capture.controller === "argent");
if (argentTargets.length) checks.push(argentCheck(environment, argentTargets.some(target => target.app.platform === "android" && target.capture.simulator !== "local")));
if (config.capture.launch !== "off") {
const ffmpeg = run("ffmpeg", ["-version"], project);
checks.push(ffmpeg.status === 0
@@ -630,13 +631,20 @@ export function summarizeCaptureCheck(plan: SourceCapturePlan, bundle: CaptureBu
* Capture the committed HEAD the way the baseline workflow does, with the local EAS login. This is
* the only doctor step that starts an EAS build and a simulator session.
*/
/** An Argent capture records one appearance, and fetches the pinned Argent release on first use. agent-device, the default, is not reported. */
function argentCheck(environment: NodeJS.ProcessEnv): DoctorCheck {
/**
* Argent fetches its pinned release on first use, and cannot switch the appearance itself (software-mansion/argent#1509):
* the host does, through `simulator:control` on an iOS EAS session or simctl and adb on this Mac. Only an Android EAS
* session has no host to ask, so its captures record one appearance. agent-device, the default, is not reported.
*/
function argentCheck(environment: NodeJS.ProcessEnv, androidOnEas: boolean): DoctorCheck {
const cached = controllerCached("argent", environment);
return { step: 3, id: "controller", status: "warn", code: "CONTROLLER_APPEARANCE_FIXED",
message: `Device controller: Argent ${ARGENT_VERSION} (${cached ? "cached on this machine" : "fetched from npm on the first capture, about 27 MB"}). `
+ "Argent cannot switch the simulator's appearance yet (software-mansion/argent#1509), so each capture records one appearance: the app's forced one, else light.",
nextAction: "Nothing to do unless you need dark-mode screenshots; use agent-device (the default) for those." };
const release = `Device controller: Argent ${ARGENT_VERSION} (${cached ? "cached on this machine" : "fetched from npm on the first capture, about 27 MB"}). `;
return androidOnEas
? { step: 3, id: "controller", status: "warn", code: "CONTROLLER_APPEARANCE_FIXED",
message: `${release}Argent cannot switch the appearance on an Android EAS Simulator session yet (software-mansion/argent#1509), so those captures record one appearance: the app's forced one, else light.`,
nextAction: "Nothing to do unless you need dark-mode screenshots on Android; use agent-device (the default) for those." }
: { step: 3, id: "controller", status: "pass",
message: `${release}The simulator's host switches the appearance, so every appearance the config asks for is captured.` };
}
export async function runCaptureCheck(options: { project: string; output: string; platform?: ExplorePlatform; controller?: string | undefined; simulator?: string | undefined; repack?: string | undefined; target?: string | undefined; environment?: NodeJS.ProcessEnv; runner?: CommandRunner; progress?: Progress }): Promise<CaptureCheckResult> {
+110
View File
@@ -0,0 +1,110 @@
import { copyFile, mkdir, readdir, readFile, writeFile, appendFile, lstat } from "node:fs/promises";
import { dirname, join, relative } from "node:path";
import { bundleColorSchemes, mergeSchemeOutcomes, routeIdOfScreen, schemeScreenId, type ColorScheme } from "../capture/appearance.js";
import type { CaptureOutcome } from "../contracts/capture.js";
import type { CaptureBundle } from "../contracts/capture.js";
import type { ExploreMap } from "../types.js";
const LIST_FIELDS = ["artifacts", "screens", "variants", "observations", "elements", "actionAttempts", "transitions", "journeys", "frontier", "gaps"] as const;
async function readJson<T>(path: string): Promise<T> { return JSON.parse(await readFile(path, "utf8")) as T; }
async function copyTree(source: string, destination: string, skip: (relativePath: string) => boolean, root = source): Promise<void> {
for (const entry of await readdir(source, { withFileTypes: true })) {
const from = join(source, entry.name), to = join(destination, entry.name), rel = relative(root, from);
if (skip(rel)) continue;
if (entry.isDirectory()) { await mkdir(to, { recursive: true }); await copyTree(from, to, skip, root); continue; }
if (!entry.isFile()) continue;
if ((await lstat(to).catch(() => undefined))) continue;
await mkdir(dirname(to), { recursive: true });
await copyFile(from, to);
}
}
/**
* A secondary appearance session's result: its export, or the error it failed with, when it left none. In brentvatne/euxy's
* main baseline run 39506374824 the dark session's install failed after the light session had captured 16 of 16 routes.
*/
export type SecondaryCaptureExport = { directory: string; scheme: ColorScheme } | { scheme: ColorScheme; error: NonNullable<CaptureOutcome["error"]> };
/**
* Fold the exports of a side's secondary appearance sessions into its primary export, in place: the frames, outcomes,
* launch evidence and map records of each `<id>@<scheme>` screen, and their files. Each session captured its passes as
* one run would (`appearancePass`), so records only append; a secondary export never replaces a primary record. A session
* that failed is listed in the bundle's `sessionFailures`, adds a failed `<id>@<scheme>` outcome carrying its error for
* every selected route, and leaves the run partial, so its appearance reads as unavailable with its own reason (a route
* the primary lost too would otherwise show the primary's) instead of going missing.
*/
export async function mergeCaptureExports(primary: string, secondaries: readonly SecondaryCaptureExport[]): Promise<void> {
if (secondaries.length === 0) return;
const bundle = await readJson<CaptureBundle>(join(primary, "bundle.json"));
const map = await readJson<ExploreMap>(join(primary, "screen-map.json"));
const known = { screens: new Set(map.screens.map(screen => screen.id)), artifacts: new Set(map.artifacts.map(artifact => artifact.id)) };
// The outcomes recombine as one run's passes do (mergeSchemeOutcomes): a secondary pass records a verified frame under
// `<id>@<scheme>` and a lost route under `<id>`, which the merge folds into the primary record the same way.
const schemes = bundleColorSchemes(bundle);
const passes: Record<string, CaptureOutcome>[] = schemes.map(() => ({}));
for (const [id, outcome] of Object.entries(bundle.outcomes)) if (!routeIdOfScreen(id).scheme) passes[0]![id] = outcome;
const failedSessions: string[] = [];
const failedOutcomes: Record<string, CaptureOutcome> = {};
for (const secondary of secondaries) {
const { scheme } = secondary;
if ("error" in secondary) {
failedSessions.push(`The ${scheme} session failed (${secondary.error.code}): ${secondary.error.message.replace(/\.+$/, "")}.`);
(bundle.sessionFailures ??= []).push({ scheme, error: secondary.error });
// A primary that renders no such appearance has no slot for its frames; the failure stays on the bundle alone.
if (!schemes.includes(scheme)) continue;
for (const id of bundle.selectedSurfaceIds) {
failedOutcomes[schemeScreenId(id, scheme)] = { status: "failed", attempts: 0, landing: "unverified", reason: secondary.error.message, error: secondary.error };
}
continue;
}
const other = await readJson<CaptureBundle>(join(secondary.directory, "bundle.json"));
const pass = passes[schemes.indexOf(scheme)];
// A session that could not show its appearance (Argent without appearance.set captures one) has nothing to add, as
// one run would have no pass for it.
if (!pass && !bundleColorSchemes(other).includes(scheme)) continue;
if (!pass) throw new Error(`The ${scheme} session's export is not an appearance of the primary export (${schemes.join(", ")})`);
for (const [id, outcome] of Object.entries(other.outcomes)) pass[routeIdOfScreen(id).routeId] = outcome;
const otherMap = await readJson<ExploreMap>(join(secondary.directory, "screen-map.json"));
for (const [id, image] of Object.entries(other.images)) if (routeIdOfScreen(id).scheme && !bundle.images[id]) bundle.images[id] = image;
for (const [scheme, record] of Object.entries(other.launch ?? {})) { bundle.launch ??= {}; bundle.launch[scheme] ??= record; }
for (const [id, context] of Object.entries(other.captureContexts ?? {})) { bundle.captureContexts ??= {}; bundle.captureContexts[id] ??= context; }
for (const field of LIST_FIELDS) {
const records = (otherMap[field] ?? []) as Array<Record<string, unknown>>;
const target = ((map as unknown as Record<string, unknown[]>)[field] ??= []);
for (const record of records) {
if (field === "screens" && known.screens.has(record.id as string)) continue;
if (field === "artifacts" && known.artifacts.has(record.id as string)) continue;
if (field === "screens") known.screens.add(record.id as string);
if (field === "artifacts") known.artifacts.add(record.id as string);
target.push(record);
}
}
// Files: artifacts and launch evidence are new paths; diagnostics append to the primary's lines.
await copyTree(secondary.directory, primary, rel => rel === "bundle.json" || rel === "screen-map.json" || rel === "screen-map.schema.json" || rel === "manifest.json"
|| rel === "execution.json" || rel.startsWith("capture-diagnostics"));
const diagnostics = join(secondary.directory, "capture-diagnostics");
for (const entry of await readdir(diagnostics, { withFileTypes: true }).catch(() => [])) {
if (!entry.isFile()) continue;
const to = join(primary, "capture-diagnostics", entry.name);
await mkdir(dirname(to), { recursive: true });
if (entry.name.endsWith(".jsonl")) await appendFile(to, await readFile(join(diagnostics, entry.name)));
else if (!(await lstat(to).catch(() => undefined))) await copyFile(join(diagnostics, entry.name), to);
}
}
// An open gap a secondary pass left makes the run partial, as it does one run's (a complete run has none), and so does
// a secondary session that failed. A partial run names why, as the map's schema requires.
if (map.gaps.some(gap => gap.status === "open") || failedSessions.length) {
map.run.status = "partial";
map.run.stopReason = [map.run.stopReason ?? (failedSessions.length ? undefined : "Some source targets could not be captured; see bundle outcomes."),
...failedSessions].filter(Boolean).join(" ");
}
const manifest = await readJson<{ files: unknown[] }>(join(primary, "manifest.json"));
manifest.files = map.artifacts.map(artifact => ({ path: artifact.path, sha256: artifact.sha256, bytes: artifact.bytes, mimeType: artifact.mimeType }));
await writeFile(join(primary, "screen-map.json"), JSON.stringify(map));
bundle.outcomes = { ...mergeSchemeOutcomes(passes, schemes), ...failedOutcomes };
await writeFile(join(primary, "bundle.json"), JSON.stringify(bundle));
await writeFile(join(primary, "manifest.json"), `${JSON.stringify(manifest, null, 2)}\n`);
}
+28 -9
View File
@@ -1,14 +1,14 @@
import type { RunTarget } from "../types.js";
import { spawnSync } from "node:child_process";
import { randomBytes } from "node:crypto";
import { readFileSync } from "node:fs";
import { readFileSync, readdirSync } from "node:fs";
import { copyFile, cp, lstat, mkdir, mkdtemp, readdir, readFile, realpath, rename, rm, stat, utimes, writeFile } from "node:fs/promises";
import { tmpdir } from "node:os";
import { join, relative, resolve } from "node:path";
import { runCiReview } from "./pipeline.js";
import { resolveCommit } from "./workflow-steps.js";
import { ciLayout, installCommand, type PackageManager } from "./workflow-install.js";
import { configPlatforms, readExploreConfig, resolveTargetConfig, configTargets, withRunOverrides, DEFAULT_EXPLORE_CONFIG_PATH, type ExploreConfig, type ExplorePlatform } from "../config/explore-config.js";
import { configPlatforms, readExploreConfig, resolveTargetConfig, configTargets, posedTargets, withRunOverrides, DEFAULT_EXPLORE_CONFIG_PATH, type ExploreConfig, type ExplorePlatform } from "../config/explore-config.js";
import { AGENT_PROVIDERS, defaultExploreConfig } from "../config/setup.js";
import type { PrepareCiExplorationOptions } from "../capture/agent/exploration.js";
import { createCiAgentSession } from "../capture/agent/session.js";
@@ -26,6 +26,7 @@ import { abandonSharedBuilds } from "../capture/shared-build.js";
import { exploreCacheRoot } from "../shared/cache-root.js";
import { packageBin } from "../shared/package-paths.js";
import { silentProgress, type Progress } from "../shared/progress.js";
import { maxParallelTargets } from "./session-plan.js";
/**
* `capture:review --base REF [--head REF]` on a laptop: the same review CI runs (`runCiReview`), on two commits of the
@@ -44,7 +45,7 @@ export type LocalReviewOptions = {
platform?: ExplorePlatform | undefined;
/** `--target`: capture this config.yaml target alone; with `platform`, the targets of that platform. */
target?: string | undefined;
/** `--parallel`: how many targets capture at once (each runs a simulator or emulator per side); all of them by default. */
/** `--parallel`: how many targets capture at once (each runs a simulator or emulator per side); by default all of them, up to capture.maxSessions. */
parallel?: number | undefined;
scope?: string | undefined;
expiresIn?: string | undefined;
@@ -285,8 +286,26 @@ export async function keepCapture(exported: string, target: string): Promise<boo
}
}
/** The session state files a review writes: `<side>-session`, and `<side>-session-<scheme>` for appearance passes in their own sessions. */
function sessionStateFiles(output: string): string[] {
const rank = (name: string) => ["head", "base", "baseline"].indexOf(name.split("-")[0]!);
try { return readdirSync(output).filter(name => /^(head|base|baseline)-session(-[a-z]+)?$/.test(name)).sort((a, b) => rank(a) - rank(b) || a.localeCompare(b)); }
catch { return []; }
}
/**
* A finished target keeps only its report: the captures, maps and checkouts are the bulk of the run. A session file
* stays, since it names a session no capture confirmed stopped; the review stops it on its way out (stopRecordedSessions).
*/
export async function pruneTargetOutput(output: string): Promise<void> {
const sessions = new Set(sessionStateFiles(output));
for (const entry of await readdir(output)) {
if (entry !== "report" && entry !== "report.zip" && !sessions.has(entry)) await rm(join(output, entry), { recursive: true, force: true });
}
}
function recordedSessionIds(outputs: string[]): string[] {
return outputs.flatMap(output => ["head-session", "base-session"].flatMap(name => {
return outputs.flatMap(output => sessionStateFiles(output).flatMap(name => {
try { const id = readFileSync(join(output, name), "utf8").trim(); return /^[A-Za-z0-9][A-Za-z0-9._-]{0,199}$/.test(id) ? [id] : []; } catch { return []; }
}));
}
@@ -307,7 +326,7 @@ export async function stopRecordedSessions(outputs: string[], environment: NodeJ
simctl: (args: string[]) => { status: number | null } = args => spawnSync("xcrun", ["simctl", ...args], { env: environment, encoding: "utf8", timeout: 120_000 }),
adb: (args: string[]) => { status: number | null } = args => spawnSync(androidTools(environment).adb, args, { env: environment, encoding: "utf8", timeout: 60_000 })): Promise<string[]> {
const stopped: string[] = [];
for (const output of outputs) for (const name of ["head-session", "base-session"]) {
for (const output of outputs) for (const name of sessionStateFiles(output)) {
const file = join(output, name);
const id = (await readFile(file, "utf8").catch(() => "")).trim();
if (!/^[A-Za-z0-9][A-Za-z0-9._-]{0,199}$/.test(id)) continue;
@@ -479,7 +498,7 @@ export async function runLocalReview(options: LocalReviewOptions): Promise<Local
target, ...(targets.length > 1 ? { targets } : {}) };
progress(`${target.id}: reviewing${target.id === platform ? "" : ` (${target.label})`}`);
const result = await runCiReview({ projects, output, config, presentation, runId, runAttempt: "1", state, environment: platformEnvironment,
...(platforms.length > 1 ? { platforms } : {}), ...(targets.length > 1 ? { targets } : {}),
...(platforms.length > 1 ? { platforms } : {}), ...(targets.length > 1 ? { targets } : {}), posedTargets: posedTargets(rawConfig),
// Targets of one platform capture the same commits from the same build; it is built and repacked once.
...(targets.length > 1 ? { buildShare: join(work, ".builds") } : {}),
subscription, useLocalEasSession, progress: message => progress(`${target.id}: ${message}`),
@@ -518,8 +537,7 @@ export async function runLocalReview(options: LocalReviewOptions): Promise<Local
if (side === "head" && head.uncommitted) continue;
await keepCapture(join(output, `${side}-export`), captureCacheDirectory(environment, target.id, side === "base" ? base.sha : head.sha));
}
// The report is all that is kept; the captures, maps and checkouts are the bulk of the run.
for (const entry of await readdir(output)) if (entry !== "report" && entry !== "report.zip") await rm(join(output, entry), { recursive: true, force: true });
await pruneTargetOutput(output);
abort.signal.throwIfAborted();
return report;
};
@@ -537,7 +555,8 @@ export async function runLocalReview(options: LocalReviewOptions): Promise<Local
await abandonSharedBuilds(join(work, ".builds"), owner => owner === join(work, target.id, `${side}-export`), failure.code);
}
};
const settled = await settleLimited(targets, options.parallel ?? targets.length,
// Each target needs a simulator session: no more run at once than capture.maxSessions, unless --parallel says.
const settled = await settleLimited(targets, options.parallel ?? maxParallelTargets(rawConfig.capture.maxSessions, targets.length),
target => captureTarget(target).catch(async (error: unknown) => { await releaseBuilds(target, error).catch(() => undefined); throw error; }));
// One target that fails costs that target's report, not the others': they are kept, and the failure is named.
// Only an interrupted review, or one where every target failed, fails as a whole.
+282 -29
View File
@@ -1,17 +1,25 @@
import { systemRegions } from "../contracts/capture.js";
import type { RunTarget } from "../types.js";
import { spawnSync } from "node:child_process";
import { mkdir, open, readFile, realpath, writeFile } from "node:fs/promises";
import { execFile, spawnSync } from "node:child_process";
import { lstat, mkdir, mkdtemp, open, readFile, realpath, rm, writeFile } from "node:fs/promises";
import { join, relative, resolve, isAbsolute } from "node:path";
import { tmpdir } from "node:os";
import { setTimeout as delay } from "node:timers/promises";
import { promisify } from "node:util";
import { buildVerifiedPlan } from "./verified-plan.js";
import { runCiMap, type CommandRunner } from "../capture/run.js";
import { routeIdOfScreen } from "../capture/appearance.js";
import { runCiMap, type CiMapOptions, type CommandRunner } from "../capture/run.js";
import type { RepackShare } from "../capture/eas-build.js";
import { resolveColorSchemes, routeIdOfScreen, type ColorScheme } from "../capture/appearance.js";
import { baseEnvironment } from "../capture/common.js";
import { DEFAULT_MAX_SESSIONS, planSideSessions, sessionShare } from "./session-plan.js";
import { mergeCaptureExports, type SecondaryCaptureExport } from "./export-merge.js";
import { abandonSharedBuilds } from "../capture/shared-build.js";
import { prepareCaptureReview, readCaptureBundle } from "../review/capture-review.js";
import { quarantineReviewContract } from "../contracts/review-validation.js";
import { agentCaptureConstraints } from "../capture/agent/source-context.js";
import { captureArtifactCacheMayMatch, captureRecipeConfig, restoreCaptureArtifactCache, writeCaptureArtifactCache } from "./artifact-cache.js";
import { captureArtifactCacheMayMatch, captureRecipeConfig, incompleteCaptureReason, restoreCaptureArtifactCache, writeCaptureArtifactCache } from "./artifact-cache.js";
import { runCiMapOnWorker } from "./capture-worker.js";
import { DEFAULT_BASELINE_WAIT_MINUTES, awaitPendingBaseline } from "./baseline-wait.js";
import { writeCiReportInput, readAppDirectory } from "../report/artifact.js";
import { assessPublicEvidence, readExportedMap, type PublicEvidenceSide } from "../review/evidence-assessment.js";
import { parseRuntimeExecutionContext } from "../review/runtime-adapter.js";
@@ -22,10 +30,10 @@ import { viewerLinks } from "../report/viewer-links.js";
import { DEFAULT_EXPLORE_CONFIG_PATH, captureRepack, type ExploreConfig, type ExplorePlatform } from "../config/explore-config.js";
import { prepareCiExploration, type PrepareCiExplorationOptions } from "../capture/agent/exploration.js";
import { generateAgentChangeNotes } from "../capture/agent/change-notes.js";
import { CiError, serializeCiError, toCiError } from "../shared/ci-errors.js";
import { CiError, environmentSecrets, serializeCiError, toCiError } from "../shared/ci-errors.js";
import type { SourceCapturePlan } from "../source/capture-plan.js";
import { createSourceChangeSummaries } from "../source/change-summary.js";
import { assertNoCaptureHealthIssues, partitionCaptureHealthIssues, renderCaptureIssueAnnotation, renderComparisonDiagnosticAnnotation, renderNotesUnavailableAnnotation, renderReviewGapAnnotation, sortedCaptureHealthIssues } from "../capture/health.js";
import { assertNoCaptureHealthIssues, capturedEvidenceInDoubt, partitionCaptureHealthIssues, renderCaptureIssueAnnotation, renderComparisonDiagnosticAnnotation, renderNotesUnavailableAnnotation, renderReviewGapAnnotation, sortedCaptureHealthIssues } from "../capture/health.js";
import { EXPLORE_PACKAGE, packageBin } from "../shared/package-paths.js";
import { agentDeviceLauncher, ensureArgentToolServer, ensureController } from "../shared/controller-cache.js";
import { silentProgress, type Progress } from "../shared/progress.js";
@@ -63,6 +71,8 @@ async function recipe(config: ExploreConfig, plan: SourceCapturePlan, side: "bas
/** The capture's clock and host-side pause; tests pair them with an injected runner so readiness budgets are not real time. */
type CaptureClock = { now?: (() => number) | undefined; sleep?: ((milliseconds: number) => void) | undefined };
const execFileAsync = promisify(execFile);
function captureOptions(config: ExploreConfig, environment: NodeJS.ProcessEnv, runner?: CommandRunner, clock: CaptureClock = {}) {
return { ...(runner && clock.now ? { now: clock.now } : {}), ...(runner && clock.sleep ? { sleep: clock.sleep } : {}), platform: config.app.platform, bundleId: config.app.bundleId,
...(config.app.scheme ? { scheme: config.app.scheme } : {}), device: config.device.name, ...(config.device.pose ? { devicePose: config.device.pose } : {}),
@@ -72,6 +82,21 @@ function captureOptions(config: ExploreConfig, environment: NodeJS.ProcessEnv, r
...(config.capture.simulator ? { simulator: config.capture.simulator } : {}),
environment, ...(runner ? { runner } : {}), easCommand: packageBin("eas-cli", "eas"), agentDeviceCommand: agentDeviceLauncher(environment), repackCommand: packageBin("@expo/repack-app", "repack-app") };
}
/**
* The hosted repack this capture shares with the other matrix jobs of its GitHub run attempt (RepackShare): only inside
* a GitHub Actions run, whose jobs are the only ones that meet, and only when another target repacks the same app
* (`peers`, in config order, from hostedRepackPeers). A run attempt's repacks are never used by another attempt or run,
* such as the baseline workflow's: "Re-run failed jobs" after an EAS environment value was fixed rebundles with it.
* Each target's `rank` (its place among the peers) staggers how long it waits for another's repack (sharedRepackWaitMs);
* a re-run attempt runs only the jobs it re-runs, so its leader may not be one of them, and its stagger starts at once.
*/
export function repackShareOption(config: ExploreConfig, environment: NodeJS.ProcessEnv, peers: readonly string[] | undefined): { repackShare?: RepackShare } {
const target = config.target?.id;
if (!peers || peers.length < 2 || !target || !peers.includes(target) || captureRepack(config) !== "eas") return {};
const run = environment.GITHUB_RUN_ID ?? "", attempt = environment.GITHUB_RUN_ATTEMPT ?? "1", repository = environment.GITHUB_REPOSITORY ?? "";
if (!/^[1-9][0-9]*$/.test(run) || !/^[1-9][0-9]*$/.test(attempt) || !/^[A-Za-z0-9_.-]+\/[A-Za-z0-9_.-]+$/.test(repository)) return {};
return { repackShare: { scope: `${repository}#${run}.${attempt}`, leader: peers[0]!, target, rank: peers.indexOf(target), ...(attempt === "1" ? {} : { rerun: true }) } };
}
/**
* Head and base run in parallel worker threads when nothing ties them together: an agent session is
* shared, an injected runner belongs to this thread, and a plausible baseline needs head's runtime profile.
@@ -84,6 +109,148 @@ async function prefetchControllers(config: ExploreConfig, environment: NodeJS.Pr
if (config.capture.controller === "argent" && config.capture.simulator === "local") await ensureArgentToolServer({ environment, progress });
}
/**
* The appearances a capture of this side renders, read before any session starts so the sessions can divide them. The
* app config comes from `eas config` in a child process, as the capture reads it, with the capture's restricted
* environment: never evaluated in this process, which holds the review's secrets. Unknown: undefined.
*/
async function expectedColorSchemes(config: ExploreConfig, project: string, environment: NodeJS.ProcessEnv, easCommand: string): Promise<ColorScheme[] | undefined> {
// Argent switches the appearance only through serve-sim on an EAS iOS session; elsewhere it captures one.
if (config.capture.controller === "argent" && config.capture.simulator !== "local" && config.app.platform !== "ios") return undefined;
if (config.capture.appearance !== "auto") return resolveColorSchemes(config.capture.appearance, undefined, config.app.platform);
const easEnvironment = { ...baseEnvironment(environment), ...(environment.EXPO_TOKEN ? { EXPO_TOKEN: environment.EXPO_TOKEN } : {}) };
try {
const { stdout } = await execFileAsync(easCommand, ["config", "--platform", config.app.platform, "--profile", config.app.easProfile, "--json", "--non-interactive"],
{ cwd: project, env: easEnvironment, timeout: 5 * 60_000, maxBuffer: 16 * 1024 * 1024 });
const app = (JSON.parse(stdout) as { appConfig?: Record<string, unknown> }).appConfig;
return app && typeof app === "object" ? resolveColorSchemes("auto", app, config.app.platform) : undefined;
} catch {
// The capture reads the same config and reports why it cannot; until then the side keeps one session.
return undefined;
}
}
/**
* Failures that fail the side even when only a secondary appearance session had them. A session Explore could not
* confirm stopped, or whose identity it could not verify, must never be hidden behind a partial report; a checkout that
* changed, evidence that failed its integrity checks, or a failure Explore did not classify means the primary's frames
* may not be what they claim either.
*/
const SIDE_FAILURE_CODES = new Set(["SESSION_CLEANUP_FAILED", "SESSION_HANDOFF_UNVERIFIED", "CAPTURE_WORKER_EXITED", "EXPLORE_FAILED"]);
/**
* Whether a secondary appearance session's failure can be reported as that appearance's frames failing, keeping the
* primary's: only when the capture reported it itself (a worker that died before its cleanup ran rejects otherwise), its
* session is proven stopped (the capture removes its session file only after a confirmed stop), and nothing in it
* casts doubt on the primary's evidence.
*/
async function secondaryFailureKept(error: unknown, sessionStatePath: string | undefined): Promise<boolean> {
if (!(error instanceof CiError)) return false;
if (SIDE_FAILURE_CODES.has(error.code) || capturedEvidenceInDoubt(error.code)
|| (error.details as { allocation?: unknown } | undefined)?.allocation === "unknown") return false;
return !(sessionStatePath && await lstat(sessionStatePath).then(() => true, () => false));
}
/**
* Run one side's capture: in one session, or when `passes` splits its appearances, one worker session per pass. The
* primary pass exports to `mapOptions.outputDirectory`; the others beside it, named by their appearances, then merge.
* `runPass` replaces the worker capture of each pass (tests).
*/
export async function captureSideInSessions(mapOptions: CiMapOptions, passes: readonly (readonly ColorScheme[])[] | undefined,
progress: Progress, signal: AbortSignal | undefined, worker: boolean,
runPass?: (passOptions: CiMapOptions) => Promise<Awaited<ReturnType<typeof runCiMap>>>): Promise<Awaited<ReturnType<typeof runCiMap>>> {
const runOnWorker = runPass ?? ((passOptions: CiMapOptions) =>
runCiMapOnWorker((({ progress: _progress, runner: _runner, agent: _agent, signal: _signal, ...rest }) => rest)(passOptions), progress, signal));
if (!passes || passes.length <= 1) return worker ? runOnWorker(mapOptions) : runCiMap(mapOptions);
const schemes = passes.flat();
progress(`${mapOptions.side}: capturing ${passes.map(pass => pass.join("+")).join(" and ")} in separate sessions`);
// The passes install the same app: they share one build and repack, as the targets of a local review do.
const ownBuildShare = mapOptions.buildShare ? undefined : await mkdtemp(join(tmpdir(), "expo-explore-builds-"));
const buildShare = mapOptions.buildShare ?? ownBuildShare!;
const settled = await Promise.allSettled(passes.map(async (pass, index) => {
const suffix = index === 0 ? "" : `-${pass.join("-")}`;
const outputDirectory = `${mapOptions.outputDirectory}${suffix}`;
try {
return await runOnWorker({ ...mapOptions, mapDirectory: `${mapOptions.mapDirectory}${suffix}`, outputDirectory, buildShare,
...(mapOptions.sessionStatePath ? { sessionStatePath: `${mapOptions.sessionStatePath}${suffix}` } : {}), appearancePass: { all: schemes, capture: pass } });
} catch (error) {
// A pass that dies mid-build must not leave the others polling for its result.
await abandonSharedBuilds(buildShare, owner => owner === outputDirectory, toCiError(error).code).catch(() => undefined);
throw error;
}
}));
// The build folder holds the repacked app, which the sessions have installed.
if (ownBuildShare) await rm(ownBuildShare, { recursive: true, force: true });
// Only the primary appearance's session fails the side, besides a stopped review and a secondary failure that cannot be
// kept (secondaryFailureKept). In brentvatne/euxy's main baseline run 39506374824 the light session captured 16 of 16
// routes and the dark session's install failed, which failed the whole target and lost its baseline. Such a failure is
// kept in the report instead: every frame of its appearance is failed with its error, and the run is partial.
const suffixOf = (pass: readonly ColorScheme[]) => `-${pass.join("-")}`;
for (const [index, outcome] of settled.entries()) {
if (outcome.status !== "rejected" || index === 0) continue;
const statePath = mapOptions.sessionStatePath ? `${mapOptions.sessionStatePath}${suffixOf(passes[index]!)}` : undefined;
if (signal?.aborted || !await secondaryFailureKept(outcome.reason, statePath)) throw outcome.reason;
}
const primary = settled[0]!;
if (primary.status === "rejected") throw primary.reason;
const secrets = [mapOptions.expoToken, ...environmentSecrets(mapOptions.environment ?? {})];
const secondaries = passes.slice(1).map((pass, index): SecondaryCaptureExport => {
const outcome = settled[index + 1]!;
if (outcome.status === "fulfilled") return { directory: `${mapOptions.outputDirectory}${suffixOf(pass)}`, scheme: pass[0]! };
const error = serializeCiError(outcome.reason, { secrets });
progress(`${mapOptions.side}: the ${pass.join("+")} session failed (${error.code}: ${error.message}); keeping the ${passes[0]!.join("+")} captures, and the report marks every ${pass.join("+")} frame failed`);
return { scheme: pass[0]!, error: { code: error.code, stage: error.stage, message: error.message, nextAction: error.nextAction, retryable: error.retryable } };
});
await mergeCaptureExports(mapOptions.outputDirectory, secondaries);
const result = primary.value;
return secondaries.some(secondary => "error" in secondary) ? { ...result, runStatus: "partial" } : result;
}
/**
* This target's share of `capture.maxSessions`: the targets are this review's (`targets`), or the config's one, and
* `posedTargets` (foldables) take the sessions that do not divide evenly.
*/
function targetSessionShare(config: ExploreConfig, targets: readonly RunTarget[] | undefined, posedTargets: readonly string[] | undefined): number {
const targetIds: string[] = (targets ?? []).map(target => target.id);
if (targetIds.length === 0) targetIds.push(...(config.targets ? Object.keys(config.targets) : [config.target?.id ?? "default"]));
return sessionShare({ maxSessions: config.capture.maxSessions, targets: targetIds, target: config.target?.id ?? targetIds[0]!,
prefer: id => posedTargets?.includes(id) ? 1 : 0 });
}
/**
* The appearance passes one side spreads over sessions of its own, out of its target's share of `capture.maxSessions`;
* undefined keeps the side in one session. The side's app config is read only when its share and routes could pay for
* a split.
*/
export async function sideAppearancePasses(config: ExploreConfig, environment: NodeJS.ProcessEnv,
input: { share: number; sides: 1 | 2; routes: number; project: string; easCommand?: string }): Promise<ColorScheme[][] | undefined> {
// A pass observes its launch only when capture.launch asks for it.
const plan = (schemes: readonly ColorScheme[]) => planSideSessions({ share: input.share, sides: input.sides, routes: input.routes, schemes,
...(config.capture.launch === "off" ? { launchSeconds: 0 } : {}) }).passes;
if (plan(["light", "dark"]).length < 2) return undefined;
const schemes = await expectedColorSchemes(config, input.project, environment, input.easCommand ?? packageBin("eas-cli", "eas"));
const passes = schemes ? plan(schemes) : [];
return passes.length > 1 ? passes.map(pass => [...pass]) : undefined;
}
/**
* Base's start in a review that waits for a baseline still being captured; head has already started. Base captures as
* soon as the wait gives up, beside head when the sides capture concurrently, and a downloaded baseline is restored
* once head finishes, since restoring checks it against head's runtime profile. A head that fails ends the wait, and
* base is never started for a review that has already failed.
*/
export async function captureBaseAfterWait(input: { waited: Promise<string | undefined> | undefined; head: Promise<void>; stopWaiting: () => void;
captureBase: (cacheDirectory?: string) => Promise<void> }): Promise<void> {
if (!input.waited) return input.captureBase();
let headFailed = false;
input.head.catch(() => { headFailed = true; input.stopWaiting(); });
const waited = await input.waited;
if (headFailed) return;
if (!waited) return input.captureBase();
await input.head;
return input.captureBase(waited);
}
export function captureSidesConcurrently(input: { plan: SourceCapturePlan; agent: boolean; injectedRunner: boolean; baselineMayExist: boolean }): boolean {
return !input.agent && !input.injectedRunner && !input.baselineMayExist
&& input.plan.headCaptureSurfaceIds.length > 0 && input.plan.baseCaptureSurfaceIds.length > 0;
@@ -111,15 +278,42 @@ export type PublicCiOptions = {
platforms?: ExplorePlatform[];
/** Every target config.yaml captures, in config order; recorded in the report so the Viewer links the sibling targets' reports. */
targets?: RunTarget[];
/** The targets with a device pose (foldables), which take the sessions `capture.maxSessions` does not divide evenly. */
posedTargets?: readonly string[];
/** The targets whose jobs repack the same app as this one, in config order (hostedRepackPeers): they share one hosted repack per commit. */
repackPeers?: readonly string[];
baseProject: string; headProject: string; pullRequestPath: string; output: string;
config: ExploreConfig; configPath?: string; environment?: NodeJS.ProcessEnv; transport?: typeof fetch;
now?: () => number; sleep?: (milliseconds: number) => void; onComment?: (state: PublicCommentState, body: string) => void;
runner?: CommandRunner;
/** Internal dependency injection; configuration cannot replace the authenticated CLI session. */
agentSessionFactory?: PrepareCiExplorationOptions["createSession"];
/** Internal dependency injection for tests: replaces the wait for a baseline run that EXPLORE_BASELINE_RUN names. */
awaitBaseline?: BaselineWait;
progress?: Progress;
/** Stops the captures between steps (a cancelled job): a session is still stopped, and a hosted repack run cancelled. */
signal?: AbortSignal;
};
/** Waits for a baseline of the base commit still being captured: its downloaded export's directory, or undefined. */
type BaselineWait = (signal: AbortSignal) => Promise<string | undefined>;
/**
* The wait for a baseline still capturing the base commit: for the run `ci:find-baseline` named in
* EXPLORE_BASELINE_RUN, which the wait checks again by the same trust rules, for up to capture.baselineWaitMinutes. A
* workflow that names no run, or a config that sets 0, captures the base as before.
*/
function pendingBaselineWait(options: Pick<PublicCiOptions, "awaitBaseline" | "config" | "target" | "platform" | "output" | "progress">,
environment: NodeJS.ProcessEnv, transport: typeof fetch, baseSha: string): BaselineWait | undefined {
if (options.awaitBaseline) return options.awaitBaseline;
const run = Number(environment.EXPLORE_BASELINE_RUN);
const minutes = options.config.capture.baselineWaitMinutes ?? DEFAULT_BASELINE_WAIT_MINUTES;
if (!Number.isSafeInteger(run) || run <= 0 || minutes === 0) return undefined;
const target = options.target ?? options.platform ?? options.config.target?.id ?? "ios";
return signal => awaitPendingBaseline({ run, sha: baseSha, target, timeoutMs: minutes * 60_000, directory: join(resolve(options.output), "baseline-download"),
environment, transport, signal, ...(options.progress ? { progress: options.progress } : {}) });
}
function assertRunIdentity(environment: NodeJS.ProcessEnv): void {
if (!/^[1-9][0-9]*$/.test(environment.GITHUB_RUN_ID ?? "") || !/^[1-9][0-9]*$/.test(environment.GITHUB_RUN_ATTEMPT ?? "")) {
throw new CiError({ code: "CI_RUN_IDENTITY_MISSING", stage: "preflight", message: "The GitHub run ID or attempt is missing.",
@@ -150,18 +344,20 @@ async function prepareCiRun(options: Pick<PublicCiOptions, "baseProject" | "head
links: viewerLinks({ repository: presentation.repository, headSha: presentation.head.sha, runId: environment.GITHUB_RUN_ID!,
runAttempt: environment.GITHUB_RUN_ATTEMPT!, configPath: repositoryConfigPath }),
startedAt: now(), updatedAt: now(), build: { platform: config.app.platform, easProfile: config.app.easProfile }, artifactOnly: true,
...(config.target ? { target: config.target } : {}), ...(options.targets && options.targets.length > 1 ? { targets: options.targets } : {}) };
...(config.target ? { target: config.target } : {}), ...(options.targets && options.targets.length > 1 ? { targets: options.targets } : {}),
...(config.report?.comment?.map ? { map: true } : {}) };
return { projects, output, state };
}
/** A review of any two commits (`capture:review --compare FILE`): the same capture as a PR's, with no PR to verify or comment on. */
export async function runCompareCi(options: Omit<PublicCiOptions, "pullRequestPath" | "transport" | "onComment"> & { comparePath: string }) {
export async function runCompareCi(options: Omit<PublicCiOptions, "pullRequestPath" | "onComment"> & { comparePath: string }) {
const environment = options.environment ?? process.env;
assertRunIdentity(environment);
const comparison = projectComparePresentation(JSON.parse(await readFile(options.comparePath, "utf8")));
const { projects, output, state } = await prepareCiRun(options, environment, comparison);
const assertCurrentHead = async () => { for (const side of ["base", "head"] as const) assertCheckout(projects[side], comparison[side].sha); };
return runCiReview({ ...options, projects, output, state, presentation: comparison,
const awaitBaseline = pendingBaselineWait(options, environment, options.transport ?? fetch, comparison.base.sha);
return runCiReview({ ...options, projects, output, state, presentation: comparison, ...(awaitBaseline ? { awaitBaseline } : {}),
environment, runId: environment.GITHUB_RUN_ID!, runAttempt: environment.GITHUB_RUN_ATTEMPT!,
assertCurrentHead, postComment: async () => {} });
}
@@ -232,7 +428,8 @@ export async function runPublicCi(options: PublicCiOptions) {
for (const side of ["base", "head"] as const) assertCheckout(projects[side], pr[side].sha);
await assertCurrentPr();
};
return runCiReview({ ...options, projects, output, state, presentation: pr,
const awaitBaseline = pendingBaselineWait(options, environment, transport, pr.base.sha);
return runCiReview({ ...options, projects, output, state, presentation: pr, ...(awaitBaseline ? { awaitBaseline } : {}),
environment, runId: environment.GITHUB_RUN_ID!, runAttempt: environment.GITHUB_RUN_ATTEMPT!,
assertCurrentHead, postComment, canPublish: () => !prVerificationFailed });
}
@@ -252,14 +449,23 @@ export async function runCiReview(options: {
platforms?: ExplorePlatform[];
/** Every target config.yaml captures; `config.target` is the one this review captures. */
targets?: RunTarget[];
/** The targets with a device pose (foldables), which take the sessions `capture.maxSessions` does not divide evenly. */
posedTargets?: readonly string[];
/** A folder shared with the other targets' reviews, so the targets of one platform build each commit once. */
buildShare?: string;
/** The targets whose matrix jobs repack the same app as this one, in config order: they share one hosted repack per commit. */
repackPeers?: readonly string[];
measure?: Parameters<typeof assessPublicEvidence>[0]["measure"];
state: PublicCommentState; environment: NodeJS.ProcessEnv;
assertCurrentHead: () => Promise<void>; postComment: () => Promise<void>; canPublish?: () => boolean;
now?: () => number; sleep?: (milliseconds: number) => void; runner?: CommandRunner; subscription?: boolean; useLocalEasSession?: boolean;
agentSessionFactory?: PrepareCiExplorationOptions["createSession"];
prepareCapture?: () => Promise<void>;
/**
* Waits for a baseline of the base commit that is still being captured, when no downloaded baseline matches. It runs
* alongside head, and base waits for it only where base would start capturing.
*/
awaitBaseline?: BaselineWait;
progress?: Progress;
/** Stops the captures between steps; model requests take it through agentSessionFactory. */
signal?: AbortSignal;
@@ -267,15 +473,18 @@ export async function runCiReview(options: {
const { projects, output, config, environment, state, assertCurrentHead, postComment } = options;
const progress = options.progress ?? silentProgress;
// The controllers are fetched before any EAS work, so a blocked download fails before a session is paid for. agent-device
// also compares the screenshots. An injected runner (tests) runs no real controller.
// also compares the screenshots. An injected runner (tests) stands in for the controllers' calls, but not for that
// comparison: it still runs the real agent-device, which fetches itself on first use.
if (!options.runner) await prefetchControllers(config, environment, progress);
const pr = options.presentation;
const now = options.now ?? Date.now;
const secretValues = Object.entries(environment).filter(([key]) => /(?:TOKEN|SECRET|PASSWORD|API_KEY)/i.test(key)).flatMap(([, value]) => value ? [value] : []);
const secretValues = environmentSecrets(environment);
let stage = "plan";
let usagePhase: AgentUsagePhase = "preflight";
let preserveReport: (() => Promise<void>) | undefined;
let reportWritten = false;
// Stops a baseline wait still polling once the review no longer needs it: done, failed, or stopped.
const stopWaiting = new AbortController();
try {
progress(`planning the review of ${pr.head.sha.slice(0, 8)} against base ${pr.base.sha.slice(0, 8)} (scope ${config.exploration.scope})`);
const plan = await buildVerifiedPlan({ baseProject: projects.base, headProject: projects.head,
@@ -306,18 +515,38 @@ export async function runCiReview(options: {
// A baseline that may be restored needs the head runtime profile first; otherwise both sides capture at once.
const baselineMayExist = plan.baseCaptureSurfaceIds.length > 0 && await captureArtifactCacheMayMatch({ ...cacheRecipes.base,
...(environment.EXPLORE_CAPTURE_CACHE ? { sourceDirectory: environment.EXPLORE_CAPTURE_CACHE } : {}), commit: pr.base.sha });
const concurrent = captureSidesConcurrently({ plan, agent: !!exploration.agent, injectedRunner: !!options.runner, baselineMayExist });
const captureSide = async (side: "head" | "base") => {
// With no finished baseline to reuse, one still being captured is waited for: from now, alongside head, and base
// waits for it only where it would start. Head is never delayed by it (brentvatne/euxy's review run 39348956443).
const waitedBaseline = !baselineMayExist && plan.baseCaptureSurfaceIds.length > 0 && options.awaitBaseline
? options.awaitBaseline(options.signal ? AbortSignal.any([stopWaiting.signal, options.signal]) : stopWaiting.signal).catch(() => undefined) : undefined;
const share = targetSessionShare(config, options.targets, options.posedTargets);
// A share of one session captures the sides in turn.
const concurrent = captureSidesConcurrently({ plan, agent: !!exploration.agent, injectedRunner: !!options.runner, baselineMayExist }) && share >= 2;
// Each side's appearances split by its own app config: a pull request can change userInterfaceStyle. Sides taken in
// turn (a baseline may be restored) each have the whole share. Passes run on worker threads, which cannot take an
// injected runner or an agent session.
const sidePasses = { head: undefined as ColorScheme[][] | undefined, base: undefined as ColorScheme[][] | undefined };
if (!exploration.agent && !options.runner) await Promise.all((["head", "base"] as const).map(async side => {
sidePasses[side] = await sideAppearancePasses(config, environment, { share, sides: concurrent ? 2 : 1, routes: plan[`${side}CaptureSurfaceIds`].length, project: projects[side] });
}));
const split = (["head", "base"] as const).filter(side => sidePasses[side]);
progress(`sessions: ${share} of ${config.capture.maxSessions ?? DEFAULT_MAX_SESSIONS} for this target; `
+ (split.length ? `${split.join(" and ")} ${split.length === 1 ? "captures its" : "capture their"} appearances in separate sessions`
: concurrent ? "one session per side" : "the sides in turn"));
const captureSide = async (side: "head" | "base", cacheDirectory = environment.EXPLORE_CAPTURE_CACHE) => {
const directory = join(output, `${side}-export`);
const selectedIds = plan[`${side}CaptureSurfaceIds`];
const cacheRecipe = cacheRecipes[side];
let reuseDirectory: string | undefined;
if (side === "base" && selectedIds.length && runtimeProfile && environment.EXPLORE_CAPTURE_CACHE) {
const restored = await restoreCaptureArtifactCache({ ...cacheRecipe, sourceDirectory: environment.EXPLORE_CAPTURE_CACHE,
if (side === "base" && selectedIds.length && runtimeProfile && cacheDirectory) {
const restored = await restoreCaptureArtifactCache({ ...cacheRecipe, sourceDirectory: cacheDirectory,
commit: pr.base.sha, selectedIds, directory: join(output, "base-cache"), runtimeProfile,
...(exploration.agent ? { captureContexts: exploration.agent.contexts } : {}) });
if (restored) reuseDirectory = join(output, "base-cache");
progress(restored ? `base: found the baseline captures of ${pr.base.sha.slice(0, 8)}` : `base: no compatible baseline of ${pr.base.sha.slice(0, 8)}; capturing the base too`);
const incomplete = restored ? undefined : await incompleteCaptureReason(cacheDirectory, pr.base.sha);
progress(restored ? `base: found the baseline captures of ${pr.base.sha.slice(0, 8)}`
: incomplete ? `base: the baseline of ${pr.base.sha.slice(0, 8)} has nothing to reuse (${incomplete}); capturing the base too`
: `base: no compatible baseline of ${pr.base.sha.slice(0, 8)}; capturing the base too`);
}
const mapOptions = { ...captureOptions(config, environment, options.runner, { now: options.now, sleep: options.sleep }), progress,
...(options.useLocalEasSession ? { useLocalEasSession: true } : {}),
@@ -325,10 +554,10 @@ export async function runCiReview(options: {
mapDirectory: join(output, `${side}-map`), outputDirectory: directory,
name: config.app.name ?? pr.repository, buildId: pr.origin === "local" ? `local-${pr.runId}-${side}` : pr.origin === "compare" ? `compare-${options.runId}-${side}` : `pr-${pr.number}-${side}`, buildVersion: pr[side].sha.slice(0, 8), gitCommit: pr[side].sha,
planPath: join(output, "execution-plan.json"), side, sessionStatePath: join(output, `${side}-session`), ...(options.buildShare ? { buildShare: options.buildShare } : {}),
...repackShareOption(config, environment, options.repackPeers),
...(reuseDirectory ? { reuseDirectory } : {}), ...(options.signal ? { signal: options.signal } : {}) };
const captured = concurrent
? await runCiMapOnWorker((({ progress: _progress, runner: _runner, agent: _agent, signal: _signal, ...rest }) => rest)(mapOptions), progress, options.signal)
: await runCiMap(mapOptions);
// A restored baseline already holds every appearance pass, so it never splits.
const captured = await captureSideInSessions(mapOptions, reuseDirectory ? undefined : sidePasses[side], progress, options.signal, concurrent);
const bundle = await readCaptureBundle(directory);
bundle.profile = cacheRecipe.profile;
await writeFile(join(directory, "bundle.json"), JSON.stringify(bundle));
@@ -346,12 +575,17 @@ export async function runCiReview(options: {
}
await writeCaptureArtifactCache(directory, cacheRecipe);
};
const captureBase = (head: Promise<void>) => captureBaseAfterWait({ waited: waitedBaseline, head, stopWaiting: () => stopWaiting.abort(),
captureBase: cacheDirectory => captureSide("base", cacheDirectory) });
if (waitedBaseline) progress(`base waits for the baseline run of ${pr.base.sha.slice(0, 8)} still being captured; head captures meanwhile`);
if (concurrent) {
stage = "capture-head"; state.phase = "capturing"; state.capture.current = "head"; usagePhase = "capture-head";
progress("capturing head and base at the same time in two simulator sessions");
progress(waitedBaseline ? "capturing head; base captures in a second simulator session if the baseline cannot be reused"
: "capturing head and base at the same time in two simulator sessions");
await postComment();
// Let both sides finish or fail on their own so every started session is stopped by its owner.
const settled = await Promise.allSettled([captureSide("head"), captureSide("base")]);
const head = captureSide("head");
const settled = await Promise.allSettled([head, captureBase(head)]);
for (const [index, outcome] of settled.entries()) {
if (outcome.status === "rejected") { stage = index === 0 ? "capture-head" : "capture-base"; throw outcome.reason; }
}
@@ -360,15 +594,16 @@ export async function runCiReview(options: {
stage = `capture-${side}`; state.phase = "capturing"; state.capture.current = side;
usagePhase = `capture-${side}`;
await postComment();
await captureSide(side);
await (side === "head" ? captureSide("head") : captureBase(Promise.resolve()));
}
}
delete state.capture.current;
stage = "compare";
progress("comparing base and head captures");
const sides = { base: await evidenceSide(join(output, "base-export")), head: await evidenceSide(join(output, "head-export")) };
const evidence = await assessPublicEvidence({ platform: config.app.platform, plan, ...sides, ...(options.measure ? { measure: options.measure } : {}), profile: `${config.app.easProfile}:${config.app.bundleId}`, agentDeviceCommand: agentDeviceLauncher(options.environment) });
const result = await prepareCaptureReview({ plan, pullRequest: pr, baseDirectory: sides.base.directory, headDirectory: sides.head.directory,
const system = systemRegions({ platform: config.app.platform, device: config.device.name, pose: config.device.pose });
const evidence = await assessPublicEvidence({ platform: config.app.platform, systemRegions: system, plan, ...sides, ...(options.measure ? { measure: options.measure } : {}), profile: `${config.app.easProfile}:${config.app.bundleId}`, agentDeviceCommand: agentDeviceLauncher(options.environment) });
const result = await prepareCaptureReview({ plan, pullRequest: pr, baseDirectory: sides.base.directory, headDirectory: sides.head.directory, systemRegions: system,
runId: options.runId, runAttempt: options.runAttempt, assertCurrentHead, assessments: evidence.surfaces,
...(options.platforms ? { platforms: options.platforms } : {}),
...(config.target ? { target: config.target } : {}), ...(options.targets ? { targets: options.targets } : {}),
@@ -456,16 +691,26 @@ export async function runCiReview(options: {
try { await postComment(); } catch { /* A stale PR or inaccessible GitHub must not replace the original failure. */ }
}
throw error;
} finally {
stopWaiting.abort();
}
}
export async function runBaselineCi(options: { platform?: ExplorePlatform; project: string; sha: string; output: string; config: ExploreConfig; environment?: NodeJS.ProcessEnv; runner?: CommandRunner;
/** Every target config.yaml captures, so this baseline takes its target's share of `capture.maxSessions`. */
targets?: RunTarget[];
/** The targets with a device pose (foldables), which take the sessions `capture.maxSessions` does not divide evenly. */
posedTargets?: readonly string[];
/** The targets whose baseline jobs repack the same app as this one, in config order: they share one hosted repack. */
repackPeers?: readonly string[];
now?: () => number; sleep?: (milliseconds: number) => void;
agentSessionFactory?: PrepareCiExplorationOptions["createSession"]; progress?: Progress;
/** Use the local `eas login` session instead of EXPO_TOKEN (expo-explore doctor --capture). */
useLocalEasSession?: boolean;
/** Agent mode with the local claude or codex login instead of the provider's API key. */
subscription?: boolean }) {
subscription?: boolean;
/** Stops the capture between steps (a cancelled job): its session is still stopped, and a hosted repack run cancelled. */
signal?: AbortSignal }) {
const environment = options.environment ?? process.env;
assertCheckout(options.project, options.sha);
if (!options.runner) await prefetchControllers(options.config, environment, options.progress ?? silentProgress);
@@ -485,12 +730,20 @@ export async function runBaselineCi(options: { platform?: ExplorePlatform; proje
onUsage: recordAgentUsage(output, () => usagePhase), ...(options.subscription ? { subscription: true } : {}) });
const directory = join(output, "baseline-export");
usagePhase = "capture-baseline";
const result = await runCiMap({ ...captureOptions(options.config, environment, options.runner, { now: options.now, sleep: options.sleep }), progress,
const baselineOptions: CiMapOptions = { ...captureOptions(options.config, environment, options.runner, { now: options.now, sleep: options.sleep }), progress,
...(options.useLocalEasSession ? { useLocalEasSession: true } : {}),
...(exploration.agent ? { agent: exploration.agent } : {}), project: options.project,
mapDirectory: join(output, "baseline-map"), outputDirectory: directory, name: options.config.app.name ?? options.config.app.bundleId,
buildId: `baseline-${options.sha.slice(0, 12)}`, buildVersion: options.sha.slice(0, 8), gitCommit: options.sha,
side: "head", planPath: join(output, "execution-plan.json"), sessionStatePath: join(output, "baseline-session") });
side: "head", planPath: join(output, "execution-plan.json"), sessionStatePath: join(output, "baseline-session"),
...repackShareOption(options.config, environment, options.repackPeers), ...(options.signal ? { signal: options.signal } : {}) };
const share = targetSessionShare(options.config, options.targets, options.posedTargets);
// Passes on worker threads cannot take an injected runner or an agent session.
const passes = !options.runner && !exploration.agent
? await sideAppearancePasses(options.config, environment, { share, sides: 1, routes: plan.headCaptureSurfaceIds.length, project: options.project }) : undefined;
progress(`sessions: ${share} of ${options.config.capture.maxSessions ?? DEFAULT_MAX_SESSIONS} for this target; `
+ (passes ? `capturing ${passes.map(pass => pass.join("+")).join(" and ")} in separate sessions` : "one session"));
const result = await captureSideInSessions(baselineOptions, passes, progress, options.signal, passes !== undefined);
const cacheRecipe = await recipe(options.config, plan, "head", environment, exploration.instructionsSha256);
if (options.subscription) cacheRecipe.toolsCommit += ":local-subscription";
const bundle = await readCaptureBundle(directory); bundle.profile = cacheRecipe.profile;
+70
View File
@@ -0,0 +1,70 @@
import type { ColorScheme } from "../capture/appearance.js";
/** One route capture on a remote session, with the bridge thread (observed ~5 s; 9.4 s before it). */
export const ROUTE_SECONDS = 5;
/** Observing the app launch once per appearance. */
export const LAUNCH_SECONDS = 28;
/** A pass moves to its own session only when that shortens the side by at least this much. */
export const SPLIT_MIN_SAVING_SECONDS = 60;
/**
* `capture.maxSessions` when the config sets none: the simulators a typical EAS plan runs at once. A starter plan runs
* one per platform, so its config sets the ceiling lower (2 for an iOS and an Android target).
*/
export const DEFAULT_MAX_SESSIONS = 5;
export type SessionShareInput = {
/** `capture.maxSessions`: the review-wide ceiling; absent means DEFAULT_MAX_SESSIONS. */
maxSessions: number | undefined;
/** The review's target ids in config order, and the one this job captures. */
targets: readonly string[];
target: string;
/** Targets a split pays off most for take the sessions that do not divide evenly: foldables and large screens first. */
prefer?: (target: string) => number;
};
/**
* This job's share of the review's session ceiling: the ceiling divided over the targets, the remainder going to the
* preferred targets (foldables) in config order, and unused when there are none. Never below 1 (a side then captures after the other): under a ceiling below the
* number of targets, at most that many target jobs run at once (maxParallelTargets).
*/
export function sessionShare(input: SessionShareInput): number {
const count = Math.max(1, input.targets.length);
const ceiling = input.maxSessions ?? DEFAULT_MAX_SESSIONS;
const base = Math.floor(ceiling / count);
let remainder = ceiling - base * count;
// A session left over goes to a preferred target only; without one it stays unused (two targets of five: two each).
const ranked = input.targets.filter(id => (input.prefer?.(id) ?? 0) > 0)
.sort((a, b) => input.prefer!(b) - input.prefer!(a) || input.targets.indexOf(a) - input.targets.indexOf(b));
const extra = new Set<string>();
for (const id of ranked) { if (remainder <= 0) break; extra.add(id); remainder -= 1; }
return Math.max(1, base + (extra.has(input.target) ? 1 : 0));
}
/** How many target jobs one review runs at once: each needs a session, so never more than the ceiling. */
export function maxParallelTargets(maxSessions: number | undefined, targets: number): number {
return Math.max(1, Math.min(targets, maxSessions ?? DEFAULT_MAX_SESSIONS));
}
export type SidePlan = {
/** Whether head and base capture at the same time. */
concurrentSides: boolean;
/** The appearance passes of one side, each on its own session; one entry means the side keeps a single session. */
passes: readonly (readonly ColorScheme[])[];
};
/**
* How one target's job spends its share of sessions: both sides at once when it has two, and a side's appearance passes
* in separate sessions only when the pass moved out shortens the side by at least a minute (a session costs its fixed
* boot and install time, so a few routes are not worth one). Sessions the work cannot fill stay unused: the ceiling is
* a limit, not a target.
*/
export function planSideSessions(input: { share: number; sides: 1 | 2; routes: number; schemes: readonly ColorScheme[];
minSavingSeconds?: number; routeSeconds?: number; launchSeconds?: number }): SidePlan {
const minSaving = input.minSavingSeconds ?? SPLIT_MIN_SAVING_SECONDS, perRoute = input.routeSeconds ?? ROUTE_SECONDS, launch = input.launchSeconds ?? LAUNCH_SECONDS;
const concurrentSides = input.sides === 1 || input.share >= 2;
const perSide = Math.floor(input.share / (concurrentSides ? input.sides : 1));
const schemes = [...input.schemes];
const passSeconds = input.routes * perRoute + launch;
const splitAppearances = schemes.length > 1 && perSide >= schemes.length && passSeconds >= minSaving;
return { concurrentSides, passes: splitAppearances ? schemes.map(scheme => [scheme]) : [schemes] };
}
+72 -19
View File
@@ -30,7 +30,7 @@ export async function writeCaptureOutputs(inputPath: string, environment: Enviro
await writeStepOutputs(environment, { preview: (await readCiReportInput(inputPath)).preview?.hosting ?? "" });
}
function githubContext(environment: Environment) {
export function githubContext(environment: Environment) {
const repository = environment.GITHUB_REPOSITORY ?? "";
const token = environment.GITHUB_TOKEN?.trim();
if (!/^[A-Za-z0-9_.-]+\/[A-Za-z0-9_.-]+$/.test(repository)) throw new CiError({ code: "CI_RUN_IDENTITY_MISSING", stage: "preflight",
@@ -41,7 +41,7 @@ function githubContext(environment: Environment) {
}
/** One GitHub REST read; 404 is an answer (undefined), every other failure stops the step with what to do. */
async function githubGet<T>(path: string, context: { token: string }, transport: typeof fetch): Promise<T | undefined> {
export async function githubGet<T>(path: string, context: { token: string }, transport: typeof fetch): Promise<T | undefined> {
let response: Response;
try {
response = await transport(`https://api.github.com/${path}`, { redirect: "error", signal: AbortSignal.timeout(30_000),
@@ -102,57 +102,109 @@ async function trustedAuthors(context: { repository: string; token: string }, tr
catch (error) { return { logins: [], invalid: error instanceof Error ? error.message : String(error) }; }
}
type GithubArtifact = { id: number; expired: boolean; workflow_run?: { id?: number } };
type GithubRun = { conclusion: string | null; workflow_id: number; head_sha: string; head_branch: string; event: string; head_repository?: { full_name?: string } };
export type GithubArtifact = { id: number; name?: string; expired: boolean; workflow_run?: { id?: number } };
export type GithubRun = { id: number; status?: string; conclusion: string | null; workflow_id: number; head_sha: string; head_branch: string; event: string;
head_repository?: { full_name?: string } };
/**
* `ci:find-baseline`: the newest successful baseline-workflow artifact of exactly this commit, uploaded from the
* default branch of this repository. Its id and run go to the step outputs for actions/download-artifact.
* What every baseline lookup starts from: the GitHub context, a checked commit and target, the default branch and the
* baseline workflow's id, and the artifact names a baseline of this target may carry. Undefined when the repository
* has no baseline workflow.
*/
export async function findBaselineArtifact(options: { sha: string; target?: string;
/** What callers from before targets pass: a platform is the id of its default target. */ platform?: string; environment?: Environment; transport?: typeof fetch }): Promise<{ id: number; run: number } | undefined> {
export async function baselineLookup(options: { sha: string; target?: string; platform?: string; environment?: Environment; transport?: typeof fetch;
/** Wraps each read, as a wait that polls for minutes retries GitHub's transient failures. */ retry?: <T>(read: () => Promise<T>) => Promise<T> }) {
const environment = options.environment ?? process.env; const transport = options.transport ?? fetch;
const context = githubContext(environment);
if (!SHA.test(options.sha)) throw new CiError({ code: "SHA_INVALID", stage: "preflight", message: "The base commit must be a full SHA.", nextAction: "Pass --sha with a 40-character commit SHA." });
const target = options.target ?? options.platform ?? "ios";
if (!isTargetId(target)) throw new CiError({ code: "TARGET_INVALID", stage: "preflight", message: `${target} is not a target id.`,
nextAction: "Pass --target with an id from targets: in .expo-explore/config.yaml (ios or android without targets:)." });
const repository = await githubGet<{ default_branch: string }>(`repos/${context.repository}`, context, transport);
const workflow = await githubGet<{ id: number }>(`repos/${context.repository}/actions/workflows/explore-baseline.yml`, context, transport);
const read = <T>(path: string) => githubGet<T>(path, context, transport);
const get = <T>(path: string) => options.retry ? options.retry(() => read<T>(path)) : read<T>(path);
const repository = await get<{ default_branch: string }>(`repos/${context.repository}`);
const workflow = await get<{ id: number }>(`repos/${context.repository}/actions/workflows/explore-baseline.yml`);
if (!repository || !workflow) return undefined;
// Baselines carry the target suffix; target ios also accepts the unsuffixed name older single-platform baseline workflows uploaded.
const names = [`explore-baseline-${options.sha}-${target}`, ...(target === "ios" ? [`explore-baseline-${options.sha}`] : [])];
return { context, get, target, sha: options.sha, defaultBranch: repository.default_branch, workflow: workflow.id, names };
}
/**
* Whether a run may supply this commit's baseline: a run of the baseline workflow for exactly this commit, on the
* default branch of this repository, started by a push or a manual dispatch. A finished run's artifact is reused, and
* a run still capturing is waited for, only when it passes this check; nothing from a pull request's head selects it.
*/
export function trustedBaselineRun(run: GithubRun | undefined, lookup: { workflow: number; sha: string; defaultBranch: string; context: { repository: string } }): run is GithubRun {
return !!run && run.workflow_id === lookup.workflow && run.head_sha === lookup.sha && run.head_branch === lookup.defaultBranch
&& ["push", "workflow_dispatch"].includes(run.event) && run.head_repository?.full_name === lookup.context.repository;
}
/**
* `ci:find-baseline`: the newest baseline-workflow artifact of exactly this commit, uploaded from the default branch of
* this repository by a finished run. Its id and run go to the step outputs for actions/download-artifact. A run that
* failed counts: a target's baseline job uploads its artifact only after every step before it passed, so another
* target's failure (one flaky simulator session, as in brentvatne/euxy#172) does not make this target's baseline
* unusable. A run still in progress or cancelled does not; findPendingBaselineRun finds one still capturing.
*/
export async function findBaselineArtifact(options: { sha: string; target?: string;
/** What callers from before targets pass: a platform is the id of its default target. */ platform?: string; environment?: Environment; transport?: typeof fetch }): Promise<{ id: number; run: number } | undefined> {
const lookup = await baselineLookup(options);
if (!lookup) return undefined;
const { context, get } = lookup;
const artifacts: GithubArtifact[] = [];
for (const name of names) for (let page = 1; page <= 10; page++) {
const listed = await githubGet<{ artifacts: GithubArtifact[] }>(`repos/${context.repository}/actions/artifacts?name=${encodeURIComponent(name)}&per_page=100&page=${page}`, context, transport);
for (const name of lookup.names) for (let page = 1; page <= 10; page++) {
const listed = await get<{ artifacts: GithubArtifact[] }>(`repos/${context.repository}/actions/artifacts?name=${encodeURIComponent(name)}&per_page=100&page=${page}`);
artifacts.push(...listed?.artifacts ?? []);
if ((listed?.artifacts.length ?? 0) < 100) break;
}
for (const artifact of artifacts.filter(candidate => !candidate.expired).sort((a, b) => b.id - a.id)) {
if (!artifact.workflow_run?.id) continue;
const run = await githubGet<GithubRun>(`repos/${context.repository}/actions/runs/${artifact.workflow_run.id}`, context, transport);
if (!run || run.conclusion !== "success" || run.workflow_id !== workflow.id || run.head_sha !== options.sha
|| run.head_branch !== repository.default_branch || !["push", "workflow_dispatch"].includes(run.event)
|| run.head_repository?.full_name !== context.repository) continue;
const run = await get<GithubRun>(`repos/${context.repository}/actions/runs/${artifact.workflow_run.id}`);
if (!trustedBaselineRun(run, lookup) || (run.conclusion !== "success" && run.conclusion !== "failure")) continue;
return { id: artifact.id, run: artifact.workflow_run.id };
}
return undefined;
}
/**
* A baseline run of exactly this commit that has not finished yet (queued, waiting or in progress), passing the same
* trust check as a finished one. In brentvatne/euxy's review run 39348956443 main's baseline of the base commit had
* started 15 s before the review, so no artifact existed yet. A run already capturing is preferred to one queued
* behind it in the baseline's concurrency group, since it finishes first; then the newest.
*/
export async function findPendingBaselineRun(options: Parameters<typeof baselineLookup>[0]): Promise<{ run: number } | undefined> {
const lookup = await baselineLookup(options);
if (!lookup) return undefined;
const listed = await lookup.get<{ workflow_runs?: GithubRun[] }>(`repos/${lookup.context.repository}/actions/workflows/${lookup.workflow}/runs?head_sha=${lookup.sha}`
+ `&branch=${encodeURIComponent(lookup.defaultBranch)}&per_page=100`);
const pending = (listed?.workflow_runs ?? []).filter(run => run.status !== "completed" && Number.isSafeInteger(run.id) && run.id > 0 && trustedBaselineRun(run, lookup))
.sort((a, b) => Number(b.status === "in_progress") - Number(a.status === "in_progress") || b.id - a.id);
return pending[0] ? { run: pending[0].id } : undefined;
}
/**
* `ci:find-baseline` as the workflow runs it: a baseline is optional, so a lookup that fails (a GitHub 5xx, a rate
* limit) is a warning and no baseline, never an error. A `ci:*` error would write explore-diagnostics/error.json,
* which the failure comment would then report as the cause of an unrelated failure, possibly on another target.
* With no finished baseline, it names a trusted baseline run still capturing the commit as `pending`, which
* `capture:review` waits for alongside head.
*/
export async function findOptionalBaseline(options: Parameters<typeof findBaselineArtifact>[0]):
Promise<{ baseline?: { id: number; run: number }; warning?: { code: string; message: string } }> {
Promise<{ baseline?: { id: number; run: number }; pending?: { run: number }; warning?: { code: string; message: string } }> {
try {
const baseline = await findBaselineArtifact(options);
return baseline ? { baseline } : {};
if (baseline) return { baseline };
} catch (error) {
return { warning: { code: error instanceof CiError ? error.code : "BASELINE_LOOKUP_FAILED",
message: `No baseline is reused, so capture captures the base too: ${error instanceof Error ? error.message : String(error)}` } };
}
try {
const pending = await findPendingBaselineRun(options);
return pending ? { pending } : {};
} catch (error) {
return { warning: { code: error instanceof CiError ? error.code : "BASELINE_LOOKUP_FAILED",
message: `No baseline run in progress is waited for, so capture captures the base too: ${error instanceof Error ? error.message : String(error)}` } };
}
}
/**
@@ -196,7 +248,8 @@ export async function cleanupAfterCapture(options: { output: string; checkouts:
const environment = options.environment ?? process.env;
const result: { stopped: string[]; failures: Array<{ code: string; message: string; nextAction: string }> } = { stopped: [], failures: [] };
const sessionFiles: string[] = [];
for (const name of ["head-session", "base-session", "baseline-session"]) {
// A side that captures an appearance in a session of its own (capture.maxSessions) records it as `<side>-session-<scheme>`.
for (const name of ["head", "base", "baseline"].flatMap(side => ["", "-light", "-dark"].map(scheme => `${side}-session${scheme}`))) {
const path = resolve(options.output, name);
try { await lstat(path); sessionFiles.push(path); } catch { /* That side never started a session. */ }
}
+4
View File
@@ -5,6 +5,7 @@ import { resolve } from "node:path";
import { ExploreStore } from "./map/store.js";
import { QUERY_CONTRACT } from "./types.js";
import { EXPLORE_VERSION } from "./shared/package-paths.js";
import { isTargetId } from "./config/explore-config.js";
import { command, resultContract, argumentError, validateCommandArguments, flag, has, setStore } from "./cli/args.js";
import { help } from "./cli/help.js";
import { listScreens, getScreen, listElements, listObservations, getObservation, nextFrontier, listAttempts, getAttempt, listTransitions, getTransition, neighbors, pathQuery, listJourneys, getJourney, listFrontier, listGaps, summary, search, status, validate, exportArtifacts } from "./cli/map-commands.js";
@@ -81,8 +82,11 @@ try {
await handler();
} catch (error) {
if (command.startsWith("capture:") || command.startsWith("source:") || command.startsWith("ci:") || command.startsWith("config:")) {
// The failure comment names the target from this: a capture, report or publication of one matrix target.
const target = (() => { try { return flag("target"); } catch { return undefined; } })();
const diagnostic = await reportCiError(ciErrorForCommand(error, command), {
...(process.env.RUNNER_TEMP ? { diagnosticsPath: resolve(process.env.RUNNER_TEMP, "explore-diagnostics", "error.json") } : {}),
...(isTargetId(target) ? { target } : {}),
...(has("json") ? { write: (line: string) => { if (process.env.GITHUB_ACTIONS === "true") console.log(line); } } : {}),
});
if (has("json")) console.error(JSON.stringify({ contract: resultContract, command, ok: false, error: diagnostic.message, diagnostic }));
+3 -3
View File
@@ -90,10 +90,10 @@ const flagKinds: Record<string, Record<string, "value" | "boolean">> = {
"explore:binary": { archive:"value", output:"value", platform:"value", controller:"value", device:"value", "bundle-id":"value", "launch-url":"value", hints:"value", "allow-destructive":"value",
"explore:binary": { replay:"value", archive:"value", output:"value", platform:"value", controller:"value", simulator:"value", device:"value", "bundle-id":"value", "launch-url":"value", hints:"value", "allow-destructive":"value",
"max-duration-minutes":"value", "max-screens":"value", "max-actions":"value", provider:"value", model:"value", owner:"value", project:"value",
upload:"boolean", "no-upload":"boolean", "expires-in":"value", alias:"value" },
"explore:session": { session:"value", "env-file":"value", "bundle-id":"value", "app-name":"value", "app-version":"value", output:"value", platform:"value", controller:"value", device:"value", "launch-url":"value", hints:"value",
upload:"boolean", "no-upload":"boolean", "expires-in":"value", alias:"value", "no-open":"boolean", port:"value" },
"explore:session": { replay:"value", session:"value", "env-file":"value", "bundle-id":"value", "app-name":"value", "app-version":"value", output:"value", platform:"value", controller:"value", device:"value", "launch-url":"value", hints:"value",
"allow-destructive":"value", "max-duration-minutes":"value", "max-screens":"value", "max-actions":"value", provider:"value", model:"value", owner:"value", project:"value",
upload:"boolean", "expires-in":"value", alias:"value", "no-open":"boolean", port:"value" },
"ci:report": { input:"value", output:"value", target:"value", platform:"value" },
+23 -19
View File
@@ -5,18 +5,10 @@ import { openViewerInBrowser, startViewerServer, waitForViewerShutdown } from ".
import { writeExplorationReport } from "../binary/report.js";
import { assertHostingExport, deployExplorationReport, zipReportDirectory } from "../report/preview-deploy.js";
import { keepReports } from "./report-commands.js";
import { warnIgnoredExpiry } from "../config/explore-config.js";
import { argumentError, flag, flags, has, nonNegativeInteger, oneOf, output, positional, positiveInteger } from "./args.js";
import { readReplayPlan } from "../binary/replay.js";
const MAX_EXPIRY_DAYS = 90;
/** `7d`, `30d`, `never`: how long the hosted link serves the report. */
export function expiresInDays(value: string): number | null {
if (value === "never") return null;
const match = /^(\d{1,3})d$/.exec(value);
const days = match ? Number(match[1]) : NaN;
if (!Number.isSafeInteger(days) || days < 1 || days > MAX_EXPIRY_DAYS) argumentError(`--expires-in must be 1d to ${MAX_EXPIRY_DAYS}d, or never (got ${value})`, "expires-in");
return days;
}
/** `--alias NAME` plus a fixed token from EXPLORE_PREVIEW_TOKEN: the hosted link stays the same from one run to the next (validated by the deploy). */
function stableLinkOptions(): { alias?: string; token?: string } {
@@ -35,16 +27,21 @@ export async function exploreBinary(): Promise<void> {
"Run expo-explore explore:binary <path to .apk, .app, .tar.gz, or simulator .ipa>.");
if (has("upload") && has("no-upload")) argumentError("Pass either --upload or --no-upload", "no-upload");
const outputDirectory = resolve(process.cwd(), flag("output") ?? defaultOutputDirectory(archive));
const expiresAfterDays = expiresInDays(flag("expires-in") ?? "7d");
if (has("expires-in")) warnIgnoredExpiry("--expires-in");
const platform = flag("platform") === undefined ? undefined : oneOf("platform", ["ios", "android"] as const);
const provider = flag("provider") === undefined ? undefined : oneOf("provider", ["anthropic", "openai"] as const);
const controller = flag("controller") === undefined ? undefined : oneOf("controller", ["agent-device", "argent"] as const);
const simulator = flag("simulator") === undefined ? "eas" : oneOf("simulator", ["eas", "local"] as const);
// On EAS Simulator the report is hosted unless --no-upload; locally it opens in the Viewer unless --upload, so a
// local run needs no Expo account at all.
const upload = simulator === "local" ? has("upload") : !has("no-upload");
const port = flag("port") === undefined ? undefined : nonNegativeInteger("port");
const started = Date.now();
const say = (message: string) => { process.stderr.write(`[explore ${Math.round((Date.now() - started) / 1000)}s] ${message}\n`); };
const allowDestructive = flags("allow-destructive").flatMap(value => value.split(",")).map(value => value.trim()).filter(Boolean);
if (!has("no-upload")) await assertHostingExport();
if (upload) await assertHostingExport();
const result = await runBinaryExploration({
archive: resolve(process.cwd(), archive), output: outputDirectory, progress: say,
archive: resolve(process.cwd(), archive), output: outputDirectory, progress: say, simulator,
...(platform ? { platform } : {}), ...(provider ? { provider } : {}), ...(controller ? { controller } : {}),
...(flag("device") ? { device: flag("device")! } : {}), ...(flag("bundle-id") ? { bundleId: flag("bundle-id")! } : {}),
...(flag("launch-url") ? { launchUrl: flag("launch-url")! } : {}), ...(flag("hints") ? { hints: flag("hints")! } : {}),
@@ -53,20 +50,22 @@ export async function exploreBinary(): Promise<void> {
...(flag("max-duration-minutes") ? { maxDurationMinutes: positiveInteger("max-duration-minutes") } : {}),
...(flag("max-screens") ? { maxScreens: positiveInteger("max-screens") } : {}),
...(flag("max-actions") ? { maxActions: positiveInteger("max-actions") } : {}),
...(flag("replay") ? { replay: await readReplayPlan(resolve(process.cwd(), flag("replay")!)) } : {}),
});
say(`exploration ${result.exploration.stopReason}: ${result.exploration.screens} screens, ${result.exploration.observations} captures, ${result.exploration.actions} actions`);
const report = await writeExplorationReport(result, join(outputDirectory, "report"));
await keepReports(async () => [{ platform: result.archive.platform, zip: await zipReportDirectory(report.reportPath) }]);
let hosted: Awaited<ReturnType<typeof deployExplorationReport>> | undefined;
if (!has("no-upload")) {
if (upload) {
say("hosting the report on EAS Hosting");
hosted = await deployExplorationReport({ reportPath: report.reportPath, platform: result.archive.platform, project: flag("project") ?? DEFAULT_BINARY_PROJECT, owner: result.session.owner, expiresAfterDays, ...stableLinkOptions() });
const owner = result.session.owner || flag("owner");
hosted = await deployExplorationReport({ reportPath: report.reportPath, platform: result.archive.platform, project: flag("project") ?? DEFAULT_BINARY_PROJECT, ...(owner ? { owner } : {}), ...stableLinkOptions() });
say(`report: ${hosted.previewUrl}${hosted.expiresAt ? ` (expires ${hosted.expiresAt})` : ""}`);
}
const summary = { ok: true,
app: { name: result.archive.appName, bundleId: result.archive.bundleId, platform: result.archive.platform, ...(result.archive.version ? { version: result.archive.version } : {}) },
archive: { path: result.archive.path, kind: result.archive.kind, sha256: result.archive.sha256, bytes: result.archive.bytes },
device: { model: result.runtime.deviceModel, osVersion: result.runtime.osVersion }, session: { id: result.session.id, owner: result.session.owner, project: result.session.project },
device: { model: result.runtime.deviceModel, osVersion: result.runtime.osVersion }, session: { id: result.session.id, simulator, ...(simulator === "eas" ? { owner: result.session.owner, project: result.session.project } : {}) },
model: result.model, exploration: { stopReason: result.exploration.stopReason, ...(result.exploration.summary ? { summary: result.exploration.summary } : {}),
screens: result.exploration.screens, observations: result.exploration.observations, actions: result.exploration.actions, durationMs: result.durationMs },
map: result.mapPath, report: report.reportPath, ...(hosted ? { url: hosted.previewUrl, expiresAt: hosted.expiresAt, deploymentUrl: hosted.deploymentUrl, ...(hosted.aliasUrl ? { aliasUrl: hosted.aliasUrl } : {}) } : {}),
@@ -75,7 +74,11 @@ export async function exploreBinary(): Promise<void> {
console.log(`${summary.app.name} (${summary.app.bundleId}, ${summary.app.platform}) explored on ${summary.device.model} ${summary.device.osVersion}: ${summary.exploration.screens} screens, ${summary.exploration.actions} actions, stopped because ${summary.exploration.stopReason}.`);
if (hosted) console.log(`Hosted report: ${hosted.previewUrl}${hosted.expiresAt ? ` (expires ${hosted.expiresAt})` : ""}`);
console.log(`Local report: ${report.reportPath}`);
console.log(`Next: ${summary.next.join(" | ")}`);
if (simulator === "eas" || hosted || has("no-open")) { console.log(`Next: ${summary.next.join(" | ")}`); return; }
const viewer = await startViewerServer({ root: outputDirectory, ...(port === undefined ? {} : { port }) });
console.log(`Viewer: ${viewer.url} (Ctrl+C to stop)`);
openViewerInBrowser(viewer.url);
await waitForViewerShutdown(viewer);
}
/** `--env-file` wins; otherwise the `.env.eas-simulator` that `eas simulator:start --out-config-type dotenv` wrote in the current directory, unless the environment already carries a session. */
@@ -91,7 +94,7 @@ export async function exploreSession(): Promise<void> {
const session = positional() ?? flag("session");
const bundleId = flag("bundle-id");
const outputDirectory = resolve(process.cwd(), flag("output") ?? defaultOutputDirectory(bundleId ?? "session"));
const expiresAfterDays = expiresInDays(flag("expires-in") ?? "7d");
if (has("expires-in")) warnIgnoredExpiry("--expires-in");
const platform = flag("platform") === undefined ? undefined : oneOf("platform", ["ios", "android"] as const);
const provider = flag("provider") === undefined ? undefined : oneOf("provider", ["anthropic", "openai"] as const);
const controller = flag("controller") === undefined ? undefined : oneOf("controller", ["agent-device", "argent"] as const);
@@ -112,6 +115,7 @@ export async function exploreSession(): Promise<void> {
...(flag("max-duration-minutes") ? { maxDurationMinutes: positiveInteger("max-duration-minutes") } : {}),
...(flag("max-screens") ? { maxScreens: positiveInteger("max-screens") } : {}),
...(flag("max-actions") ? { maxActions: positiveInteger("max-actions") } : {}),
...(flag("replay") ? { replay: await readReplayPlan(resolve(process.cwd(), flag("replay")!)) } : {}),
});
say(`exploration ${result.exploration.stopReason}: ${result.exploration.screens} screens, ${result.exploration.observations} captures, ${result.exploration.actions} actions`);
const report = await writeExplorationReport(result, join(outputDirectory, "report"));
@@ -119,7 +123,7 @@ export async function exploreSession(): Promise<void> {
let hosted: Awaited<ReturnType<typeof deployExplorationReport>> | undefined;
if (has("upload")) {
say("hosting the report on EAS Hosting");
hosted = await deployExplorationReport({ reportPath: report.reportPath, platform: result.app.platform, project: flag("project") ?? DEFAULT_BINARY_PROJECT, ...(flag("owner") ? { owner: flag("owner")! } : {}), expiresAfterDays, ...stableLinkOptions() });
hosted = await deployExplorationReport({ reportPath: report.reportPath, platform: result.app.platform, project: flag("project") ?? DEFAULT_BINARY_PROJECT, ...(flag("owner") ? { owner: flag("owner")! } : {}), ...stableLinkOptions() });
say(`report: ${hosted.previewUrl}${hosted.expiresAt ? ` (expires ${hosted.expiresAt})` : ""}`);
}
const summary = { ok: true, app: result.app, device: { model: result.runtime.deviceModel, osVersion: result.runtime.osVersion },
+50 -14
View File
@@ -15,12 +15,13 @@ import { installCiFiles } from "../ci/workflow-install.js";
import { readRouteEntry } from "../ci/route-entry.js";
import { buildVerifiedPlan } from "../ci/verified-plan.js";
import { formatCaptureCheck, formatDoctorReport, referencedSecretNames, runCaptureCheck, runDoctor } from "../ci/doctor.js";
import { readExploreConfig, readExploreInstructions, resolveTargetConfig, configTargets, isTargetId, withRunOverrides, previewDays, configPlatforms as listPlatforms, EXPLORE_PLATFORMS, DEFAULT_EXPLORE_CONFIG_PATH, PREVIEW_DEFAULT_DAYS, type ExplorePlatform } from "../config/explore-config.js";
import { readExploreConfig, readExploreInstructions, resolveTargetConfig, configTargets, posedTargets, hostedRepackPeers, isTargetId, withRunOverrides, warnIgnoredExpiry, configPlatforms as listPlatforms, EXPLORE_PLATFORMS, DEFAULT_EXPLORE_CONFIG_PATH, type ExplorePlatform } from "../config/explore-config.js";
import { argumentError, output, required, has, flag, flags, nonNegativeInteger, oneOf, positiveInteger } from "./args.js";
import { help } from "./help.js";
import { CiError } from "../shared/ci-errors.js";
import { cleanupAfterCapture, findOptionalBaseline, resolveCompareRefs, resolvePullRequest, writeCaptureOutputs, writeStepOutputs } from "../ci/workflow-steps.js";
import { stderrProgress } from "../shared/progress.js";
import { maxParallelTargets } from "../ci/session-plan.js";
export async function doctor(): Promise<void> {
if (has("help")) return help();
@@ -109,13 +110,20 @@ export async function ciPullRequest(): Promise<void> {
output({ ok: true, pullRequest: presentation, author });
}
/** `ci:find-baseline --sha SHA [--platform P]`: a reusable baseline artifact of the base commit, if one exists. */
/**
* `ci:find-baseline --sha SHA [--platform P]`: a reusable baseline artifact of the base commit, if one exists, or else
* the baseline run still capturing it (`pending`), which `capture:review` waits for alongside head.
*/
export async function ciFindBaseline(): Promise<void> {
const platform = platformFlag(), target = targetFlag();
const { baseline, warning } = await findOptionalBaseline({ sha: required("sha"), ...(platform ? { platform } : {}), ...(target ? { target } : {}) });
const { baseline, pending, warning } = await findOptionalBaseline({ sha: required("sha"), ...(platform ? { platform } : {}), ...(target ? { target } : {}) });
if (baseline) await writeStepOutputs(process.env, { id: String(baseline.id), run: String(baseline.run) });
if (pending) {
await writeStepOutputs(process.env, { pending: String(pending.run) });
console.error(`No finished baseline of ${required("sha").slice(0, 8)} yet; baseline run ${pending.run} is still capturing it; capture:review waits for it while head captures, up to capture.baselineWaitMinutes.`);
}
if (warning && process.env.GITHUB_ACTIONS) console.log(`::warning title=${warning.code}::${warning.message.replace(/[\r\n]+/g, " ")}`);
output({ ok: true, baseline: baseline ?? null, ...(warning ? { warning } : {}) });
output({ ok: true, baseline: baseline ?? null, ...(pending ? { pending } : {}), ...(warning ? { warning } : {}) });
}
/** `ci:compare-refs --project DIR --base REF --head REF [--output FILE]`: the two commits a comparison captures. */
@@ -159,10 +167,14 @@ export async function configPlatforms(): Promise<void> {
output({ ok: true, platforms: configTargets(config).map(target => target.id) });
}
/** The targets config.yaml captures, in order, with their platforms and labels: the review workflow's matrix. */
/**
* The targets config.yaml captures, in order, with their platforms and labels: the review workflow's matrix, and how
* many of its jobs run at once (`maxParallel`), since each needs a session of capture.maxSessions.
*/
export async function configTargetList(): Promise<void> {
const config = await readExploreConfig(flag("project") ?? process.cwd(), flag("config") ?? DEFAULT_EXPLORE_CONFIG_PATH);
output({ ok: true, targets: configTargets(config) });
const targets = configTargets(config);
output({ ok: true, targets, maxParallel: maxParallelTargets(config.capture.maxSessions, targets.length) });
}
async function exists(path: string): Promise<boolean> {
@@ -215,6 +227,23 @@ export async function sourceVerify(): Promise<void> {
output({ ok: true, deterministic: true, routes: plan.inventorySurfaceIds.length });
}
/**
* Stops a CI capture GitHub cancels: the runner sends SIGINT, then SIGTERM. The first stops the captures between steps,
* so each stops its session and cancels a hosted repack run it waits on (a blocked eas-cli command gets the signal too
* and returns); the second quits at once. Without it a cancelled job left its EAS workflow run repacking.
*/
function ciInterrupt(): AbortSignal {
const controller = new AbortController();
const interrupt = (signal: NodeJS.Signals) => {
if (controller.signal.aborted) process.exit(130);
process.stderr.write(`\nStopping the capture (${signal}): its sessions are stopped and its hosted repack cancelled. A second signal quits now.\n`);
controller.abort(new CiError({ code: "CAPTURE_CANCELLED", stage: "capture", message: `The capture was stopped (${signal}).`, nextAction: "Rerun the workflow when you are ready." }));
};
process.on("SIGINT", interrupt);
process.on("SIGTERM", interrupt);
return controller.signal;
}
export async function captureReview(): Promise<void> {
// Refs without the CI flags: a review on this machine, with your own logins (see runLocalReview).
if (!has("base-project") && !has("head-project") && !has("pull-request") && !has("compare")) return localReview();
@@ -222,13 +251,17 @@ export async function captureReview(): Promise<void> {
const configPath = flag("config") ?? DEFAULT_EXPLORE_CONFIG_PATH;
const selection = targetSelection();
const rawConfig = await readExploreConfig(headProject, configPath);
const config = withRunOverrides(resolveTargetConfig(rawConfig, selection), { scope: flag("scope"), expiresIn: flag("expires-in"), controller: flag("controller"), simulator: flag("simulator"), repack: flag("repack") });
// A release comparison's workflow may still pass --expires-in; ci:deploy says once that report.preview.expiresAfter is ignored.
if (has("expires-in")) warnIgnoredExpiry("--expires-in");
const overrides = { scope: flag("scope"), expiresIn: flag("expires-in"), controller: flag("controller"), simulator: flag("simulator"), repack: flag("repack") };
const config = withRunOverrides(resolveTargetConfig(rawConfig, selection), overrides);
if (has("pull-request") === has("compare")) argumentError("Pass --pull-request FILE for a pull request, or --compare FILE for any two commits.", has("compare") ? "compare" : "pull-request");
// A matrix job names its target; its artifacts, report and comment section carry the target's id.
const named = selection.target !== undefined || selection.platform !== undefined;
const common = { ...(named ? { target: config.target!.id } : {}), ...(selection.platform ? { platform: selection.platform } : {}),
platforms: listPlatforms(rawConfig), targets: configTargets(rawConfig), baseProject: required("base-project"), headProject,
output: required("output"), config, configPath, progress: stderrProgress() };
platforms: listPlatforms(rawConfig), targets: configTargets(rawConfig), posedTargets: posedTargets(rawConfig), repackPeers: hostedRepackPeers(rawConfig, config, overrides),
baseProject: required("base-project"), headProject,
output: required("output"), config, configPath, progress: stderrProgress(), signal: ciInterrupt() };
const result = has("compare") ? await runCompareCi({ ...common, comparePath: required("compare") })
: await runPublicCi({ ...common, pullRequestPath: required("pull-request") });
// The deploy job runs only when the reviewed config asks for a hosted report.
@@ -238,13 +271,17 @@ export async function captureReview(): Promise<void> {
export async function captureBaseline(): Promise<void> {
const project = required("project");
const config = withRunOverrides(resolveTargetConfig(await readExploreConfig(project, flag("config") ?? DEFAULT_EXPLORE_CONFIG_PATH), targetSelection()), { controller: flag("controller"), simulator: flag("simulator"), repack: flag("repack") });
output({ ok: true, ...await runBaselineCi({ platform: config.app.platform, project, sha: required("sha"), output: required("output"), config, progress: stderrProgress() }) });
const rawConfig = await readExploreConfig(project, flag("config") ?? DEFAULT_EXPLORE_CONFIG_PATH);
const overrides = { controller: flag("controller"), simulator: flag("simulator"), repack: flag("repack") };
const config = withRunOverrides(resolveTargetConfig(rawConfig, targetSelection()), overrides);
output({ ok: true, ...await runBaselineCi({ platform: config.app.platform, project, sha: required("sha"), output: required("output"), config, targets: configTargets(rawConfig), posedTargets: posedTargets(rawConfig),
repackPeers: hostedRepackPeers(rawConfig, config, overrides), progress: stderrProgress(), signal: ciInterrupt() }) });
}
/** `capture` (or `capture:review`) `[--base REF] [--head REF | --committed]`: a local review, opened in the Viewer or hosted with --upload. */
async function localReview(): Promise<void> {
const platform = platformFlag(), target = targetFlag();
if (has("expires-in")) warnIgnoredExpiry("--expires-in");
const parallel = has("parallel") ? positiveInteger("parallel") : undefined;
const result = await runLocalReview({ project: flag("project") ?? process.cwd(), base: flag("base"), head: flag("head"), ...(target ? { target } : {}),
...(parallel ? { parallel } : {}),
@@ -276,9 +313,8 @@ async function localReview(): Promise<void> {
...(result.head.uncommitted?.length ? { uncommittedFiles: result.head.uncommitted.length } : {}) });
if (has("upload")) {
const preview = result.config.report?.preview;
const hosted = await deployLocalReports({ reports: zips, project: previewProjectSlug(result.repository, preview?.project), ...(preview?.owner ? { owner: preview.owner } : {}),
// null is expiresAfter: never, which must not fall back to the default.
expiresAfterDays: has("expires-in") ? previewDays(required("expires-in"), "--expires-in") : preview ? preview.expiresAfterDays : PREVIEW_DEFAULT_DAYS });
if (preview?.expiresAfterIgnored) warnIgnoredExpiry("report.preview.expiresAfter");
const hosted = await deployLocalReports({ reports: zips, project: previewProjectSlug(result.repository, preview?.project), ...(preview?.owner ? { owner: preview.owner } : {}) });
// Hosted, the temporary copies are not needed (a failed upload keeps them for report:upload); --output keeps its own.
const copied = has("output");
if (!copied) await rm(result.work, { recursive: true, force: true });
+15 -10
View File
@@ -18,9 +18,9 @@ const GET_STARTED = `Get started (run these in the Expo app's directory):
doctor --capture [--project DIRECTORY] [--output DIRECTORY] [--target ID | --platform ios|android] [--controller agent-device|argent] [--simulator eas|local] [--repack eas|local] [--yes]
(captures the committed HEAD like the baseline workflow, with your eas login, and reports each route)
capture [--base REF] [--head REF | --committed] [--target ID | --platform ios|android] [--parallel N] [--scope all|changes]
[--controller agent-device|argent] [--simulator eas|local] [--repack eas|local] [--upload [--expires-in 7d|30d|never]] [--no-open] [--no-cache] [--output DIRECTORY] [--project DIRECTORY]
(on this machine: checks out both commits in temporary worktrees, captures them with your eas login and your
local claude or codex login, and opens the report; the head is HEAD with your uncommitted changes, as eas build
[--controller agent-device|argent] [--simulator eas|local] [--repack eas|local] [--upload] [--no-open] [--no-cache] [--output DIRECTORY] [--project DIRECTORY]
(on this machine: checks out both commits in temporary worktrees, captures them with your eas login (in agent
mode, also your local claude or codex login), and opens the report; the head is HEAD with your uncommitted changes, as eas build
uploads them (--committed, or cli.requireCommit in eas.json, leaves them out); the base defaults to where HEAD
left the default branch;
without .expo-explore/config.yaml it uses the defaults setup would write;
@@ -31,7 +31,7 @@ const GET_STARTED = `Get started (run these in the Expo app's directory):
machine, instead of EAS Simulator, and removes it afterwards (or set capture.simulator in config.yaml); builds
still run on EAS, and the host switches appearance for either controller; the app is repacked with the route
observer on this machine too (--repack eas, or capture.repack, keeps the hosted EAS job);
with targets in config.yaml it reviews them at once (--parallel N at most N at a time; the targets of a platform
with targets in config.yaml it reviews them at once, up to capture.maxSessions (--parallel N at most N at a time; the targets of a platform
share each commit's build), --target ID one of them, --platform those of a platform)
view [--catalog FILE] [--port N] [--no-open]
(opens a report in the Viewer)
@@ -39,23 +39,27 @@ const GET_STARTED = `Get started (run these in the Expo app's directory):
(the runs kept on this machine: local reviews, explorations and uploaded reports, for 30 days or up to 2 GB;
in a terminal, an arrow-key picker: Enter opens a run in the local Viewer, u hosts it, d deletes it;
ID is a run's id from reports list, or latest; EXPLORE_INTRO=off skips the title animation)
report:upload <report.zip | report directory>... [--owner ACCOUNT] [--project SLUG] [--expires-in 7d|30d|never] [--config FILE] [--json]
report:upload <report.zip | report directory>... [--owner ACCOUNT] [--project SLUG] [--config FILE] [--json]
(hosts reports already on this machine on EAS Hosting behind one link, with your eas login: one per platform
of the same review, from capture:review --output or a downloaded explore-review artifact, or one explore:binary report)`;
const BINARY = `Prebuilt apps (no source needed; runs on EAS Simulator with your local claude or codex login):
explore:binary <app.apk | app.app | app.tar.gz | simulator .ipa> [--output DIRECTORY] [--no-upload] [--json]
const BINARY = `Prebuilt apps (no source needed; runs with your local claude or codex login):
explore:binary <app.apk | app.app | app.tar.gz | simulator .ipa> [--output DIRECTORY] [--simulator eas|local] [--no-upload] [--json]
(installs the archive in a fresh EAS Simulator session, lets the model explore it, writes a map
and report under --output, hosts the report on EAS Hosting, and prints the link)
--simulator local runs it on a simulator or emulator it starts on this machine instead, with no EAS login,
and opens the report in the local Viewer (--upload hosts it instead, --no-open skips the Viewer, --port N);
iOS needs Xcode, Android the SDK (ANDROID_HOME) with platform-tools, cmdline-tools, the emulator and a system image for this Mac
Flags, all optional: --device NAME --bundle-id ID --launch-url URL --hints "what to focus on"
--controller agent-device|argent (the session's device controller; agent-device by default)
--controller agent-device|argent (the device controller; agent-device by default)
(a sign-in in --hints is written as {{secret:NAME}}: Explore types the value of the environment variable NAME,
and the model and the report only see the reference)
--max-duration-minutes N --max-screens N --max-actions N --provider anthropic|openai --model ID
--owner ACCOUNT --project SLUG --expires-in 7d|30d|never --allow-destructive "Label,Other label"
--owner ACCOUNT --project SLUG --allow-destructive "Label,Other label"
--replay FILE (takes the steps in FILE instead of a model's, so every run and device takes the same path; no model access needed, and not with --provider, --model or --hints)
--alias NAME (one EAS Hosting host per app that always serves the newest run; with EXPLORE_PREVIEW_TOKEN
set to a fixed 32-character token, the hosted link is the same every run)
(iOS needs a simulator build; a device .ipa and Android .aab are rejected before any session starts)
(iOS needs a simulator build; a device .ipa and Android .aab are rejected before any device starts)
explore:session [SESSION_ID] [--bundle-id ID] [--output DIRECTORY] [--upload] [--json]
(explores an app that is already installed on an EAS Simulator session you started — run it where
"eas simulator:start --type agent-device --out-config-type dotenv" (or --type argent) wrote .env.eas-simulator,
@@ -91,6 +95,7 @@ const CI = `CI (the installed workflows run these; they manage remote EAS Simula
(resolves the live PR and refuses it unless it is open, from this repository, and by a trusted author)
ci:compare-refs --project DIRECTORY --base REF --head REF [--output FILE]
ci:find-baseline --sha COMMIT [--target ID | --platform ios|android]
(a finished baseline's artifact of the commit, or else the default branch's baseline run still capturing it)
ci:after-capture --output TEMP_DIRECTORY --checkout DIRECTORY...
(stops the capture's EAS Simulator sessions and checks capture left each checkout unchanged)
(the ci:* commands write their results to $GITHUB_OUTPUT for the next steps)`;
+6 -7
View File
@@ -3,7 +3,7 @@ import { readFile, stat } from "node:fs/promises";
import { basename, dirname, join, relative, resolve, sep } from "node:path";
import { DEFAULT_BINARY_PROJECT } from "../binary/run.js";
import { pruneCaptureCache, repositoryName } from "../ci/local-review.js";
import { readExploreConfig, previewDays, DEFAULT_EXPLORE_CONFIG_PATH, PREVIEW_DEFAULT_DAYS, type ExploreConfig } from "../config/explore-config.js";
import { readExploreConfig, warnIgnoredExpiry, DEFAULT_EXPLORE_CONFIG_PATH, type ExploreConfig } from "../config/explore-config.js";
import { assertHostingExport, deployLocalReports, previewProjectSlug, readReportZip, zipReportDirectory, MAX_REPORT_BYTES, type ReportZipSummary } from "../report/preview-deploy.js";
import { startLocalReportServer } from "../report/local-report-server.js";
import { findSavedRun, listSavedRuns, pruneSavedRuns, readSavedReports, removeSavedRun, reportTarget, saveRun, savedRunsDirectory, SAVED_RUN_LIMITS, type SavedReport, type SavedRun } from "../report/saved-runs.js";
@@ -11,10 +11,9 @@ import { openViewerInBrowser } from "../map/viewer-server.js";
import { formatBytes, friendlyTime, kindLabel, runPicker, runTitle } from "./reports-picker.js";
import { pickIntro } from "./title-intros.js";
import { CiError, toCiError } from "../shared/ci-errors.js";
import { argumentError, flag, has, nonNegativeInteger, output, positionals, required } from "./args.js";
import { argumentError, flag, has, nonNegativeInteger, output, positionals } from "./args.js";
/** explore:binary's default: an exploration link is for a quick look. */
const EXPLORATION_DEFAULT_DAYS = 7;
/** A report directory (review.json or exploration.json and assets/), its parent holding report.zip, or a report ZIP. */
async function readReport(path: string): Promise<{ path: string; zip: Uint8Array; summary: ReportZipSummary }> {
@@ -94,12 +93,12 @@ async function hostReports(reports: ReadReport[]): Promise<void> {
const found = review ? await repositoryPreview(review.repository) : undefined;
const preview = found?.preview;
if (review) process.stderr.write(found ? `report.preview from ${found.source}\n`
: `no config of ${review.repository} here: hosting with the defaults (pass --config, or --owner, --project and --expires-in)\n`);
: `no config of ${review.repository} here: hosting with the defaults (pass --config, or --owner and --project)\n`);
const project = flag("project") ?? (review ? previewProjectSlug(review.repository, preview?.project) : DEFAULT_BINARY_PROJECT);
const owner = flag("owner") ?? preview?.owner;
const expiresAfterDays = has("expires-in") ? previewDays(required("expires-in"), "--expires-in")
: preview ? preview.expiresAfterDays : review ? PREVIEW_DEFAULT_DAYS : EXPLORATION_DEFAULT_DAYS;
const hosted = await deployLocalReports({ project, ...(owner ? { owner } : {}), expiresAfterDays,
if (has("expires-in")) warnIgnoredExpiry("--expires-in");
else if (preview?.expiresAfterIgnored) warnIgnoredExpiry("report.preview.expiresAfter");
const hosted = await deployLocalReports({ project, ...(owner ? { owner } : {}),
reports: reports.map(report => ({ platform: report.summary.platform, target: report.summary.target, zip: report.zip,
...(report.summary.changes === undefined ? {} : { changes: report.summary.changes }), ...(report.summary.device ? { device: report.summary.device } : {}) })),
ownerHint: "Pass --owner with the Expo account that should host the report.",
+70 -27
View File
@@ -94,6 +94,16 @@ export type ExploreConfig = {
/** `exclude` names routes, as `requiresRoute` does, that stay in the map but are never captured. */
exploration: { scope: "changes" | "all"; exclude?: string[] };
capture: { maxMinutes: number; maxActions: number;
/**
* The most EAS Simulator sessions one review may run at once, across its targets and sides; absent:
* DEFAULT_MAX_SESSIONS (5). Below the number of targets, that many target jobs run at once, one session each.
*/
maxSessions?: number;
/**
* How many minutes a review waits for the default branch's baseline run of its base commit, still capturing when
* the review starts, before it captures the base itself; 0 never waits. Absent: DEFAULT_BASELINE_WAIT_MINUTES (15).
*/
baselineWaitMinutes?: number;
/** Where the route observer is bundled into the app, when the config chooses; captureRepack gives the default. */
repack?: "eas" | "local"; appearance: AppearanceSetting; launch: LaunchSetting; reset: RouteResetSetting; unattributed: UnattributedSetting; readinessSeconds?: number;
/** The device controller; absent means agent-device, so configs that never chose one keep their cache recipes. */
@@ -104,7 +114,9 @@ export type ExploreConfig = {
setup?: ExploreSetupAction[];
agent?: { provider: "openai" | "anthropic"; model: string; instructions?: string };
/** Hosted previews: one token-gated EAS Hosting deployment per review, see docs/private-review.md. */
report?: { preview?: { hosting: "eas"; /** null: the link never expires; the deployment lives until someone deletes it. */ expiresAfterDays: number | null; owner?: string; project?: string } };
report?: { preview?: { hosting: "eas"; /** `expiresAfter` was set; it no longer applies (see IGNORED_EXPIRY). */ expiresAfterIgnored?: true; owner?: string; project?: string };
/** `map`: the PR comment draws a Mermaid map of the changed screens and the screens that lead to them. Off by default. */
comment?: { map: boolean } };
/**
* `trustedAuthors`: GitHub logins (e.g. `devin-ai-integration[bot]`) whose same-repository pull
* requests the review workflow captures although they lack write access. The workflow reads this
@@ -113,8 +125,16 @@ export type ExploreConfig = {
ci?: { trustedAuthors?: string[] };
};
export const PREVIEW_MAX_DAYS = 365;
export const PREVIEW_DEFAULT_DAYS = 30;
/**
* What `report.preview.expiresAfter` and `--expires-in` say now that Explore sets no expiry: a hosted preview lasts as long
* as its EAS Hosting deployment. Both are still accepted, so configs and copied workflows keep working.
*/
export const IGNORED_EXPIRY = "no longer applies: a hosted preview lasts as long as its EAS Hosting deployment, which EAS deletes after 30 days on Free and 90 on paid plans unless an alias points at it. Delete the deployment (eas deploy:delete) to remove it sooner.";
/** Says once, on stderr (a warning annotation in GitHub Actions), that an expiry setting is ignored. */
export function warnIgnoredExpiry(setting: string, environment: NodeJS.ProcessEnv = process.env): void {
process.stderr.write(`${environment.GITHUB_ACTIONS === "true" ? "::warning::" : ""}${setting} ${IGNORED_EXPIRY}\n`);
}
const MAX_CONFIG_BYTES = 64 * 1024;
const UNSAFE_CHARACTERS = /[\u0000-\u001f\u007f]/;
@@ -179,6 +199,11 @@ function positiveInteger(value: unknown, label: string, fallback?: number): numb
return value;
}
function boundedInteger(value: unknown, label: string, minimum: number, maximum: number): number {
if (typeof value !== "number" || !Number.isSafeInteger(value) || value < minimum || value > maximum) fail(`${label} must be a whole number from ${minimum} to ${maximum}`, label);
return value;
}
function customScheme(value: unknown, label: string): string {
const result = text(value, label, 128).toLowerCase();
if (!/^[a-z][a-z0-9+.-]*$/.test(result) || RESERVED_SCHEMES.has(result)) {
@@ -285,7 +310,7 @@ export function validateExploreConfig(value: unknown): ExploreConfig {
const exploration = optionalRecord(root.exploration, "exploration");
exactKeys(exploration, ["scope", "exclude"], "exploration");
const capture = optionalRecord(root.capture, "capture");
exactKeys(capture, ["maxMinutes", "maxActions", "repack", "appearance", "launch", "reset", "unattributed", "readinessSeconds", "controller", "simulator"], "capture");
exactKeys(capture, ["maxMinutes", "maxActions", "maxSessions", "baselineWaitMinutes", "repack", "appearance", "launch", "reset", "unattributed", "readinessSeconds", "controller", "simulator"], "capture");
const controller = literal(capture.controller, ["agent-device", "argent"] as const, "capture.controller", "agent-device");
const simulator = literal(capture.simulator, ["eas", "local"] as const, "capture.simulator", "eas");
const mode = literal(root.mode, ["deterministic", "agent"] as const, "mode", "deterministic");
@@ -332,6 +357,8 @@ export function validateExploreConfig(value: unknown): ExploreConfig {
capture: {
maxMinutes: positiveInteger(capture.maxMinutes, "capture.maxMinutes", 30),
maxActions: positiveInteger(capture.maxActions, "capture.maxActions", 500),
...(capture.maxSessions === undefined ? {} : { maxSessions: positiveInteger(capture.maxSessions, "capture.maxSessions", 2) }),
...(capture.baselineWaitMinutes === undefined ? {} : { baselineWaitMinutes: boundedInteger(capture.baselineWaitMinutes, "capture.baselineWaitMinutes", 0, 60) }),
...(capture.repack === undefined ? {} : { repack: literal(capture.repack, ["eas", "local"] as const, "capture.repack") }),
appearance: parseAppearance(capture.appearance, "capture.appearance"),
launch: parseLaunchSetting(capture.launch, "capture.launch"),
@@ -430,6 +457,11 @@ export function configTargets(config: ExploreConfig): RunTarget[] {
label: target.label ?? `${target.device ?? (target.platform === config.app.platform ? config.device.name : DEFAULT_ANDROID_DEVICE)}${target.pose ? ` (${target.pose})` : ""}` }));
}
/** The targets with a device pose (foldables): they take the sessions `capture.maxSessions` does not divide evenly. */
export function posedTargets(config: ExploreConfig): string[] {
return Object.entries(config.targets ?? {}).filter(([, target]) => target.pose).map(([id]) => id);
}
/**
* The flat configuration capture consumes for one target, with `target` naming it. `--target` names one; `--platform`
* names the target of that platform, when only one has it; with neither, the config must have a single target.
@@ -505,8 +537,8 @@ export function resolvePlatformConfig(config: ExploreConfig, platform: ExplorePl
/**
* One run's choices over the reviewed config, from the workflow that started it: a release comparison can capture
* every screen (`scope: all`) and keep its hosted link (`expiresIn: never`). `expiresIn` changes nothing when the
* config hosts no report.
* every screen (`scope: all`). `expiresIn` (`--expires-in`) is accepted from workflows written before Explore stopped
* setting an expiry, and changes nothing.
*/
export function withRunOverrides(config: ExploreConfig, overrides: { scope?: string | undefined; expiresIn?: string | undefined; controller?: string | undefined; simulator?: string | undefined; repack?: string | undefined }): ExploreConfig {
let result = config;
@@ -528,11 +560,6 @@ export function withRunOverrides(config: ExploreConfig, overrides: { scope?: str
if (overrides.scope !== "changes" && overrides.scope !== "all") fail("--scope must be changes or all", "--scope");
result = { ...result, exploration: { ...result.exploration, scope: overrides.scope } };
}
if (overrides.expiresIn !== undefined) {
// Validated even without hosting, so a typo in a release workflow surfaces before hosting is turned on.
const expiresAfterDays = previewDays(overrides.expiresIn, "--expires-in");
if (result.report?.preview) result = { ...result, report: { ...result.report, preview: { ...result.report.preview, expiresAfterDays } } };
}
return result;
}
@@ -544,16 +571,17 @@ export function captureRepack(config: ExploreConfig): "eas" | "local" {
return config.capture.repack ?? (config.capture.simulator === "local" ? "local" : "eas");
}
/** `expiresAfter` is a whole number of days (`30d`, `30 days`, or `30`), at most PREVIEW_MAX_DAYS, or `never` (null). */
export function previewDays(value: unknown, field: string): number | null {
if (value === undefined) return PREVIEW_DEFAULT_DAYS;
if (typeof value === "string" && /^\s*never\s*$/i.test(value)) return null;
const match = typeof value === "number" ? [null, String(value)] : typeof value === "string" ? /^\s*(\d{1,3})\s*(?:d|days?)?\s*$/i.exec(value) : null;
const days = match ? Number(match[1]) : Number.NaN;
if (!Number.isInteger(days) || days < 1 || days > PREVIEW_MAX_DAYS) {
fail(`${field} must be between 1 and ${PREVIEW_MAX_DAYS} days, such as 30d, or never`, field);
}
return days;
/**
* The targets, in config order, whose captures repack the same app as `selected` on EAS: one platform, EAS profile and
* bundle id. Their matrix jobs make one hosted repack per commit between them, led by the first (RepackShare); a target
* no other one shares an app with gets just itself.
*/
export function hostedRepackPeers(config: ExploreConfig, selected: ExploreConfig, overrides: Parameters<typeof withRunOverrides>[1] = {}): string[] {
if (captureRepack(selected) !== "eas" || !selected.target) return [];
const sameApp = (other: ExploreConfig) => other.app.platform === selected.app.platform && other.app.easProfile === selected.app.easProfile
&& other.app.bundleId === selected.app.bundleId && captureRepack(other) === "eas";
return configTargets(config).filter(target => target.id === selected.target!.id || sameApp(withRunOverrides(resolveTargetConfig(config, { target: target.id }), overrides)))
.map(target => target.id);
}
const GITHUB_LOGIN = /^[A-Za-z0-9](?:[A-Za-z0-9]|-(?=[A-Za-z0-9])){0,38}(?:\[bot\])?$/;
@@ -588,9 +616,23 @@ function trustedAuthorList(value: unknown): string[] {
function reportConfig(value: unknown): ExploreConfig["report"] {
if (value === undefined) return undefined;
const report = record(value, "report");
exactKeys(report, ["preview"], "report");
if (report.preview === undefined) return {};
const preview = record(report.preview, "report.preview");
exactKeys(report, ["preview", "comment"], "report");
const comment = commentConfig(report.comment);
const preview = previewConfig(report.preview);
return { ...(preview ? { preview } : {}), ...(comment ? { comment } : {}) };
}
function commentConfig(value: unknown): NonNullable<ExploreConfig["report"]>["comment"] {
if (value === undefined) return undefined;
const comment = record(value, "report.comment");
exactKeys(comment, ["map"], "report.comment");
if (typeof comment.map !== "boolean") fail("report.comment.map must be true or false", "report.comment.map");
return { map: comment.map };
}
function previewConfig(value: unknown): NonNullable<ExploreConfig["report"]>["preview"] {
if (value === undefined) return undefined;
const preview = record(value, "report.preview");
exactKeys(preview, ["hosting", "expiresAfter", "owner", "project"], "report.preview");
const owner = optionalText(preview.owner, "report.preview.owner", 64);
if (owner !== undefined && !/^[a-z0-9][a-z0-9_-]*$/.test(owner)) {
@@ -600,12 +642,13 @@ function reportConfig(value: unknown): ExploreConfig["report"] {
if (project !== undefined && !/^[a-z0-9][a-z0-9-]*[a-z0-9]$/.test(project)) {
fail("report.preview.project must be a lowercase EAS project slug", "report.preview.project");
}
return { preview: {
return {
hosting: literal(preview.hosting, ["eas"] as const, "report.preview.hosting"),
expiresAfterDays: previewDays(preview.expiresAfter, "report.preview.expiresAfter"),
// Accepted from configs written before Explore stopped setting an expiry; the commands that host say it is ignored.
...(preview.expiresAfter === undefined ? {} : { expiresAfterIgnored: true as const }),
...(owner === undefined ? {} : { owner }),
...(project === undefined ? {} : { project }),
} };
};
}
/**
+64 -4
View File
@@ -93,6 +93,12 @@ export type CaptureBundle = {
}>;
/** The exact observed frames explicitly accepted by the agent, never replacement screenshots. */
agentAcceptance?: { policy: typeof AGENT_ACCEPTANCE_POLICY; frames: Record<string, AgentAcceptedFrame> };
/**
* The appearance passes that failed while the primary one captured (a secondary session that failed, or a later pass of
* one session), each with its error. Every selected route then has a failed `<id>@<scheme>` outcome carrying it, when
* the bundle renders that appearance. Absent when every pass finished.
*/
sessionFailures?: Array<{ scheme: "light" | "dark"; error: NonNullable<CaptureOutcome["error"]> }>;
};
/** Bind acceptance receipts to captured images and, when supplied, their exported observations. */
@@ -144,11 +150,13 @@ export function assertAgentAcceptance(bundle: CaptureBundle,
* capture, and the Android emulator's clock and indicators sit inside the same fraction of a
* Pixel frame. Its glyphs differ between sessions even after `--normalize-status-bar`, and within one
* session whenever the clock or an indicator redraws, so evidence never treats a difference that
* lies entirely inside this band as an app change. The Viewer marks the same band.
* lies entirely inside this band as an app change; a difference that crosses its edge counts only for
* its pixels below it. The Viewer marks the same band. Android: the glyphs of `sdk_gphone64` on Android 15
* (720x1600) reach 6.4 % of the frame, measured on euxy #164, so its band is 7.
*/
export const STATUS_BAR_BAND_PERCENT = 6.5;
/** The top band of a portrait screenshot that holds the system status bar, per platform. */
export const STATUS_BAR_BAND_PERCENT_BY_PLATFORM: Record<"ios" | "android", number> = { ios: STATUS_BAR_BAND_PERCENT, android: 5 };
export const STATUS_BAR_BAND_PERCENT_BY_PLATFORM: Record<"ios" | "android", number> = { ios: STATUS_BAR_BAND_PERCENT, android: 7 };
export function statusBarBandPercent(platform: "ios" | "android" | undefined): number {
return STATUS_BAR_BAND_PERCENT_BY_PLATFORM[platform ?? "ios"];
}
@@ -164,6 +172,47 @@ export function homeIndicatorBandPercent(platform: "ios" | "android" | undefined
return HOME_INDICATOR_BAND_PERCENT_BY_PLATFORM[platform ?? "ios"];
}
/**
* An area of a screenshot the system draws, in percent of the screenshot's width and height. A difference that
* lies entirely inside one is never an app change; the verdict leaves it out and the Viewer blurs it.
*/
export type SystemRegion = { kind: "status-bar" | "home-indicator"; x: number; y: number; width: number; height: number };
/**
* iPhone Duo (iOS 27) has no status bar across the top: its clock and Wi-Fi indicator stack in one pill in the
* right-hand rail, 28 points from the edge, lower on the outer (closed) panel than on the inner one. Measured on
* EAS Simulator captures of both panels; the pill is the same 40×64 points on each, only its top differs. The app's own
* rail controls start about 3% of the height below it on the outer panel, so the padding stays small.
*/
export const DUO_PANELS = {
outer: { width: 466, height: 678, statusBar: { x: 398, y: 87, width: 40, height: 64 } },
inner: { width: 951, height: 669, statusBar: { x: 883, y: 34, width: 40, height: 64 } },
} as const;
const DUO_STATUS_BAR_PADDING_POINTS = 4;
const isFoldable = (platform: "ios" | "android" | undefined, device: string | undefined) => platform !== "android" && /\bduo\b/i.test(device ?? "");
/**
* The areas the verdict leaves out for a capture: the platform's status-bar band and, on iOS, its home-indicator
* band; on an iPhone foldable, the lit panel's status-bar pill instead of a top band. `pose` is the capture's device pose
* (closed lights the outer panel; open and half-open the inner one); a foldable with none boots closed.
*/
export function systemRegions(options: { platform?: "ios" | "android" | undefined; device?: string | undefined; pose?: string | undefined }): SystemRegion[] {
const regions: SystemRegion[] = [];
if (isFoldable(options.platform, options.device)) {
const panel = options.pose === "open" || options.pose === "half-open" ? DUO_PANELS.inner : DUO_PANELS.outer;
const pad = DUO_STATUS_BAR_PADDING_POINTS;
const percent = (value: number, of: number) => Math.round((value / of) * 10000) / 100;
regions.push({ kind: "status-bar",
x: percent(panel.statusBar.x - pad, panel.width), y: percent(panel.statusBar.y - pad, panel.height),
width: percent(panel.statusBar.width + 2 * pad, panel.width), height: percent(panel.statusBar.height + 2 * pad, panel.height) });
} else {
regions.push({ kind: "status-bar", x: 0, y: 0, width: 100, height: statusBarBandPercent(options.platform) });
}
const home = homeIndicatorBandPercent(options.platform);
if (home > 0) regions.push({ kind: "home-indicator", x: 0, y: 100 - home, width: 100, height: home });
return regions;
}
/** Deterministic pixel comparison of one surface's base and head screenshots. */
export type PublicVisualVerdict = {
verdict: "unchanged" | "modified" | "needs-review";
@@ -187,10 +236,12 @@ export type PublicVisualVerdict = {
* content. Absent when any region changed, or moves were not measured.
*/
allRegionsMoved?: true;
/** Diff regions inside the iOS status-bar band that were excluded from the verdict. */
/** Diff regions inside the status-bar band that were excluded from the verdict, or clipped to their part below it. */
ignoredStatusBarRegions?: number;
/** Diff regions that lay entirely inside the home-indicator band and were left out of the verdict. */
/** Diff regions that lay inside the home-indicator band and were left out of the verdict, or clipped to their part above it. */
ignoredHomeIndicatorRegions?: number;
/** The system-drawn areas the verdict left out, present when a difference inside one was ignored. */
systemRegions?: SystemRegion[];
};
/**
* Evidence-based reviewer status. `status` on the surface stays source-derived because capture
@@ -206,6 +257,11 @@ export type PublicSurfaceReview = {
description: string;
/** Every deterministic accessibility fact for this surface, in reviewer priority order. */
facts?: Array<{ category: string; templateId: string; bullet: string }>;
/**
* The screenshots match because the change is past the edge of the screen: every element its facts name reaches
* beyond the captured screen. Only on a `modified` verdict whose screenshots match.
*/
offScreen?: true;
};
/** An evidence-bound explanation. It never changes a deterministic comparison verdict. */
export type PublicChangeNote = {
@@ -248,6 +304,8 @@ export type PublicReview = {
statusBarBandPercent?: number;
/** The share of each screenshot's height, from the bottom, the verdict ignored as the home indicator. */
homeIndicatorBandPercent?: number;
/** Every system-drawn area the verdict leaves out of each screenshot; the bands above, or a foldable's rail pill. */
systemRegions?: SystemRegion[];
/** Every platform the config captures, so the Viewer can offer the sibling platform's report of the same run. */
platforms?: Array<"ios" | "android">;
/** The target this report captured; absent in reports from before targets, whose target is their platform. */
@@ -277,6 +335,8 @@ export type PublicReview = {
note?: PublicChangeNote;
/** The appearance this surface rendered in; a secondary appearance is its own surface, `<id>@<scheme>`. */
colorScheme?: "light" | "dark";
/** Named in `exploration.exclude`: kept for its neighbors' navigation context, never captured on either side. */
excluded?: true;
}>;
};
export function validateCapturePlan(value: unknown): SourceCapturePlan {
+13 -2
View File
@@ -57,6 +57,7 @@ function assertReviewFields(surface: Surface, enforceComparisonEvidence: boolean
|| (["modified", "unchanged", "variant"].includes(review.status) && (!surface.base || !surface.head))
|| (surface.sourceScope !== undefined && !["added", "removed"].includes(review.status)
&& !SCOPE_VERDICTS[surface.sourceScope]?.includes(review.status))
|| (review.offScreen !== undefined && (review.offScreen !== true || review.status !== "modified" || surface.visual?.verdict !== "unchanged"))
|| (review.facts !== undefined && (!Array.isArray(review.facts) || review.facts.length > 100
|| review.facts.some((fact) => !fact || typeof fact.category !== "string" || typeof fact.templateId !== "string"
|| typeof fact.bullet !== "string" || !fact.bullet.trim() || fact.bullet.length > MAX_REVIEW_DESCRIPTION)))) {
@@ -75,11 +76,17 @@ function assertReviewFields(surface: Surface, enforceComparisonEvidence: boolean
}
}
/** A region's move between the screenshots: finite, and small enough to be layout drift rather than another screen. */
/**
* The farthest a region may move between the screenshots, in screenshot pixels: within the largest screenshot a
* device takes (an element recentred on a wide display moves hundreds of pixels), not an unbounded number.
*/
const MAX_SHIFT = 8192;
/** A region's move between the screenshots: finite, and within a screenshot. */
function validShift(shift: unknown): boolean {
if (!shift || typeof shift !== "object") return false;
const { x, y } = shift as { x?: unknown; y?: unknown };
return typeof x === "number" && typeof y === "number" && Number.isFinite(x) && Number.isFinite(y) && Math.abs(x) <= 64 && Math.abs(y) <= 64;
return typeof x === "number" && typeof y === "number" && Number.isFinite(x) && Number.isFinite(y) && Math.abs(x) <= MAX_SHIFT && Math.abs(y) <= MAX_SHIFT;
}
function assertVisualVerdict(visual: PublicVisualVerdict, enforceComparisonEvidence: boolean, paired: boolean): void {
@@ -345,6 +352,10 @@ function validatePublicReview(review: PublicReview, enforceComparisonEvidence: b
throw new Error("Public review source presence or status is inconsistent");
}
ids.add(surface.id);
if (surface.excluded !== undefined && (surface.excluded !== true
|| surface.baseImage || surface.headImage || surface.baseOutcome || surface.headOutcome || surface.visual)) {
throw new Error("Public review excluded surface carries capture evidence");
}
assertReviewFields(surface, enforceComparisonEvidence);
assertChangeNote(surface, review);
if (surface.sourceChange) {
+73
View File
@@ -0,0 +1,73 @@
import type { PublicReview } from "./capture.js";
import { navigationCaptureContext } from "./navigation-graph.js";
type Surface = PublicReview["surfaces"][number];
type Side = "base" | "head";
/** Navigation is route-level: a secondary-appearance surface (`<id>@<scheme>`) rides on its route's entry. */
const SCHEME_SUFFIX = /@(light|dark)$/;
const routeOf = (id: string): string => id.replace(SCHEME_SUFFIX, "");
/**
* The surfaces a side planned to capture: every route in `all` mode, the changed routes and their navigation context in
* `changes` mode. Undefined without navigation, which leaves no route outside the context.
*/
export function captureContext(review: PublicReview, side: Side): ReadonlySet<string> | undefined {
const navigation = review.navigation?.[side];
if (!navigation) return undefined;
const inventory = review.surfaces.filter((surface) => surface[side]);
const routes = new Map<string, NonNullable<Surface[Side]>>();
for (const surface of inventory) {
const id = routeOf(surface.id);
if (!routes.has(id)) routes.set(id, { ...surface[side]!, id });
}
const selected = new Set(review.mode === "all" ? [...routes.keys()] : navigationCaptureContext(navigation,
inventory.filter((surface) => surface.status !== "unchanged").map((surface) => routeOf(surface.id)),
[...routes.values()]));
return new Set(inventory.filter((surface) => selected.has(routeOf(surface.id))).map((surface) => surface.id));
}
/** Both a screenshot of the planned route and a verified landing on it. */
export function verifiedCapture(surface: Surface, side: Side): boolean {
const outcome = side === "base" ? surface.baseOutcome : surface.headOutcome;
return Boolean(surface[side]?.kind === "screen" && (side === "base" ? surface.baseImage : surface.headImage)
&& outcome?.status === "captured" && outcome.landing === "verified");
}
/**
* Why a side that exists in its revision has no verified capture, as the report's capture gap: `accepted` when the run
* left it uncaptured on purpose (a redirect declaration, a route named in `exploration.exclude`, or a route outside the
* capture context), `open` for anything else, such as a capture that failed. Undefined for a verified capture or an
* absent side. `context` is the side's `captureContext`.
*/
export function captureGap(review: PublicReview, surface: Surface, side: Side,
context: ReadonlySet<string> | undefined): "accepted" | "open" | undefined {
const entry = surface[side];
if (!entry || verifiedCapture(surface, side)) return undefined;
const image = side === "base" ? surface.baseImage : surface.headImage;
const outcome = side === "base" ? surface.baseOutcome : surface.headOutcome;
const outsideCaptureContext = review.mode === "changes" && context !== undefined && !context.has(surface.id);
const deliberate = !image && !outcome && (entry.kind === "redirect" || surface.excluded === true || outsideCaptureContext
|| (!review.navigation?.[side] && review.mode === "changes" && surface.status === "unchanged"));
return deliberate ? "accepted" : "open";
}
/**
* The surfaces the run never captured on purpose: no screenshot of their own on either side, and every side that exists
* left uncaptured deliberately (`captureGap` is `accepted`). None is one of the report's screens. The Viewer leaves them
* out (viewer/src/utils/omitted-screens.ts reaches the same set from the map's gaps), and the PR comment skips them.
* A side with an open gap, such as a failed capture, keeps the surface: its failure is part of the report.
*/
export function neverCapturedSurfaceIds(review: PublicReview): ReadonlySet<string> {
// The Viewer refuses a review whose navigation names a route its inventory lacks; the comment still renders one,
// without a capture context, so no route counts as outside it.
const contextOf = (side: Side) => { try { return captureContext(review, side); } catch { return undefined; } };
const contexts = { base: contextOf("base"), head: contextOf("head") };
const omitted = new Set<string>();
for (const surface of review.surfaces) {
if (surface.baseImage || surface.headImage) continue;
const gaps = (["base", "head"] as const).map((side) => captureGap(review, surface, side, contexts[side]));
if (gaps.includes("accepted") && !gaps.includes("open")) omitted.add(surface.id);
}
return omitted;
}
+6 -2
View File
@@ -6,6 +6,7 @@ import {
type Observation,
type Screen,
} from "../types.js";
import { accessibilityFingerprintsComparable } from "./store.js";
/**
* Cross-run screen matching for PR-style comparisons (baseline = base branch map,
@@ -253,7 +254,8 @@ function matchEntry(match: Match): { entry: BuildComparisonEntry; review: Compar
// captured on both sides) hide a real change to the primary content.
const baselineHash = match.baseline.representative.fingerprints.accessibility;
const currentHash = match.current.representative.fingerprints.accessibility;
const verdict = baselineHash && currentHash
// Hashes from different fingerprint versions (an older Explore's map) say nothing about whether the trees differ.
const verdict = baselineHash && currentHash && accessibilityFingerprintsComparable(baselineHash, currentHash)
? (baselineHash === currentHash ? "unchanged" : "modified")
: "unknown";
const unchanged = verdict === "unchanged";
@@ -262,7 +264,9 @@ function matchEntry(match: Match): { entry: BuildComparisonEntry; review: Compar
? `Matched by ${signals}; the representative captures share an accessibility fingerprint, so the rendered tree is unchanged.`
: verdict === "modified"
? `Matched by ${signals}; the representative captures have different accessibility fingerprints, so the rendered tree changed.`
: `Matched by ${signals}; accessibility fingerprints are unavailable on at least one side, so the change state could not be verified from fingerprints.`;
: baselineHash && currentHash
? `Matched by ${signals}; the representative captures were fingerprinted by different Explore versions, so the change state could not be verified from fingerprints.`
: `Matched by ${signals}; accessibility fingerprints are unavailable on at least one side, so the change state could not be verified from fingerprints.`;
const entry: BuildComparisonEntry = {
id: `match-${match.current.screen.id}`,
relationship: "one-to-one",
+23 -5
View File
@@ -14,6 +14,7 @@ import { dirname, extname, isAbsolute, join, relative, resolve, sep } from "node
import { fileURLToPath } from "node:url";
import { Ajv2020, type ErrorObject, type ValidateFunction } from "ajv/dist/2020.js";
import { isSystemUiNode } from "../capture/controller-output.js";
import { keyboardNodes } from "../capture/keyboard.js";
import { validateNavigationGraph } from "../contracts/navigation-graph.js";
import {
CONTRACT,
@@ -67,11 +68,26 @@ export const normalizeSelector = (input?: string): string | undefined => {
return input.replace(/\s+/g, " ").trim() || undefined;
};
const VOLATILE_SNAPSHOT_NODE_KEYS = new Set(["ref"]);
/** A ref names a node within one read; which field has the focus can change between two reads of one screen. */
const VOLATILE_SNAPSHOT_NODE_KEYS = new Set(["ref", "focused"]);
/**
* The accessibility fingerprint's algorithm, written as its prefix. Version 2 leaves the system keyboard and the focus out
* of the tree; a hash from before carries no prefix. Two fingerprints of different versions say nothing about whether the
* trees differ, so a comparison treats them as not comparable rather than changed (see accessibilityFingerprintsComparable).
*/
const ACCESSIBILITY_FINGERPRINT_VERSION = "a2";
export function accessibilityFingerprintsComparable(left: string, right: string): boolean {
const version = (hash: string) => /^(a\d+):/.exec(hash)?.[1] ?? "a1";
return version(left) === version(right);
}
/** Canonical JSON of the node tree only: controller wrappers carry per-session diagnostics and
* ref generations that would otherwise make identical screens fingerprint differently, and Android
* system UI nodes (status bar clock and signal, keyboard) change between emulator sessions. */
* system UI nodes (status bar clock and signal, keyboard) change between emulator sessions. The
* system keyboard on either platform is left out as the elements leave it out, so a keyboard up on
* one side only does not make two captures of one screen differ. */
function normalizeJsonSnapshotNodes(input: string): string | undefined {
if (!/^\s*\{/.test(input)) return undefined;
let value: unknown;
@@ -86,10 +102,12 @@ function normalizeJsonSnapshotNodes(input: string): string | undefined {
if (!data || typeof data !== "object" || !Array.isArray((data as { nodes?: unknown }).nodes)) {
return undefined;
}
const nodes = (data as { nodes: unknown[] }).nodes;
const keyboard: Set<unknown> = keyboardNodes(nodes.filter((entry): entry is Record<string, unknown> => !!entry && typeof entry === "object" && !Array.isArray(entry)));
const canonical = (node: unknown): unknown => {
if (Array.isArray(node)) {
return node
.filter((entry) => !(entry && typeof entry === "object" && isSystemUiNode(entry as Record<string, unknown>)))
.filter((entry) => !(entry && typeof entry === "object" && (isSystemUiNode(entry as Record<string, unknown>) || keyboard.has(entry))))
.map(canonical);
}
if (!node || typeof node !== "object") return node;
@@ -100,7 +118,7 @@ function normalizeJsonSnapshotNodes(input: string): string | undefined {
.map(([key, entry]) => [key, canonical(entry)]),
);
};
return `nodes\0${JSON.stringify(canonical((data as { nodes: unknown[] }).nodes))}`;
return `nodes\0${JSON.stringify(canonical(nodes))}`;
}
export const normalizeSnapshot = (input: string): string =>
@@ -118,7 +136,7 @@ export async function fingerprintState(options: {
snapshotText?: string;
}): Promise<ObservationFingerprints> {
const accessibility = options.snapshotText
? sha256(`snapshot\0${normalizeSnapshot(options.snapshotText)}`)
? `${ACCESSIBILITY_FINGERPRINT_VERSION}:${sha256(`snapshot\0${normalizeSnapshot(options.snapshotText)}`)}`
: undefined;
const screenshot = options.screenshotPath ? await readFile(options.screenshotPath) : undefined;
const visual = screenshot ? sha256(Buffer.concat([Buffer.from("visual\0"), screenshot])) : undefined;
+62 -26
View File
@@ -1,10 +1,10 @@
/**
* The agent-facing view of one hosted report: `/p/<token>/api/v1/...` JSON resources, the Markdown
* index agents get when they fetch the token page without asking for HTML, and the block the Viewer
* shell carries for fetchers that convert HTML to text. It reads the decrypted archive directly, so
* it only picks the fields it publishes from the report JSON and treats everything else as untrusted.
* shell carries for fetchers that convert HTML to text. It reads the report's own JSON (from a ZIP, or from the static
* manifest a hosted preview serves), so it only picks the fields it publishes and treats everything else as untrusted.
*/
import { readZipDirectory, readZipEntry, type ZipEntry } from "./zip-reader.js";
import { readZipDirectory, readZipEntry } from "./zip-reader.js";
export const AGENT_API_VERSION = "v1";
/** The largest review.json or exploration.json a hosted report reads. */
@@ -22,8 +22,10 @@ export type LoadedReport = {
file: "review.json" | "exploration.json";
raw: Uint8Array;
json: Json;
zip: Uint8Array;
entries: Map<string, ZipEntry>;
/** Every file of the report by name (`assets/<sha256>.png`), with its size. */
entries: Map<string, { size: number }>;
/** One file's bytes, at most `limit` of them; undefined when there is no such file or it is larger. Rejects when it cannot be read now. */
read: (name: string, limit: number) => Promise<Uint8Array | undefined>;
};
const rec = (value: unknown): Json => (typeof value === "object" && value !== null && !Array.isArray(value) ? value as Json : {});
@@ -34,25 +36,40 @@ const bool = (value: unknown): boolean | undefined => (typeof value === "boolean
const strings = (value: unknown): string[] => arr(value).filter((item): item is string => typeof item === "string");
const defined = <T extends Json>(object: T): T => Object.fromEntries(Object.entries(object).filter(([, value]) => value !== undefined)) as T;
/** Parses the decrypted archive; undefined when it is not a report this API knows how to read. */
/** Parses a report ZIP; undefined when it is not a report this API knows how to read. */
export async function loadReport(zip: Uint8Array): Promise<LoadedReport | undefined> {
const entries = readZipDirectory(zip);
if (!entries) return undefined;
const file = entries.has("review.json") ? "review.json" : entries.has("exploration.json") ? "exploration.json" : undefined;
const directory = readZipDirectory(zip);
if (!directory) return undefined;
const file = directory.has("review.json") ? "review.json" : directory.has("exploration.json") ? "exploration.json" : undefined;
if (!file) return undefined;
const raw = await readZipEntry(zip, entries.get(file)!, MAX_METADATA_BYTES);
const raw = await readZipEntry(zip, directory.get(file)!, MAX_METADATA_BYTES);
if (!raw) return undefined;
const read = async (name: string, limit: number) => {
const entry = directory.get(name);
return entry ? await readZipEntry(zip, entry, limit) : undefined;
};
return reportFrom(file, raw, directory, read);
}
/** The report a metadata file describes, with its files and a way to read them; undefined when it is not a report this API reads. */
export function reportFrom(file: string, raw: Uint8Array, entries: Map<string, { size: number }>, read: LoadedReport["read"]): LoadedReport | undefined {
let parsed: unknown;
try { parsed = JSON.parse(new TextDecoder().decode(raw)); } catch { return undefined; }
return reportFromJson(file, raw, parsed, entries, read);
}
/** `reportFrom` for a metadata file already parsed: `parsed` is `raw` as JSON. */
export function reportFromJson(file: string, raw: Uint8Array, parsed: unknown, entries: Map<string, { size: number }>, read: LoadedReport["read"]): LoadedReport | undefined {
if (file !== "review.json" && file !== "exploration.json") return undefined;
const root = rec(parsed);
if (file === "review.json") {
const review = rec(root.review);
if (review.contract !== "explore.review" || !Array.isArray(review.surfaces)) return undefined;
return { kind: "pr-review", file, raw, json: review, zip, entries };
return { kind: "pr-review", file, raw, json: review, entries, read };
}
const report = root.kind === "explore.exploration-report" ? root : rec(root.report);
if (report.kind !== "explore.exploration-report" || typeof report.map !== "object") return undefined;
return { kind: "exploration", file, raw, json: report, zip, entries };
return { kind: "exploration", file, raw, json: report, entries, read };
}
export type AgentContext = {
@@ -159,7 +176,8 @@ const edgeItem = (edge: Json): Json => defined({ from: str(edge.from), to: str(e
* means a visible change. A screen whose source changed but whose captures match is unchanged, including in reports
* from before 0.19.1, which called it "needs review". A modified screen whose screenshots match changed only in what
* assistive technology reads: `accessibility-only`. A screen whose capture failed is `unavailable`, and one whose
* source changed but that has no verdict was not compared: `not-compared`.
* source changed but that has no verdict was not compared: `not-compared`. A route named in `exploration.exclude` is
* `excluded`: never captured on either side.
*/
/** Both sides captured at their verified destination route, as a verdict requires (isAccessibilityOnlyChange in the contract). */
function verifiedPair(surface: Json): boolean {
@@ -177,6 +195,8 @@ function captureUnavailable(surface: Json, side: "base" | "head"): boolean {
}
function verdictOf(surface: Json): string | undefined {
// Named in exploration.exclude: never captured, so there is nothing to compare or to miss.
if (surface.excluded === true) return "excluded";
// A capture that failed is unavailable, as the Viewer says, whatever the source or an older verdict claims.
if (str(surface.status) !== "removed" && captureUnavailable(surface, "head")) return "unavailable";
if (["modified", "unchanged"].includes(str(surface.status) ?? "") && captureUnavailable(surface, "base")) return "unavailable";
@@ -189,9 +209,9 @@ function verdictOf(surface: Json): string | undefined {
return status ?? str(surface.status);
}
const STATUS_ORDER = ["modified", "accessibility-only", "added", "removed", "needs-review", "unavailable", "not-compared", "unchanged", "variant"];
const STATUS_ORDER = ["modified", "accessibility-only", "added", "removed", "needs-review", "unavailable", "not-compared", "excluded", "unchanged", "variant"];
const STATUS_WORDS: Record<string, string> = { "accessibility-only": "accessibility only (screenshots match, the accessibility tree does not)",
"needs-review": "need review", "not-compared": "not compared", variant: "screen variants" };
"needs-review": "need review", "not-compared": "not compared", excluded: "excluded by config (exploration.exclude), never captured", variant: "screen variants" };
/** "3 modified, 1 added, 1 removed, 30 unchanged": every status counted, so the parts add up to the total. */
function screenCounts(screens: Json): string {
@@ -234,24 +254,39 @@ function explorationScreenItem(context: AgentContext, map: Json, screen: Json):
viewerUrl: ID.test(id) ? viewerUrl(context, { screen: id }) : undefined,
});
}
function screenshotUrl(context: AgentContext, map: Json, screen: Json): string | undefined {
/** The screenshot a screen is shown by: its latest observation's. */
function screenshotArtifact(map: Json, screen: Json): Json | undefined {
const observationId = strings(screen.observationIds).at(-1);
const observation = arr(map.observations).map(rec).find(o => o.id === observationId);
const artifact = arr(map.artifacts).map(rec).find(a => a.id === observation?.screenshotArtifactId);
const sha = str(artifact?.sha256);
return arr(map.artifacts).map(rec).find(a => a.id === observation?.screenshotArtifactId);
}
function screenshotUrl(context: AgentContext, map: Json, screen: Json): string | undefined {
const sha = str(screenshotArtifact(map, screen)?.sha256);
return sha && SHA256.test(sha) ? withPlatform(context, apiUrl(context, `/assets/${sha}`)) : undefined;
}
const elementItem = (element: Json): Json => defined({
id: str(element.id ?? element.elementId), label: str(element.label), role: str(element.role), value: str(element.value), identifier: str(element.identifier),
bounds: element.bounds === undefined ? undefined : rec(element.bounds), actions: strings(element.actions), risk: str(element.risk), actionable: bool(element.actionable), enabled: bool(element.enabled), visible: bool(element.visible),
});
/**
* One element, its bounds on the axes of the screenshot it is listed beside. A screenshot stored turned a quarter
* clockwise (an opened iPhone Duo's) is `turnedHeight` points tall on the tree's axes, where the map keeps the bounds:
* there the portrait point (x, y) shows at (H - y, x).
*/
const elementItem = (element: Json, turnedHeight?: number): Json => {
const bounds = element.bounds === undefined ? undefined : rec(element.bounds);
const [x, y, width, height] = [num(bounds?.x), num(bounds?.y), num(bounds?.width), num(bounds?.height)];
const shown = bounds && turnedHeight !== undefined && bounds.unit === "points" && x !== undefined && y !== undefined && width !== undefined && height !== undefined
? { x: turnedHeight - y - height, y: x, width: height, height: width, unit: "points" } : bounds;
return defined({
id: str(element.id ?? element.elementId), label: str(element.label), role: str(element.role), value: str(element.value), identifier: str(element.identifier),
bounds: shown, actions: strings(element.actions), risk: str(element.risk), actionable: bool(element.actionable), enabled: bool(element.enabled), visible: bool(element.visible),
});
};
const transitionItem = (transition: Json): Json => defined({
id: str(transition.id), from: str(transition.fromScreenId), to: str(transition.toScreenId), actionKind: str(transition.actionKind), elementId: str(transition.elementId), attempts: strings(transition.attemptIds).length,
});
function explorationScreenDetail(context: AgentContext, map: Json, screen: Json): Json {
const id = str(screen.id);
const transitions = arr(map.transitions).map(rec);
const elements = arr(map.elements).map(rec).filter(element => element.screenId === id).map(elementItem);
const turnedHeight = num(rec(screenshotArtifact(map, screen)?.turned).height);
const elements = arr(map.elements).map(rec).filter(element => element.screenId === id).map(element => elementItem(element, turnedHeight));
const variants = arr(map.variants).map(rec).filter(variant => variant.screenId === id).map(variant => defined({ id: str(variant.id), key: str(variant.key), name: str(variant.name), observations: strings(variant.observationIds).length }));
const gaps = arr(map.gaps).map(rec).filter(gap => gap.screenId === id).map(gapItem);
return defined({
@@ -465,9 +500,10 @@ export async function agentApiReply(context: AgentContext, loaded: LoadedReport,
}
case "assets": {
if (id === undefined) return notFound();
const entry = loaded.entries.get(`assets/${id}.png`);
if (!entry) return notFound();
const bytes = await readZipEntry(loaded.zip, entry, MAX_ASSET_BYTES);
if (!loaded.entries.has(`assets/${id}.png`)) return notFound();
let bytes: Uint8Array | undefined;
try { bytes = await loaded.read(`assets/${id}.png`, MAX_ASSET_BYTES); }
catch { return json({ error: "unavailable", message: "The screenshot could not be read; retry shortly" }, 503); }
if (!bytes) return notFound();
return { status: 200, contentType: "image/png", body: bytes as BodyInit, headers: { ETag: `"${id}"`, "Content-Length": String(bytes.byteLength) } };
}
+15 -5
View File
@@ -8,13 +8,15 @@ import { compareTitleMarkdown, projectReviewPresentation, type CompareReviewPres
import { renderPublicComment, platformSectionMerge, reviewChangeCount, type PublicCommentState } from "./pr-comment-body.js";
import { isTargetId, type ExplorePlatform } from "../config/explore-config.js";
import { upsertStickyComment } from "./pr-comment.js";
import { retryGitHubLookup } from "./github-lookup.js";
import { previewLinks, viewerLinks } from "./viewer-links.js";
import type { PublicReview } from "../contracts/capture.js";
import { CiError, toCiError } from "../shared/ci-errors.js";
const CONTRACT = "explore.report-input";
/** The reviewed config's hosted-preview request, carried to the credentialed deploy job. */
export type CiPreviewRequest = { hosting: "eas"; expiresAfterDays: number | null; owner?: string; project?: string };
/** The config's report.preview, handed from the capture job to the deploy job. */
export type CiPreviewRequest = { hosting: "eas"; expiresAfterIgnored?: true; owner?: string; project?: string };
export type CiReportInput = {
contract: typeof CONTRACT;
publicReview: PublicReview;
@@ -64,7 +66,8 @@ async function readInputUnchecked(inputPath: string): Promise<CiReportInput> {
}
if (input.preview !== undefined) {
const preview = input.preview;
if (preview.hosting !== "eas" || (preview.expiresAfterDays !== null && (!Number.isInteger(preview.expiresAfterDays) || preview.expiresAfterDays < 1 || preview.expiresAfterDays > 365))
// An `expiresAfterDays` from before Explore stopped setting an expiry is ignored.
if (preview.hosting !== "eas"
|| (preview.project !== undefined && !/^[a-z0-9][a-z0-9-]{0,62}[a-z0-9]$/.test(preview.project))
|| (preview.owner !== undefined && !/^[a-z0-9][a-z0-9_-]{0,63}$/.test(preview.owner))) {
throw new Error("Invalid hosted preview request");
@@ -134,6 +137,8 @@ export async function publishCiReportComment(options: {
preview?: { url: string; expiresAt: number | null };
/** Set by `ci:deploy` when the hosted preview failed: the comment falls back to the central Viewer and the ZIP. */
previewFailed?: boolean;
/** Pause between GitHub lookups that may succeed later; tests replace it so retries take no real time. */
sleep?: (milliseconds: number) => Promise<void>;
}) {
const environment = options.environment ?? process.env; const transport = options.transport ?? fetch;
const input = await readInput(options.inputPath); const review = input.publicReview; const pr = review.pullRequest;
@@ -184,16 +189,21 @@ export async function publishCiReportComment(options: {
catch (error) { throw new CiError({ code: "GITHUB_RESPONSE_INVALID", stage: "publication",
message: "GitHub returned invalid report metadata.", nextAction: "Rerun publication; if the response remains invalid, inspect the GitHub metadata request.", cause: error }); }
};
const artifact = await api(`actions/artifacts/${new URL(options.artifactUrl).pathname.split("/").at(-1)}`);
// A lookup is made again when GitHub may answer it later: no answer, a 5xx, or a 404 from the Actions API, which may
// not list a just-uploaded artifact yet. A bad token, a missing permission or a rate limit fails at once, as before.
const lookup = (path: string) => retryGitHubLookup(() => api(path), error => error instanceof CiError
&& (error.code === "GITHUB_NETWORK_FAILED" || error.code === "GITHUB_REQUEST_FAILED" && error.retryable
|| error.code === "GITHUB_RESOURCE_NOT_FOUND" && path.startsWith("actions/")), options.sleep);
const artifact = await lookup(`actions/artifacts/${new URL(options.artifactUrl).pathname.split("/").at(-1)}`);
if (artifact.expired || artifact.name !== `explore-review-${review.runId}-${review.runAttempt}${target ? `-${target}` : ""}` || String(artifact.workflow_run?.id) !== review.runId) {
throw new CiError({ code: artifact.expired ? "REPORT_ARTIFACT_EXPIRED" : "REPORT_ARTIFACT_MISMATCH", stage: "publication",
message: "The uploaded report artifact is unavailable or belongs to another run",
nextAction: artifact.expired ? "Rerun Explore to upload a fresh report artifact." : "Pass the final report artifact from this workflow run and attempt." });
}
const attempt = await api(`actions/runs/${review.runId}/attempts/${review.runAttempt}`);
const attempt = await lookup(`actions/runs/${review.runId}/attempts/${review.runAttempt}`);
if (String(attempt.id) !== review.runId || String(attempt.run_attempt) !== review.runAttempt) throw new CiError({ code: "REPORT_RUN_MISMATCH", stage: "publication",
message: "Report run attempt mismatch", nextAction: "Publish the artifact and report input from the same current workflow attempt." });
const current = await api(`pulls/${pr.number}`);
const current = await lookup(`pulls/${pr.number}`);
if (current.state !== "open" || current.head?.repo?.full_name !== pr.repository || current.head?.ref !== pr.head.ref) {
throw new CiError({ code: "PR_CHANGED", stage: "publication", message: "PR closed or changed branches before the report artifact was ready",
nextAction: "Run Explore for the current PR head and base before publishing a ready report." });
+69 -41
View File
@@ -15,7 +15,7 @@ export type FailureCommentOptions = {
runAttempt: string;
/** The directory the failure job downloaded diagnostics artifacts into (`<stage>/<artifact>/…/error.json`). */
diagnosticsPath?: string;
/** The failed target's id; otherwise the one the first diagnostics artifact's name carries. */
/** The failed target's id; otherwise every target the diagnostics artifacts name. */
target?: string;
/** What callers from before targets pass: a platform is the id of its default target. */
platform?: ExplorePlatform;
@@ -25,6 +25,8 @@ export type FailureCommentOptions = {
};
type Diagnostic = { code: string; stage: string; message: string; nextAction: string };
/** A failed target's diagnostic. `target` is absent when neither the artifact's name nor the file names one. */
export type LocatedFailure = { diagnostic?: Diagnostic; target?: string; platform?: ExplorePlatform };
const SHA = /^[0-9a-f]{40}$/;
// `explore-<kind>-<run>-<attempt>-<target>`: the kind has no digits, so the first number is the run and the rest is the target id.
const ARTIFACT_DIR = /^explore-[a-z-]+-\d+-\d+-([a-z][a-z0-9-]{0,31})$/;
@@ -34,11 +36,14 @@ const MAX_DIAGNOSTIC_BYTES = 32_768;
const CODE = /^[A-Z][A-Z0-9_]{1,63}$/;
const STAGE = /^[a-z][a-z0-9-]{0,63}$/;
function parseDiagnostic(text: string, stage: string): Diagnostic | undefined {
function parseJson(text: string): Record<string, unknown> | undefined {
let parsed: unknown;
try { parsed = JSON.parse(text); } catch { return undefined; }
if (!parsed || typeof parsed !== "object" || Array.isArray(parsed)) return undefined;
const value = parsed as Record<string, unknown>;
return parsed && typeof parsed === "object" && !Array.isArray(parsed) ? parsed as Record<string, unknown> : undefined;
}
function parseDiagnostic(value: Record<string, unknown> | undefined, stage: string): Diagnostic | undefined {
if (!value) return undefined;
const field = (name: string, max: number) => typeof value[name] === "string" && (value[name] as string).trim() && (value[name] as string).length <= max ? value[name] as string : undefined;
const code = field("code", 64), message = field("message", 4000), nextAction = field("nextAction", 4000), stageName = field("stage", 64) ?? stage;
if (!code || !CODE.test(code) || !message || !nextAction || !STAGE.test(stageName)) return undefined;
@@ -46,32 +51,51 @@ function parseDiagnostic(text: string, stage: string): Diagnostic | undefined {
return { code, stage: stageName, message, nextAction };
}
/** Reads one error.json; `named` is the target its artifact directory names, else the one the file records. */
async function readFailure(path: string, stage: string, named: string | undefined): Promise<LocatedFailure | undefined> {
const info = await lstat(path).catch(() => undefined);
if (!info) return undefined;
const located = (target: string | undefined): LocatedFailure => ({ ...(target ? { target } : {}), ...(target === "ios" || target === "android" ? { platform: target as ExplorePlatform } : {}) });
if (!info.isFile() || info.size > MAX_DIAGNOSTIC_BYTES) return located(named);
const value = parseJson(await readFile(path, "utf8"));
const target = named ?? (isTargetId(value?.target) ? value.target : undefined);
const diagnostic = parseDiagnostic(value, stage);
return diagnostic ? { diagnostic, ...located(target) } : located(target);
}
/**
* The error.json a failed job left behind and the target its artifact name carries (`ios` or `android` for a config
* without targets, and for artifacts of versions before targets). A report failure's diagnostic takes precedence over
* capture's; a present but unreadable file yields the generic failure rather than an earlier stage's error, and only
* Every failed target's error.json and the target it belongs to (`ios` or `android` for a config without targets, and
* for artifacts of versions before targets). actions/download-artifact puts each artifact a pattern matches in a
* directory named after it, except when the pattern matches exactly one: that one lands in the stage's directory
* itself, and the file names its target. A report failure's diagnostic takes precedence over its target's capture
* diagnostic; a present but unreadable file yields the generic failure rather than an earlier stage's error, and only
* regular files count (never symlinks out of the artifact). `platform` is set when the target id is a platform's.
*/
export async function locateFailureDiagnostic(root: string): Promise<{ diagnostic?: Diagnostic; target: string; platform?: ExplorePlatform } | undefined> {
export async function locateFailureDiagnostics(root: string): Promise<LocatedFailure[]> {
const failures: LocatedFailure[] = [];
const seen = new Set<string>();
for (const [stage, file] of [["report", "error.json"], ["capture", join("explore-diagnostics", "error.json")]] as const) {
const entries = await readdir(join(root, stage), { withFileTypes: true }).catch(() => []);
for (const entry of entries.filter(entry => entry.isDirectory() && artifactTarget(entry.name)).sort((a, b) => a.name.localeCompare(b.name))) {
const path = join(root, stage, entry.name, file);
const info = await lstat(path).catch(() => undefined);
if (!info) continue;
const target = artifactTarget(entry.name)!;
const located: { target: string; platform?: ExplorePlatform } = { target, ...(target === "ios" || target === "android" ? { platform: target as ExplorePlatform } : {}) };
if (!info.isFile() || info.size > MAX_DIAGNOSTIC_BYTES) return located;
const diagnostic = parseDiagnostic(await readFile(path, "utf8"), stage);
return diagnostic ? { diagnostic, ...located } : located;
const candidates = [...entries.filter(entry => entry.isDirectory() && artifactTarget(entry.name)).sort((a, b) => a.name.localeCompare(b.name))
.map(entry => ({ path: join(root, stage, entry.name, file), named: artifactTarget(entry.name) })), { path: join(root, stage, file), named: undefined }];
for (const { path, named } of candidates) {
const located = await readFailure(path, stage, named);
if (!located || seen.has(located.target ?? "")) continue;
seen.add(located.target ?? "");
failures.push(located);
}
}
return undefined;
return failures;
}
/** The first failure locateFailureDiagnostics finds: a report failure before a capture failure. */
export async function locateFailureDiagnostic(root: string): Promise<LocatedFailure | undefined> {
return (await locateFailureDiagnostics(root))[0];
}
/**
* The failure job's comment: a job died outside Explore's own progress comments (install, checkout, a crash
* before the pipeline posted), so the sticky comment gets a failed record for that target from the same
* before the pipeline posted), so the sticky comment gets a failed record for each failed target from the same
* renderer every other writer uses. Nothing is posted when the PR is closed; a PR that moved on is marked
* superseded so the failure lands under previous runs.
*/
@@ -94,27 +118,31 @@ export async function publishFailureComment(options: FailureCommentOptions): Pro
return { posted: false, reason: "The PR is no longer an open same-repository PR." };
}
const superseded = pr.head?.sha !== options.headSha || pr.base?.sha !== options.baseSha;
const located = options.diagnosticsPath ? await locateFailureDiagnostic(options.diagnosticsPath) : undefined;
const target = requested ?? located?.target;
// The failure job knows a target by its id alone; the comment's own record for that id supplies its label and platform.
const platform: ExplorePlatform | undefined = target === "ios" || target === "android" ? target : undefined;
const configPath = options.configPath ?? ".expo-explore/config.yaml";
const now = Date.now();
const diagnostic = located?.diagnostic;
const state: PublicCommentState = {
phase: "failed", mode: "changes", executionMode: "deterministic", configPath,
pullRequest: { repositoryId: String(pr.base?.repo?.id ?? ""), repository: options.repository, number: options.pullNumber, title: String(pr.title ?? ""), url: String(pr.html_url ?? ""),
base: { ref: String(pr.base?.ref ?? ""), sha: options.baseSha }, head: { ref: String(pr.head?.ref ?? ""), sha: options.headSha } },
// No artifact URL, so no Viewer link: the target would only pick a report that this failure never uploaded.
links: viewerLinks({ repository: options.repository, headSha: options.headSha, runId: options.runId, runAttempt: options.runAttempt, configPath }),
startedAt: now, updatedAt: now, artifactOnly: true,
...(platform ? { build: { platform, easProfile: "" } } : {}),
failure: { step: diagnostic?.stage ?? "workflow", message: diagnostic
? `${diagnostic.code}: ${diagnostic.message} Next action: ${diagnostic.nextAction}`
: "Explore could not finish this review. See the run for the error code, cause, and next action. Fix the reported issue and rerun Explore." },
...(superseded ? { superseded: true } : {}),
};
await upsertStickyComment({ repository: options.repository, issueNumber: options.pullNumber, body: renderPublicComment(state, undefined, target),
...platformSectionMerge(state, target), token, fetchImpl: transport });
const located = options.diagnosticsPath ? await locateFailureDiagnostics(options.diagnosticsPath) : [];
// Each failed target gets its own failed record: since one target's failed capture no longer stops the others'
// reports (brentvatne/euxy#172), several targets of one run can fail in different jobs.
const failures: LocatedFailure[] = requested !== undefined ? [{ ...(located.find(failure => failure.target === requested) ?? located[0]), target: requested }]
: located.length ? located : [{}];
for (const { target, diagnostic } of failures) {
// The failure job knows a target by its id alone; the comment's own record for that id supplies its label and platform.
const platform: ExplorePlatform | undefined = target === "ios" || target === "android" ? target : undefined;
const configPath = options.configPath ?? ".expo-explore/config.yaml";
const now = Date.now();
const state: PublicCommentState = {
phase: "failed", mode: "changes", executionMode: "deterministic", configPath,
pullRequest: { repositoryId: String(pr.base?.repo?.id ?? ""), repository: options.repository, number: options.pullNumber, title: String(pr.title ?? ""), url: String(pr.html_url ?? ""),
base: { ref: String(pr.base?.ref ?? ""), sha: options.baseSha }, head: { ref: String(pr.head?.ref ?? ""), sha: options.headSha } },
// No artifact URL, so no Viewer link: the target would only pick a report that this failure never uploaded.
links: viewerLinks({ repository: options.repository, headSha: options.headSha, runId: options.runId, runAttempt: options.runAttempt, configPath }),
startedAt: now, updatedAt: now, artifactOnly: true,
...(platform ? { build: { platform, easProfile: "" } } : {}),
failure: { step: diagnostic?.stage ?? "workflow", message: diagnostic
? `${diagnostic.code}: ${diagnostic.message} Next action: ${diagnostic.nextAction}`
: "Explore could not finish this review. See the run for the error code, cause, and next action. Fix the reported issue and rerun Explore." },
...(superseded ? { superseded: true } : {}),
};
await upsertStickyComment({ repository: options.repository, issueNumber: options.pullNumber, body: renderPublicComment(state, undefined, target),
...platformSectionMerge(state, target), token, fetchImpl: transport });
}
return { posted: true, ...(superseded ? { superseded: true } : {}) };
}
+25
View File
@@ -0,0 +1,25 @@
/**
* Pauses between attempts of a GitHub lookup that may succeed later, 45 seconds in all. GitHub's Actions API is
* eventually consistent just after an upload: in brentvatne/euxy's review run 39348956443, the duo-closed report job's
* `actions/artifacts/<id>` answered 404 for the artifact that job had just uploaded, while the other four targets'
* jobs found theirs.
*/
const GITHUB_LOOKUP_RETRY_DELAYS_MS = [3_000, 6_000, 12_000, 24_000] as const;
/**
* Runs a read-only GitHub lookup again after each pause while `transient` says its failure may clear on its own. The
* caller decides what is transient; any other failure, or the last one, is thrown as it is.
*/
export async function retryGitHubLookup<T>(lookup: () => Promise<T>, transient: (error: unknown) => boolean,
sleep: (milliseconds: number) => Promise<void> = milliseconds => new Promise<void>(done => setTimeout(done, milliseconds))): Promise<T> {
for (let retry = 0; ; retry += 1) {
try { return await lookup(); }
catch (error) {
const delayMs = GITHUB_LOOKUP_RETRY_DELAYS_MS[retry];
if (delayMs === undefined || !transient(error)) throw error;
// A lookup that succeeds later would otherwise leave no trace of the failure in the job log.
process.stderr.write(`GitHub lookup failed (${(error instanceof Error ? error.message : String(error)).split("\n", 1)[0]}); trying again in ${delayMs / 1000} s\n`);
await sleep(delayMs);
}
}
}
+23 -2
View File
@@ -7,11 +7,13 @@ import type { ExplorePlatform } from "../config/explore-config.js";
import type { RunTarget } from "../types.js";
import { reportTarget } from "./saved-runs.js";
import { AGENT_API_VERSION, agentApiReply, agentIndexMarkdown, loadReport, prefersHtml, type AgentContext, type LoadedReport } from "./agent-api.js";
import { manifestBody, previewAsset } from "./preview-manifest.js";
/**
* Serve review reports on this machine the way a hosted report serves them: the built Viewer at `/p/<token>`, each
* target's report at `/p/<token>/report.zip?target=…` (or `?platform=…`), and the list at `/p/<token>/reports.json`. The
* Viewer already knows this layout, so nothing in it changes. An agent on this machine gets what it gets from a hosted
* target's manifest at `/p/<token>/manifest.json?target=…` (or `?platform=…`) and its assets at
* `/p/<token>/assets/<sha256>.<ext>`, the whole ZIP at `/p/<token>/report.zip`, and the list at `/p/<token>/reports.json`.
* The Viewer already knows this layout, so nothing in it changes. An agent on this machine gets what it gets from a hosted
* link: the Markdown index when it fetches `/p/<token>` without asking for HTML, `/p/<token>/llms.txt`, and the JSON
* API at `/p/<token>/api/v1/...` (agent-api.ts, shared with the hosted Worker). Loopback only; anything else answers 404.
*/
@@ -67,6 +69,25 @@ export async function startLocalReportServer(options: { reports: LocalReport[];
return send(reply.status, body, reply.contentType, { vary: "Accept", ...reply.headers });
}
if (isPage) return send(200, await readFile(resolve(viewerRoot, "index.html")), TYPES[".html"]);
// What the Viewer opens (preview-manifest.ts, as a hosted preview serves it): the manifest, then each asset as it is
// shown. An asset is named by its SHA-256 and carries no target, so it comes from whichever report holds it.
const read = (entry: LocalReport) => {
if (!loaded.has(entry)) loaded.set(entry, loadReport(entry.zip));
return loaded.get(entry)!;
};
if (url.pathname === `${page}/manifest.json`) {
const content = report ? await read(report) : undefined;
return content ? send(200, manifestBody(content), TYPES[".json"]) : send(404, "Not found");
}
if (url.pathname.startsWith(`${page}/assets/`)) {
const path = url.pathname.slice(page.length + 1);
for (const entry of options.reports) {
const content = await read(entry);
const asset = content ? await previewAsset(content, path) : undefined;
if (asset) return send(200, asset.bytes, asset.type);
}
return send(404, "Not found");
}
if (url.pathname === `${page}/report.zip`) return report ? send(200, report.zip, "application/zip") : send(404, "Not found");
if (url.pathname === `${page}/reports.json` && options.reports.length > 1) return send(200, listing, TYPES[".json"]);
// The Viewer's own assets, at the absolute paths its index.html names; never anything outside the build.
+358 -240
View File
@@ -2,9 +2,10 @@ import type { PublicReview } from "../contracts/capture.js";
import type { RunTarget } from "../types.js";
import type { SimulatorRuntimeMetadata } from "../capture/run.js";
import { CI_COMMENT_MARKER } from "./pr-comment.js";
import { CHANGE_GROUPS, MAX_COMMENT_TARGETS, changeCountText, compareTargets, defaultTargetLabel, foldRecord, strongerGroup, historyLine, orderedRecords, readCommentRecords, recordLabel, recordLine, recordSummary, recordTarget,
type ChangeGroup, type CommentTarget, type CommentRecords, type HistoryEntry, type PlatformRecord } from "./pr-comment-record.js";
import { CHANGE_GROUPS, MAX_COMMENT_TARGETS, MAX_NAVIGATION_EDGES, MAX_NAVIGATION_NODES, compareTargets, defaultTargetLabel, foldRecord, strongerGroup, historyLine, orderedPrevious, orderedRecords, previousLine, readCommentRecords, recordLabel, recordLine, recordSummary,
type ChangeGroup, type CommentNavigation, type CommentTarget, type CommentRecords, type HistoryEntry, type NavigationGroup, type NavigationKind, type PlatformRecord } from "./pr-comment-record.js";
import { isAccessibilityOnlyChange } from "../contracts/review-validation.js";
import { neverCapturedSurfaceIds } from "../contracts/surface-capture.js";
import type { PublicCaptureHealth, PublicEvidenceSummary } from "../review/evidence-assessment.js";
import type { ReviewPresentation } from "../contracts/review.js";
import type { ViewerLinks } from "./viewer-links.js";
@@ -57,18 +58,13 @@ export type PublicCommentState = {
artifactOnly?: boolean;
/** The PR moved while this run worked; the comment keeps its report under previous runs. */
superseded?: boolean;
/** report.comment.map: draw the map of the changed screens. */
map?: true;
};
type ReviewGroup = ChangeGroup;
const GROUP_ORDER: ReviewGroup[] = CHANGE_GROUPS;
const GROUP_LABEL: Record<ReviewGroup, string> = {
added: "🟩 Added",
modified: "🟧 Modified",
removed: "🟥 Removed",
"accessibility-only": "🟦 Accessibility only",
"needs-review": "🟨 Needs review",
};
/** The body is HTML, so no blank lines are needed and the block keeps its natural spacing. */
function details(summary: string, content: string[]): string[] {
@@ -113,14 +109,6 @@ function link(label: string, destination: string): string {
const SUPERSEDED_NOTE = "The pull request moved while this run worked; its report is kept here until the current run replaces it.";
function duration(startedAt: number, updatedAt: number): string {
const total = Math.max(0, Math.round((updatedAt - startedAt) / 1000));
if (startedAt <= 0 || total > 24 * 3600) return "";
const minutes = Math.floor(total / 60);
const seconds = total % 60;
return minutes > 0 ? `${minutes}m ${String(seconds).padStart(2, "0")}s` : `${seconds}s`;
}
function plural(count: number, noun: string): string {
return `${count} ${noun}${count === 1 ? "" : "s"}`;
}
@@ -155,6 +143,16 @@ function notComparedChange(surface: Surface): boolean {
return verified && (!surface.review || surface.visual?.reason === "diff-failed");
}
/**
* An accessibility-only change whose elements lie past the edge of the screen (the review's `offScreen`), so another
* target's row can say the change is there, off-screen, rather than missing. Like notComparedChange, a comparison that
* failed or a redirect is not one.
*/
function offScreenChange(surface: Surface): boolean {
return surface.review?.offScreen === true && isAccessibilityOnlyChange(surface) && surface.visual?.reason !== "diff-failed"
&& (surface.head ?? surface.base)?.kind !== "redirect";
}
function routeOf(surface: Surface): string {
return plainText(surface.head?.route || surface.base?.route || surface.id, MAX_ROUTE_LENGTH);
}
@@ -189,12 +187,7 @@ function previewExpiry(expiresAt: number): string {
return new Date(expiresAt).toLocaleDateString("en-US", { month: "short", day: "numeric", year: "numeric", timeZone: "UTC" }) + " (UTC)";
}
/**
* Differences that are not review targets, stated once instead of per screen: screen variants
* (source-unchanged screens captured in another runtime state) and app-wide source changes that
* left some screens without a visible difference.
*/
/** "Launch: splash unchanged for 3.1 s" — the app launch per appearance, kept apart from the screen rows. */
/** "Launch: splash unchanged for 3.1 s" — the app launch per appearance, for the local text summary. */
function launchLine(review: PublicReview): string | undefined {
const launch = review.launch ?? [];
if (!launch.length) return undefined;
@@ -218,9 +211,19 @@ function launchLine(review: PublicReview): string | undefined {
return `Launch: ${parts.join("; ")}.`;
}
/** "Splash screen changed": the launch comparison when it found a difference; a steady splash is not news for the comment. */
function splashLines(review: PublicReview): string[] {
const launch = review.launch ?? [];
const changed = launch.filter(entry => entry.visual && entry.visual.verdict !== "unchanged");
if (!changed.length) return [];
const verdict = changed.every(entry => entry.visual!.verdict === "needs-review") ? "needs review" : "changed";
const schemes = changed.map(entry => entry.colorScheme).filter((scheme): scheme is NonNullable<typeof scheme> => Boolean(scheme));
return [`Splash screen ${verdict}${schemes.length && changed.length < launch.length ? ` in ${schemes.join(" and ")} mode` : ""}`];
}
function contextLines(review: PublicReview): string[] {
const variants = review.surfaces.filter(surface => surface.review?.status === "variant").length;
const appWide = review.surfaces.filter(surface => surface.sourceScope === "app-wide");
const appWide = review.surfaces.filter(surface => surface.sourceScope === "app-wide" && !surface.excluded);
const files = [...new Set(appWide.flatMap(surface => surface.sourceChange?.files.filter(file => file.relation === "global").map(file => file.path) ?? []))].sort();
const lines: string[] = [];
const launch = launchLine(review);
@@ -234,7 +237,6 @@ function contextLines(review: PublicReview): string[] {
}
const SURFACE_ID = /^[A-Za-z0-9._-]{1,200}(@(light|dark))?$/;
const MAX_ROWS_PER_GROUP = 200;
/** The run id and attempt the run link names (`…/actions/runs/<id>/attempts/<n>`). */
function runIdentity(run: string): { runId: string; runAttempt: string } {
@@ -303,32 +305,50 @@ export function platformRecordOf(state: PublicCommentState, target?: CommentTarg
if (state.review) {
const evidenceBased = state.review.surfaces.some(surface => surface.review);
record.changes = [];
// A route the run never captured on purpose is not one of the report's screens, here as in the Viewer.
const omitted = neverCapturedSurfaceIds(state.review);
for (const surface of state.review.surfaces) {
if ((surface.head ?? surface.base)?.kind === "redirect") continue;
if (omitted.has(surface.id)) continue;
const group = groupOf(surface, evidenceBased);
if (group === "unlisted") continue;
const scheme = schemeOf(surface.id);
record.changes.push({ id: surface.id, route: routeOf(surface), group, description: descriptionOf(surface, group), ...(scheme ? { scheme } : {}) });
record.changes.push({ id: surface.id, route: routeOf(surface), group, description: descriptionOf(surface, group), ...(scheme ? { scheme } : {}),
...(group === "accessibility-only" && offScreenChange(surface) ? { offScreen: true as const } : {}) });
}
const navigation = state.map ? navigationOf(state.review, new Map(record.changes.map(change => [change.id, change.group]))) : undefined;
if (navigation) record.navigation = navigation;
const notCompared = state.review.surfaces.filter(notComparedChange).length;
if (notCompared) record.notCompared = notCompared;
record.context = contextLines(state.review).filter(line => line.length);
record.context = splashLines(state.review);
const diagnostics = [...(state.evidence?.diagnostics ?? [])];
// A failed appearance pass is one issue, named first, rather than one per frame it left unavailable: in brentvatne/
// euxy's main baseline run 39506374824 the dark session's failure would have listed all 16 routes, crowding out the rest.
const sessionFailures = state.evidence?.sessionFailures ?? [];
if (sessionFailures.length) record.sessionFailures = sessionFailures.map(({ side, scheme, frames }) => ({ side, scheme, frames }));
const sessionLines = sessionFailures.map(failure => ({ side: failure.side, route: `${failure.scheme} session`, code: failure.code,
message: `${failure.message.replace(/\.+$/, "")}.${failure.frames ? ` ${plural(failure.frames, `${failure.scheme} frame`)} ${failure.frames === 1 ? "is" : "are"} unavailable.` : ""}`,
nextAction: failure.nextAction }));
const ofFailedPass = (surfaceId: string, side: "base" | "head", outcome: NonNullable<Surface["headOutcome"]>) => sessionFailures.some(failure => failure.side === side
&& schemeOf(surfaceId) === failure.scheme && outcome.error?.code === failure.code && outcome.error.message === failure.message);
for (const surface of state.review.surfaces) {
for (const side of ["base", "head"] as const) {
const outcome = surface[`${side}Outcome`];
if (!outcome || (outcome.status === "captured" && outcome.landing === "verified") || diagnostics.some(item => item.surfaceId === surface.id && item.side === side)) continue;
if (ofFailedPass(surface.id, side, outcome)) continue;
diagnostics.push({ surfaceId: surface.id, side, code: outcome.error?.code ?? (outcome.status === "captured" ? "CAPTURE_LANDING_UNVERIFIED" : "CAPTURE_UNAVAILABLE"),
stage: outcome.error?.stage ?? "capture", message: outcome.error?.message ?? outcome.reason ?? (outcome.status === "captured" ? "Destination route was not verified." : `Capture ${outcome.status}.`),
nextAction: outcome.error?.nextAction ?? "Check this side's capture diagnostics, correct the route configuration or capture failure, and rerun capture.", retryable: outcome.error?.retryable ?? false });
}
}
if (diagnostics.length) {
record.diagnostics = diagnostics.slice(0, 20).map(diagnostic => {
if (diagnostics.length || sessionLines.length) {
record.diagnostics = [...sessionLines, ...diagnostics.map(diagnostic => {
const surface = state.review!.surfaces.find(surface => surface.id === diagnostic.surfaceId);
return { ...(diagnostic.side ? { side: diagnostic.side } : {}), route: surface ? routeOf(surface) : diagnostic.surfaceId, code: diagnostic.code, message: diagnostic.message, nextAction: diagnostic.nextAction };
});
record.diagnosticCount = diagnostics.length;
// A secondary appearance's issue names its appearance, as its change row does: its light frame may be fine.
const scheme = schemeOf(diagnostic.surfaceId);
return { ...(diagnostic.side ? { side: diagnostic.side } : {}), route: `${surface ? routeOf(surface) : diagnostic.surfaceId}${surface && scheme ? ` (${scheme})` : ""}`,
code: diagnostic.code, message: diagnostic.message, nextAction: diagnostic.nextAction };
})].slice(0, 20);
record.diagnosticCount = sessionLines.length + diagnostics.length;
}
}
if (state.failure) record.failure = { step: state.failure.step, message: state.failure.message };
@@ -336,6 +356,105 @@ export function platformRecordOf(state: PublicCommentState, target?: CommentTarg
return record;
}
const NAVIGATION_GROUPS: NavigationGroup[] = ["added", "removed", "modified"];
const NAVIGATION_KINDS: NavigationKind[] = ["push", "replace", "navigate", "dismissTo", "redirect"];
/** "/p/[d]" for the route file app/(tabs)/(patterns)/p/[d].tsx: the route without its groups, which a dynamic route's variants share. */
function routePattern(route: NonNullable<Surface["head"]>): string {
const path = `/${route.logicalPath.replace(/^\/+/, "")}`.split("/").filter(segment => !/^\(.*\)$/.test(segment)).join("/");
return plainText(path.replace(/\/index$/, ""), MAX_ROUTE_LENGTH) || "/";
}
/**
* The map the comment draws: every route added, removed or changed visually (a screen whose source changed but looks
* the same is not one), the routes one step before them, and the transitions among those. A route's appearances and
* variants are one node. Switching tabs reaches every tab from every other, so tab transitions are not edges: the
* routes a tab switch shows are grouped instead. Removed routes and the transitions into them come from the base.
*/
function navigationOf(review: PublicReview, groups: Map<string, ChangeGroup>): CommentNavigation | undefined {
if (!review.navigation) return undefined;
const keyOf = new Map<string, string>();
const groupOf = new Map<string, NavigationGroup>();
const omitted = neverCapturedSurfaceIds(review);
for (const surface of review.surfaces) {
const route = surface.head ?? surface.base;
if (!route || omitted.has(surface.id)) continue;
const key = routePattern(route);
keyOf.set(surface.id.replace(/@(light|dark)$/, ""), key);
const group = groups.get(surface.id);
if (group !== "added" && group !== "removed" && group !== "modified") continue;
const current = groupOf.get(key);
if (!current || NAVIGATION_GROUPS.indexOf(group) < NAVIGATION_GROUPS.indexOf(current)) groupOf.set(key, group);
}
if (!groupOf.size || groupOf.size > MAX_NAVIGATION_NODES) return undefined;
const tabs = new Set<string>();
const links = new Map<string, { from: string; to: string; kinds: Set<NavigationKind>; base: boolean }>();
const add = (edge: NonNullable<PublicReview["navigation"]>["head"]["edges"][number], base: boolean) => {
const from = keyOf.get(edge.from), to = keyOf.get(edge.to);
if (!from || !to || from === to) return;
if (edge.kind === "tab") { tabs.add(to); return; }
const link = links.get(`${from}\n${to}`) ?? { from, to, kinds: new Set(), base };
link.kinds.add(edge.kind);
links.set(`${from}\n${to}`, link);
};
for (const edge of review.navigation.head.edges) add(edge, false);
for (const edge of review.navigation.base.edges) if (groupOf.get(keyOf.get(edge.to) ?? "") === "removed") add(edge, true);
// The routes that lead to the most changed ones first, while the map has room.
const leads = new Map<string, number>();
for (const link of links.values()) if (groupOf.has(link.to) && !groupOf.has(link.from)) leads.set(link.from, (leads.get(link.from) ?? 0) + 1);
const before = [...leads].sort(([a, m], [b, n]) => n - m || (a < b ? -1 : 1)).slice(0, MAX_NAVIGATION_NODES - groupOf.size).map(([key]) => key);
const keys = [...groupOf.keys(), ...before].sort();
const index = new Map(keys.map((key, position) => [key, position]));
const edges = [...links.values()].filter(link => index.has(link.from) && index.has(link.to))
.sort((a, b) => Number(groupOf.has(b.to)) - Number(groupOf.has(a.to)) || index.get(a.from)! - index.get(b.from)! || index.get(a.to)! - index.get(b.to)!)
.slice(0, MAX_NAVIGATION_EDGES)
.sort((a, b) => index.get(a.from)! - index.get(b.from)! || index.get(a.to)! - index.get(b.to)!)
.map(link => ({ from: index.get(link.from)!, to: index.get(link.to)!, kinds: NAVIGATION_KINDS.filter(kind => link.kinds.has(kind)), ...(link.base ? { base: true as const } : {}) }));
if (!edges.length) return undefined;
return { nodes: keys.map(key => ({ route: key, ...(groupOf.has(key) ? { group: groupOf.get(key)! } : {}), ...(tabs.has(key) ? { tab: true as const } : {}) })), edges };
}
/** A route as a quoted Mermaid label, with Mermaid's entity codes for what would end the label, the line or the fence. */
function mermaidLabel(route: string): string {
return `"${plainText(route, MAX_ROUTE_LENGTH).replace(/#/g, "#35;").replace(/"/g, "#quot;").replace(/</g, "#lt;").replace(/>/g, "#gt;").replace(/`/g, "#96;")}"`;
}
/** Outlines, not fills: GitHub draws Mermaid in the reader's theme, and a fixed fill is wrong in one of them. */
const MAP_CLASSES: Record<NavigationGroup, string> = {
added: "stroke:#1a7f37,stroke-width:3px",
modified: "stroke:#bc4c00,stroke-width:3px",
removed: "stroke:#cf222e,stroke-width:2px,stroke-dasharray:5 4",
};
function mermaidMap(navigation: CommentNavigation): string[] {
const node = (index: number) => `n${index}[${mermaidLabel(navigation.nodes[index]!.route)}]`;
const tabs = navigation.nodes.flatMap((entry, index) => entry.tab ? [index] : []);
const lines = ["```mermaid", "flowchart LR"];
if (tabs.length) lines.push(` subgraph tabs["Tabs"]`, ...tabs.map(index => ` ${node(index)}`), " end");
lines.push(...navigation.nodes.flatMap((entry, index) => entry.tab ? [] : [` ${node(index)}`]));
lines.push(...navigation.edges.map(edge => ` n${edge.from} ${edge.base ? "-.->" : "-->"}|${edge.kinds.join(", ")}| n${edge.to}`));
for (const group of ["added", "modified", "removed"] as const) {
const members = navigation.nodes.flatMap((entry, index) => entry.group === group ? [`n${index}`] : []);
if (members.length) lines.push(` classDef ${group} ${MAP_CLASSES[group]}`, ` class ${members.join(",")} ${group}`);
}
return [...lines, "```"];
}
/** One collapsed map per distinct navigation: targets with the same routes share one, named when the targets' maps differ. */
function mapSection(ready: PlatformRecord[]): string[] {
const maps = new Map<string, { navigation: CommentNavigation; records: PlatformRecord[] }>();
for (const record of ready) {
if (!record.navigation) continue;
const key = JSON.stringify(record.navigation);
maps.set(key, { navigation: record.navigation, records: [...(maps.get(key)?.records ?? []), record] });
}
return [...maps.values()].flatMap(({ navigation, records }) => {
const named = maps.size > 1 || records.length < ready.length ? `${records.map(labelOf).join(" · ")} · ` : "";
return details(`${named}Map (${plural(navigation.nodes.length, "screen")})`, ["", ...mermaidMap(navigation), "",
"Outlined: green added, orange changed, red dashed removed; the others lead to them. Tabs groups the screens a tab switch shows.", ""]);
});
}
const commitLink = (repository: string, sha: string) => link(inlineText(sha.slice(0, 8), 8), `https://github.com/${repository.split("/").map(encodeURIComponent).join("/")}/commit/${encodeURIComponent(sha)}`);
function surfaceLink(report: string, id: string): string | undefined {
@@ -345,213 +464,210 @@ function surfaceLink(report: string, id: string): string | undefined {
return url.href;
}
function resultCell(record: PlatformRecord): string {
switch (record.phase) {
case "pending": return "⏳ starting";
case "capturing": {
const capture = record.capture;
if (!capture) return "⏳ capturing";
const side = capture.current ?? (capture.done.head > 0 || capture.baseFromCache ? "head" : "base");
return `⏳ capturing ${side} ${capture.done[side]}/${capture.planned[side]}${capture.baseFromCache && side === "head" ? " · base reused" : ""}`;
}
case "failed": return `🟥 failed · ${inlineText(record.failure?.step ?? "capture", 48)}`;
case "ready": {
const unavailable = record.health ? record.health.base.unavailable + record.health.head.unavailable : 0;
const unverified = record.health ? record.health.base.unverified + record.health.head.unverified : 0;
const failed = (record.validation?.visual?.failed ?? 0) + (record.validation?.accessibility?.failed ?? 0);
const notCompared = record.notCompared ?? 0;
const icon = unavailable || unverified || failed || notCompared ? "🟨" : "🟩";
return [`${icon} ${changeCountText(record.changes?.map(change => change.group) ?? [])}`, notCompared ? `${notCompared} not compared` : undefined, unavailable ? `${unavailable} unavailable` : undefined, unverified ? `${unverified} unattributed` : undefined,
failed ? `${failed} comparison${failed === 1 ? "" : "s"} failed` : undefined, duration(record.startedAt, record.updatedAt)].filter(Boolean).join(" · ");
}
}
}
/** The report link alone; the workflow run and the report zip sit in the footer, where someone debugging looks for them. */
function reportCell(record: PlatformRecord): string {
const run = link("run", record.links.run);
if (record.phase === "failed") return `${run} · ${link("evidence", record.links.artifacts)}`;
if (record.phase !== "ready") return run;
return record.links.report ? link(record.links.report.includes(".expo.app/") ? "Open" : "Viewer", record.links.report) : record.links.deploying ? "deploying…" : run;
}
const ROW_LIMIT = 10;
const MAX_FAILURE_MESSAGE = 600;
/** A target's label as Markdown text. */
const labelOf = (record: PlatformRecord) => inlineText(recordLabel(record), 80);
const finished = (record: PlatformRecord) => record.phase === "ready" || record.phase === "failed";
const firstReport = (records: PlatformRecord[]) => records.find(record => record.phase === "ready" && record.links.report)?.links.report;
/** "platform" for the per-platform targets of a config without targets, "device" once a config names its own. */
function targetNoun(records: PlatformRecord[]): "platform" | "device" {
return records.every(record => !record.target && record.label === undefined) ? "platform" : "device";
}
/** "Workflow iOS run · zip; Android run · zip" for the ready targets, with the target named only when there are several. */
function artifactLinks(records: PlatformRecord[]): string | undefined {
function updatedText(records: PlatformRecord[]): string {
const at = Math.max(...records.map(record => record.updatedAt));
return `${new Date(at).toLocaleString("en-US", { month: "short", day: "numeric", hour: "numeric", minute: "2-digit", timeZone: "UTC" })} UTC`;
}
/** "17 screens changed · Review in Explore": what the run found, with the report as the heading's link; until a target finished, what it is doing. */
function title(records: PlatformRecord[], rows: ChangeRow[]): string {
const ready = records.filter(record => record.phase === "ready");
if (!ready.length) return undefined;
const parts = ready.map(record => {
const links = [link("run", record.links.run), record.links.download ? link("zip", record.links.download) : undefined].filter(Boolean).join(" · ");
return ready.length > 1 ? `${labelOf(record)} ${links}` : links;
});
return `Workflow ${parts.join("; ")}`;
}
function deviceCell(record: PlatformRecord): string {
const parts = [record.device ? inlineText(record.device, 80) : undefined,
record.app?.version ? inlineText(`${record.app.name ? `${record.app.name} ` : ""}${record.app.version}${record.app.build ? ` (${record.app.build})` : ""}`, 80) : undefined].filter(Boolean);
return parts.length ? parts.join(" · ") : "—";
}
type ChangeRow = { id: string; route: string; scheme?: string; group: ReviewGroup; description: string; perTarget: Partial<Record<CommentTarget, { group: ReviewGroup; href?: string }>> };
function changeRows(records: PlatformRecord[]): Map<ReviewGroup, ChangeRow[]> {
const rows = new Map<string, ChangeRow>();
for (const record of records) {
if (record.phase !== "ready") continue;
for (const change of record.changes ?? []) {
const href = record.links.report ? surfaceLink(record.links.report, change.id) : undefined;
const row = rows.get(change.id) ?? { id: change.id, route: change.route, group: change.group, description: change.description, perTarget: {}, ...(change.scheme ? { scheme: change.scheme } : {}) };
// A screen accessibility-only on one target and visibly changed on another is listed as the visible change.
if (strongerGroup(row.group, change.group) !== row.group) Object.assign(row, { group: change.group, description: change.description });
row.perTarget[recordTarget(record)] = { group: change.group, ...(href ? { href } : {}) };
rows.set(change.id, row);
}
}
const groups = new Map<ReviewGroup, ChangeRow[]>();
for (const row of rows.values()) groups.set(row.group, [...(groups.get(row.group) ?? []), row]);
return groups;
}
function renderChangeRow(row: ChangeRow, records: PlatformRecord[]): string {
const where = records.flatMap(record => {
const entry = row.perTarget[recordTarget(record)];
if (!entry) return [];
const label = labelOf(record) + (entry.group !== row.group ? ` (${GROUP_LABEL[entry.group].replace(/^\S+ /, "").toLowerCase()})` : "");
return [entry.href ? link(label, entry.href) : label];
});
const route = `\`${plainText(row.route, MAX_ROUTE_LENGTH).replace(/`/g, "")}\`${row.scheme ? ` (${inlineText(row.scheme, 20)})` : ""}`;
return `- ${route} — ${inlineText(row.description, MAX_DESCRIPTION_LENGTH)}${where.length ? ` · ${where.join(" · ")}` : ""}`;
}
function changesSection(records: PlatformRecord[], rowCap: number): string[] {
const ready = records.filter(record => record.phase === "ready");
if (!ready.length) return [];
const groups = changeRows(ready);
const lines: string[] = [];
const noun = targetNoun(records);
const total = [...groups.values()].reduce((sum, rows) => sum + rows.length, 0);
lines.push(`**Changes** · ${plural(total, "screen")}${records.length > 1 && ready.length > 1 ? ` · a screen is listed once, with a link per ${noun}` : ""}`, "");
// With screens left uncompared, nothing changed only among the ones compared.
if (total === 0) lines.push(ready.some(record => record.notCompared) ? `No compared screen changed on the reviewed ${noun}s.` : `No screen changed on the reviewed ${noun}s.`, "");
for (const group of GROUP_ORDER) {
const rows = groups.get(group) ?? [];
if (!rows.length) continue;
const shown = rows.slice(0, rowCap);
const content = ["", ...(group === "accessibility-only" ? ["Before and After look the same. What screen readers and voice control read changed.", ""] : []),
...shown.map(row => renderChangeRow(row, records))];
if (rows.length > shown.length) {
const report = ready.find(record => record.links.report)?.links.report;
content.push(`- ${rows.length - shown.length} more${report ? ` — ${link("Explore report", report)}` : ""}`);
}
content.push("");
lines.push(...details(`${GROUP_LABEL[group]} (${rows.length})`, content));
}
const seen = new Set<string>();
for (const record of ready) for (const line of record.context ?? []) {
if (seen.has(line)) continue;
seen.add(line);
const everywhere = ready.every(other => other.context?.includes(line));
lines.push(everywhere ? line : `${labelOf(record)} · ${line}`, "");
}
return lines;
}
/**
* The heading carries the report links: on a phone the table's last column is clipped, and the
* commits it reviews are a subtitle under it. A target still capturing adds its link when it is ready.
*/
function heading(records: PlatformRecord[]): string {
const ready = records.filter(record => record.phase === "ready" && record.links.report);
if (!ready.length) {
return records.some(record => record.phase === "ready" && record.links.deploying) ? "### Explore review · report deploying, not ready yet" : "### Explore review";
if (records.every(record => record.phase === "failed")) return "### Explore review · capture failed";
const active = records.filter(record => !finished(record));
const verb = active.some(record => record.phase === "capturing") ? "capturing" : "starting";
return `### Explore review · ${verb}${records.length > 1 ? ` on ${plural(active.length, targetNoun(records))}` : ""}`;
}
const links = ready.length === 1 ? [link("Open report", ready[0]!.links.report!)]
: ready.map(record => link(`${labelOf(record)} report`, record.links.report!));
return `### Explore review · ${links.join(" · ")}`;
// With screens left uncompared, nothing changed only among the ones compared.
const summary = rows.length ? `${plural(rows.length, "screen")} changed` : ready.some(record => gaps(record).length) ? "No compared screens changed" : "No screens changed";
// An appearance a pass never captured is said in the heading too: the verdict above says nothing about it.
const missing = [...new Set(ready.flatMap(record => (record.sessionFailures ?? []).map(failure => failure.scheme)))].map(scheme => `${scheme} not captured`);
const report = firstReport(ready);
const open = report ? link("Review in Explore", report) : ready.some(record => record.links.deploying) ? "report deploying…" : undefined;
return `### ${[summary, ...missing, ...(open ? [open] : [])].join(" · ")}`;
}
function validationCell(record: PlatformRecord, check: "capture" | "routes" | "visual" | "accessibility" | "inventory"): string {
if (record.phase === "pending" || record.phase === "capturing") return "⏳";
if (record.phase === "failed") {
if (check !== "capture") return "—";
const capture = record.capture;
return capture ? `🟥 ${capture.done.base + capture.done.head}/${capture.planned.base + capture.planned.head} surfaces` : "🟥 not started";
}
const health = record.health;
switch (check) {
case "capture": {
const planned = health ? health.base.planned + health.head.planned : (record.capture?.planned.base ?? 0) + (record.capture?.planned.head ?? 0);
const done = health ? health.base.captured + health.head.captured : (record.capture?.done.base ?? 0) + (record.capture?.done.head ?? 0);
const cache = record.capture?.baseFromCache ? " · base cache reused" : "";
return done < planned ? `🟨 ${done}/${planned} surfaces · ${planned - done} unavailable${cache}` : `🟩 ${done}/${planned} surfaces${cache}`;
}
case "routes": {
if (!health) return "—";
const verified = health.base.verified + health.head.verified, planned = health.base.planned + health.head.planned;
const unverified = health.base.unverified + health.head.unverified, unavailable = health.base.unavailable + health.head.unavailable;
return `${verified === planned ? "🟩" : "🟨"} ${verified}/${planned} verified${unverified ? ` · ${unverified} unattributed` : ""}${unavailable ? ` · ${unavailable} unavailable` : ""}`;
}
case "visual": {
const v = record.validation?.visual;
if (!v) return "🟨 not run";
const mark = v.compared === 0 || v.failed > 0 || v.skipped ? "🟨" : "🟩";
return `${mark} ${v.compared} compared · threshold ${v.threshold}${v.failed ? ` · ${v.failed} failed` : ""}${v.skipped ? ` · ${v.skipped} skipped` : ""}`;
}
case "accessibility": {
const a = record.validation?.accessibility;
if (!a) return "🟨 not run";
return `${a.compared === 0 || a.failed > 0 ? "🟨" : "🟩"} ${a.compared} compared · ${plural(a.described, "description")}${a.failed ? ` · ${a.failed} failed` : ""}`;
}
case "inventory": return record.validation?.routes !== undefined ? `🟩 ${plural(record.validation.routes, "route")}` : "🟩 analyzed";
}
/** Every target by name, each linking its own report, then the commits and when the comment last changed. */
function subtitle(records: PlatformRecord[], repository: string): string {
const first = records[0]!;
const targets = records.map(record => record.phase === "ready" && record.links.report ? link(labelOf(record), record.links.report) : labelOf(record));
return `<sub>${[...targets, `${commitLink(repository, first.headSha)} vs ${commitLink(repository, first.baseSha)}`, `updated ${updatedText(records)}`].join(" · ")}</sub>`;
}
function validationSummary(record: PlatformRecord): string {
/** One line for a target that is not simply done: still working, failed, or finished with captures missing. */
function statusLine(record: PlatformRecord): string | undefined {
const label = labelOf(record), run = link("run", record.links.run);
switch (record.phase) {
case "failed": return "🟥 failed";
case "pending": return `${label} · starting · ${run}`;
case "capturing": {
const capture = record.capture;
if (!capture) return `${label} · capturing · ${run}`;
const side = capture.current ?? (capture.done.head > 0 || capture.baseFromCache ? "head" : "base");
return `${label} · capturing ${side} ${capture.done[side]}/${capture.planned[side]}${capture.baseFromCache && side === "head" ? " · base reused" : ""} · ${run}`;
}
case "failed": return [`**${label} failed**`, inlineText(record.failure?.step ?? "capture", 48), record.failure?.message ? inlineText(record.failure.message, MAX_FAILURE_MESSAGE) : undefined,
run, link("evidence", record.links.artifacts)].filter(Boolean).join(" · ");
case "ready": {
const health = record.health;
if (!health) return "🟩 reviewed";
const captured = health.base.captured + health.head.captured, planned = health.base.planned + health.head.planned;
return `${captured < planned ? "🟨" : "🟩"} ${captured}/${planned} captured`;
const parts = gaps(record);
return parts.length ? `${label} · ${parts.join(" · ")} · ${run}` : undefined;
}
default: return "⏳ in progress";
}
}
function validationSection(byTarget: PlatformRecord[]): string[] {
if (!byTarget.some(record => record.phase === "ready" || record.phase === "failed")) return [];
const summary = `Validation · ${byTarget.map(record => `${labelOf(record)} ${validationSummary(record)}`).join(" · ")}`;
const header = `| Check | ${byTarget.map(labelOf).join(" | ")} |`;
const checks: [string, Parameters<typeof validationCell>[1]][] = [["Capture", "capture"], ["Route verification", "routes"], ["Visual comparison", "visual"], ["Accessibility comparison", "accessibility"], ["Source inventory", "inventory"]];
const content = ["", header, `|---|${byTarget.map(() => "---").join("|")}|`,
...checks.map(([label, check]) => `| ${label} | ${byTarget.map(record => validationCell(record, check)).join(" | ")} |`), ""];
for (const record of byTarget) {
if (record.phase === "failed" && record.failure) {
content.push(`**${labelOf(record)} failed** — ${inlineText(record.failure.message, 1200)} ${link("Open the failed run", record.links.run)} · ${link("Evidence artifact", record.links.artifacts)}`, "");
}
if (record.phase !== "ready") continue;
const health = record.health;
const gaps = health ? health.base.unavailable + health.base.unverified + health.head.unavailable + health.head.unverified : 0;
if (!gaps && !record.diagnostics?.length) continue;
const sideLine = (side: "base" | "head") => { const h = health![side]; return `${side} ${h.verified}/${h.planned} verified${h.unverified ? ` · ${h.unverified} unattributed` : ""}${h.unavailable ? ` · ${h.unavailable} unavailable` : ""}`; };
content.push(health ? `**${labelOf(record)}** · ${sideLine("base")} · ${sideLine("head")}` : `**${labelOf(record)}**`, "");
/** "dark not captured" for an appearance whose pass failed on head (or both sides), "… on base" when only the base's did. */
function missingAppearances(record: PlatformRecord): string[] {
const failures = record.sessionFailures ?? [];
return [...new Set(failures.map(failure => failure.scheme))].map(scheme => {
const sides = new Set(failures.filter(failure => failure.scheme === scheme).map(failure => failure.side));
return `${scheme} not captured${sides.size === 1 && sides.has("base") ? " on base" : ""}`;
});
}
/** What a finished target has no verdict on: captures missing or unattributed, comparisons that did not run or failed. */
function gaps(record: PlatformRecord): string[] {
const unavailable = record.health ? record.health.base.unavailable + record.health.head.unavailable : 0;
const unverified = record.health ? record.health.base.unverified + record.health.head.unverified : 0;
const failed = (record.validation?.visual?.failed ?? 0) + (record.validation?.accessibility?.failed ?? 0);
return [...missingAppearances(record), unavailable ? `${plural(unavailable, "capture")} unavailable` : undefined, unverified ? `${unverified} unattributed` : undefined,
record.notCompared ? `${record.notCompared} not compared` : undefined, failed ? `${plural(failed, "comparison")} failed` : undefined]
.filter((part): part is string => Boolean(part));
}
/** The per-screen capture problems, collapsed: the status line already counts them. */
function diagnosticsSection(records: PlatformRecord[]): string[] {
const flagged = records.filter(record => record.phase === "ready" && ((record.diagnostics?.length ?? 0) > 0 || (record.diagnosticCount ?? 0) > 0));
if (!flagged.length) return [];
const content: string[] = [""];
for (const record of flagged) {
if (records.length > 1) content.push(`**${labelOf(record)}**`, "");
for (const diagnostic of record.diagnostics ?? []) {
content.push(`- ${diagnostic.side ? `${diagnostic.side === "base" ? "Base" : "Head"} · ` : ""}${inlineText(diagnostic.route, MAX_ROUTE_LENGTH)} · ${inlineText(diagnostic.code, 80)}: ${inlineText(diagnostic.message, 160)} Next action: ${inlineText(diagnostic.nextAction, 240)}`);
}
const more = (record.diagnosticCount ?? 0) - (record.diagnostics?.length ?? 0);
if (more > 0) content.push(`- ${more} more — ${link("View run", record.links.run)}`);
if (record.diagnostics?.length) content.push("");
if (more > 0) content.push(`- ${more} more — ${link("run", record.links.run)}`);
content.push("");
}
return details(summary, content);
const total = flagged.reduce((sum, record) => sum + Math.max(record.diagnosticCount ?? 0, record.diagnostics?.length ?? 0), 0);
return details(`Capture issues (${total})`, content);
}
type ChangeRow = { id: string; route: string; scheme?: string; group: ReviewGroup; description: string; href?: string; targets: PlatformRecord[] };
/** Every changed screen once, in group order (added, removed, modified, …), linking the first target in config order that shows its group. */
function changeRows(ready: PlatformRecord[]): ChangeRow[] {
const rows = new Map<string, ChangeRow>();
for (const record of ready) {
for (const change of record.changes ?? []) {
const href = record.links.report ? surfaceLink(record.links.report, change.id) : undefined;
const row = rows.get(change.id) ?? { id: change.id, route: change.route, group: change.group, description: change.description, targets: [], ...(change.scheme ? { scheme: change.scheme } : {}) };
// A screen accessibility-only on one target and visibly changed on another is listed as the visible change.
// The link opens the target the row describes.
if (strongerGroup(row.group, change.group) !== row.group) Object.assign(row, { group: change.group, description: change.description, href });
else if (!row.href && href && change.group === row.group) row.href = href;
row.targets.push(record);
rows.set(change.id, row);
}
}
return [...rows.values()].sort((a, b) => GROUP_ORDER.indexOf(a.group) - GROUP_ORDER.indexOf(b.group));
}
function renderChangeRow(row: ChangeRow, ready: PlatformRecord[]): string {
const code = `\`${plainText(row.route, MAX_ROUTE_LENGTH).replace(/`/g, "")}\``;
const route = `${row.href ? link(code, row.href) : code}${row.scheme ? ` (${inlineText(row.scheme, 20)})` : ""}`;
const description = inlineText(row.description.replace(/\.$/, ""), MAX_DESCRIPTION_LENGTH);
let text: string;
if (row.group === "added" || row.group === "removed") {
const verb = row.group === "added" ? "Added" : "Removed";
// The default description repeats the verb and route; an agent's note does not.
text = row.description === `${verb} ${row.route} screen` ? `${verb} ${route}` : `${verb} ${route} · ${description}`;
} else {
text = `${route} · ${description}`;
}
const tag = row.group === "accessibility-only" ? "accessibility only" : row.group === "needs-review" ? "needs review" : undefined;
return `- ${[text, tag, ...whereNote(row, ready)].filter(Boolean).join(" · ")}`;
}
/**
* Where a screen's change shows. A target whose screenshots match because the changed elements lie past the edge of the
* screen has the change, off-screen, so it is named apart: "iPhone 17, Android only · off-screen on iPhone Duo (closed)"
* for Tempo's footnote below the fold of its sheet in brentvatne/euxy#172 (review run 39362642343). The targets that show
* it are still named the shorter way among the others (none when all of them do); without such a target the note is what
* it always was.
*/
function whereNote(row: ChangeRow, ready: PlatformRecord[]): string[] {
const offScreen = row.group === "modified"
? row.targets.filter(record => record.changes?.some(change => change.id === row.id && change.offScreen === true)) : [];
const shown = row.targets.filter(record => !offScreen.includes(record));
// A target with screens it has no verdict on may have changed this one too: name where it changed, not where it did not.
const unsure = ready.some(record => !row.targets.includes(record) && gaps(record).length);
const visible = unsure ? `on ${shown.map(labelOf).join(", ")}` : targetsNote(shown, ready.filter(record => !offScreen.includes(record)));
return [visible, offScreen.length ? `off-screen on ${offScreen.map(labelOf).join(", ")}` : undefined].filter((part): part is string => Boolean(part));
}
/** "Android only" / "not on Android": the targets a screen changed on, when not all of them, named the shorter way. */
function targetsNote(on: PlatformRecord[], ready: PlatformRecord[]): string | undefined {
if (ready.length < 2 || on.length >= ready.length) return undefined;
const missing = ready.filter(record => !on.includes(record));
return missing.length < on.length ? `not on ${missing.map(labelOf).join(", ")}` : `${on.map(labelOf).join(", ")} only`;
}
function changesSection(ready: PlatformRecord[], rows: ChangeRow[], rowCap: number, withMap: boolean): string[] {
if (!ready.length) return [];
const lines: string[] = [];
const shown = rows.slice(0, rowCap);
lines.push(...shown.map(row => renderChangeRow(row, ready)));
if (rows.length > shown.length) {
const report = firstReport(ready);
const more = `${plural(rows.length - shown.length, "more screen")} in the report`;
lines.push(`- ${report ? link(more, report) : more}`);
}
if (lines.length) lines.push("");
if (withMap) lines.push(...mapSection(ready));
const byLine = new Map<string, PlatformRecord[]>();
for (const record of ready) for (const line of record.context ?? []) byLine.set(line, [...(byLine.get(line) ?? []), record]);
for (const [line, records] of byLine) lines.push(records.length === ready.length ? line : `${line} · ${targetsNote(records, ready)}`, "");
return lines;
}
/**
* The previous run stays on the comment while a new one works or has only failed; this line, above it, says which,
* with each failure's step and message.
*/
function newRunLine(current: PlatformRecord[], shown: PlatformRecord[], repository: string): string {
const first = current[0]!;
const active = current.filter(record => !finished(record));
const failed = current.filter(record => record.phase === "failed");
const named = current.length > 1 || shown.length > 1 || Boolean(first.target) || first.label !== undefined;
const verb = active.some(record => record.phase === "capturing") ? "capturing" : "starting";
const working = active.length ? [`${verb}${named ? ` on ${active.map(labelOf).join(", ")}` : ""}`] : [];
const failures = failed.map(record => [named ? `**${labelOf(record)} failed**` : undefined, inlineText(record.failure?.step ?? "capture", 48),
record.failure?.message ? inlineText(record.failure.message, MAX_FAILURE_MESSAGE) : undefined].filter(Boolean).join(" · "));
const links = [link("run", first.links.run), ...(failed.length ? [link("evidence", failed[0]!.links.artifacts)] : [])];
return `> **${active.length ? "New run in progress" : "New run failed"}** for ${commitLink(repository, first.headSha)} · ${[...working, ...failures, ...links].join(" · ")}`;
}
function footer(records: PlatformRecord[]): string {
const parts: string[] = [];
const expiries = records.filter(record => record.phase === "ready" && record.links.expiresAt !== undefined).map(record => record.links.expiresAt);
if (expiries.length) parts.push(expiries.some(at => at === null) ? "Preview links work for anyone who has them until the deployment is deleted"
: `Preview links work for anyone who has them until ${previewExpiry(Math.min(...expiries.filter((at): at is number => at !== null)))}`);
const runs = records.filter((record, index, all) => all.findIndex(other => other.links.run === record.links.run) === index);
parts.push(runs.length === 1 ? link("Workflow run", runs[0]!.links.run) : runs.map(record => link(`${labelOf(record)} run`, record.links.run)).join(" · "));
parts.push(link("Config", records[0]!.links.config));
return `<sub>${parts.join(" · ")}</sub>`;
}
function historySection(repository: string, history: HistoryEntry[]): string[] {
@@ -580,42 +696,47 @@ function deploymentOf(url: string | undefined): string | undefined {
}
function deploymentMarkerOf(records: CommentRecords): string | undefined {
const names = new Set<string>();
for (const record of Object.values(records.current)) { const name = deploymentOf(record?.links.report); if (name) names.add(name); }
for (const record of [...Object.values(records.current), ...Object.values(records.previous ?? {})]) { const name = deploymentOf(record?.links.report); if (name) names.add(name); }
for (const entry of records.history) for (const item of Object.values(entry.platforms)) { const name = deploymentOf(item?.report); if (name) names.add(name); }
return names.size ? `<!-- expo-explore-deployments:${[...names].sort().join(",")} -->` : undefined;
}
/** The whole comment from its records: hidden lines first, then one review for every target of the current run, in config order. */
/**
* The whole comment from its records: hidden lines first, then the review. Until a target of the current run is ready,
* the previous run's review stays up under a line that says a new run is working or failed.
*/
export function renderCommentRecords(records: CommentRecords, existing?: string, budget = PUBLIC_COMMENT_BODY_BUDGET): string {
const current = orderedRecords(records);
const previous = orderedPrevious(records);
const hidden = [CI_COMMENT_MARKER, ...[deploymentMarkerOf(records)].filter((line): line is string => Boolean(line)),
...current.map(recordLine), ...[historyLine(records.history)].filter((line): line is string => Boolean(line))];
...current.map(recordLine), ...previous.map(previousLine), ...[historyLine(records.history)].filter((line): line is string => Boolean(line))];
if (!current.length) return [...hidden, "### Explore review", "", ...(records.history.some(entry => Object.values(entry.platforms).some(item => item?.superseded)) ? [SUPERSEDED_NOTE, ""] : []),
...historySection("", records.history)].join("\n");
const first = current[0]!;
const shown = current.some(record => record.phase === "ready") || !previous.length ? current : previous;
const first = shown[0]!;
const repository = repositoryOf(first.links.run);
const render = (rowCap: number): string => {
const lines: string[] = [...hidden, heading(current), "",
`<sub>base ${commitLink(repository, first.baseSha)} → ${commitLink(repository, first.headSha)}</sub>`, ""];
const ready = shown.filter(record => record.phase === "ready");
const rows = changeRows(ready);
const render = (rowCap: number, withMap: boolean): string => {
const lines: string[] = [...hidden];
if (shown !== current) lines.push(newRunLine(current, shown, repository), "");
lines.push(title(shown, rows), "", subtitle(shown, repository), "");
if (first.superseded) lines.push(SUPERSEDED_NOTE, "");
lines.push(`| | Device | Result | Report |`, `|---|---|---|---|`,
...current.map(record => `| ${labelOf(record)} | ${deviceCell(record)} | ${resultCell(record)} | ${reportCell(record)} |`), "");
lines.push(...changesSection(current, rowCap));
lines.push(...validationSection(current));
for (const line of shown.map(statusLine)) if (line) lines.push(line, "");
lines.push(...changesSection(ready, rows, rowCap, withMap));
lines.push(...diagnosticsSection(shown));
lines.push(...historySection(repository, records.history));
const expiries = current.filter(record => record.phase === "ready" && record.links.expiresAt !== undefined).map(record => record.links.expiresAt);
const footer: string[] = [];
if (expiries.length) footer.push(expiries.some(at => at === null) ? "Preview links work for anyone who has them until the deployment is deleted."
: `Preview links work until ${previewExpiry(Math.min(...expiries.filter((at): at is number => at !== null)))} for anyone who has them.`);
footer.push(`Managed via ${link("`" + plainText(configPathOf(first.links.config), 120).replace(/`/g, "") + "`", first.links.config)}`);
const artifacts = artifactLinks(current);
if (artifacts) footer.push(artifacts);
lines.push(footer.join(" · "));
lines.push(footer(shown));
return lines.join("\n");
};
let rowCap = MAX_ROWS_PER_GROUP;
let body = render(rowCap);
while (body.length > budget && rowCap > 1) { rowCap = Math.max(1, Math.floor(rowCap / 2)); body = render(rowCap); }
// The map goes first when the body is over budget, then the list shrinks, then the previous run's records go; once the
// current run shows, nothing renders them, so they go first.
let rowCap = ROW_LIMIT;
let body = render(rowCap, true);
if (body.length > budget && shown === current && previous.length) return renderCommentRecords({ ...records, previous: {} }, existing, budget);
if (body.length > budget) body = render(rowCap, false);
while (body.length > budget && rowCap > 1) { rowCap = Math.max(1, Math.floor(rowCap / 2)); body = render(rowCap, false); }
if (body.length > budget && previous.length) return renderCommentRecords({ ...records, previous: {} }, existing, budget);
return body;
}
@@ -623,10 +744,6 @@ function repositoryOf(run: string): string {
const match = /^https:\/\/github\.com\/([^/]+)\/([^/]+)\/actions\//.exec(run);
return match ? `${decodeURIComponent(match[1]!)}/${decodeURIComponent(match[2]!)}` : "";
}
function configPathOf(config: string): string {
const match = /\/blob\/[0-9a-f]{40}\/(.+)$/.exec(config);
return match ? decodeURIComponent(match[1]!) : "config";
}
/**
* Render the sticky pull-request comment for any phase of a public review run, folding this writer into the existing
@@ -654,8 +771,9 @@ export function reviewDevice(state: { runtime?: Pick<SimulatorRuntimeMetadata, "
export function reviewChanges(review: PublicReview): Map<string, ChangeGroup> {
const evidenceBased = review.surfaces.some(surface => surface.review);
const changes = new Map<string, ChangeGroup>();
const omitted = neverCapturedSurfaceIds(review);
for (const surface of review.surfaces) {
if ((surface.head ?? surface.base)?.kind === "redirect") continue;
if (omitted.has(surface.id)) continue;
const group = groupOf(surface, evidenceBased);
if (group !== "unlisted") changes.set(surface.id, group);
}
+120 -16
View File
@@ -56,6 +56,23 @@ export function changeCountText(groups: readonly ChangeGroup[]): string {
return [visual ? `${visual} visual change${visual === 1 ? "" : "s"}` : undefined, `${accessibility} accessibility only`].filter(Boolean).join(" · ");
}
export type NavigationGroup = "added" | "modified" | "removed";
export type NavigationKind = "push" | "replace" | "navigate" | "redirect" | "tab" | "dismissTo";
const NAVIGATION_GROUPS: NavigationGroup[] = ["added", "modified", "removed"];
const NAVIGATION_KINDS: NavigationKind[] = ["push", "replace", "navigate", "redirect", "tab", "dismissTo"];
export const MAX_NAVIGATION_NODES = 30;
export const MAX_NAVIGATION_EDGES = 80;
/**
* The comment's map: one node per route that was added, removed or changed visually, and per route one step before
* one of them; edges index `nodes`. `tab` marks a route a tab switch shows, `base` an edge into a removed route, which
* only the base revision's navigation has.
*/
export type CommentNavigation = {
nodes: { route: string; group?: NavigationGroup; tab?: true }[];
edges: { from: number; to: number; kinds: NavigationKind[]; base?: true }[];
};
export type SideHealth = { planned: number; captured: number; verified: number; unverified: number; unavailable: number };
export type PlatformRecord = {
@@ -86,14 +103,20 @@ export type PlatformRecord = {
accessibility?: { compared: number; described: number; failed: number };
routes?: number;
};
/** The review's changed surfaces, one per surface; `scheme` for a secondary appearance (`<route>@<scheme>`). */
changes?: { id: string; route: string; group: ChangeGroup; description: string; scheme?: string }[];
/**
* The review's changed surfaces, one per surface; `scheme` for a secondary appearance (`<route>@<scheme>`). `offScreen`
* marks an accessibility-only change whose elements lie past the edge of the screen.
*/
changes?: { id: string; route: string; group: ChangeGroup; description: string; scheme?: string; offScreen?: true }[];
/** Screens whose source changed and whose captures were verified, but that got no verdict: the comparison did not run or failed. */
notCompared?: number;
/** Differences stated once rather than per screen (variants, app-wide source changes). */
context?: string[];
navigation?: CommentNavigation;
diagnostics?: { side?: "base" | "head"; route: string; code: string; message: string; nextAction: string }[];
diagnosticCount?: number;
/** Appearance passes that failed while the primary one captured: that appearance was not captured on that side. */
sessionFailures?: { side: "base" | "head"; scheme: "light" | "dark"; frames: number }[];
failure?: { step: string; message: string };
/** The PR moved while this run worked; its report stays reachable but it is not the current review. */
superseded?: boolean;
@@ -109,7 +132,12 @@ export type HistoryEntry = {
};
export type HistoryItem = { phase: RecordPhase; summary: string; run: string; report?: string; download?: string; superseded?: boolean; label?: string; order?: number };
export type CommentRecords = { current: Partial<Record<CommentTarget, PlatformRecord>>; history: HistoryEntry[] };
export type CommentRecords = {
current: Partial<Record<CommentTarget, PlatformRecord>>;
/** The last finished run's records, shown while the current run has no finished record of its own. */
previous?: Partial<Record<CommentTarget, PlatformRecord>>;
history: HistoryEntry[];
};
/** The id a record is keyed by. */
export function recordTarget(record: Pick<PlatformRecord, "platform" | "target">): CommentTarget {
@@ -128,12 +156,22 @@ export function compareTargets(a: { id: CommentTarget; order?: number | undefine
/** The current records in display order. */
export function orderedRecords(records: CommentRecords): PlatformRecord[] {
return Object.values(records.current).filter((record): record is PlatformRecord => Boolean(record))
return ordered(records.current);
}
/** The previous run's records in display order. */
export function orderedPrevious(records: CommentRecords): PlatformRecord[] {
return ordered(records.previous ?? {});
}
function ordered(records: Partial<Record<CommentTarget, PlatformRecord>>): PlatformRecord[] {
return Object.values(records).filter((record): record is PlatformRecord => Boolean(record))
.sort((a, b) => compareTargets({ id: recordTarget(a), order: a.order }, { id: recordTarget(b), order: b.order }));
}
export const HISTORY_LIMIT = 5;
const RECORD_LINE = /^<!-- expo-explore-record:([a-z][a-z0-9-]{0,31}):([A-Za-z0-9_-]+) -->$/gm;
const PREVIOUS_LINE = /^<!-- expo-explore-previous:([a-z][a-z0-9-]{0,31}):([A-Za-z0-9_-]+) -->$/gm;
const HISTORY_LINE = /^<!-- expo-explore-history:([A-Za-z0-9_-]+) -->$/m;
const MAX_ENCODED = 24_000;
const MAX_DECODED = 256_000;
@@ -158,6 +196,10 @@ export function recordLine(record: PlatformRecord): string {
return `<!-- expo-explore-record:${recordTarget(record)}:${encodeRecordPayload(record)} -->`;
}
export function previousLine(record: PlatformRecord): string {
return `<!-- expo-explore-previous:${recordTarget(record)}:${encodeRecordPayload(record)} -->`;
}
export function historyLine(history: HistoryEntry[]): string | undefined {
return history.length ? `<!-- expo-explore-history:${encodeRecordPayload(history)} -->` : undefined;
}
@@ -233,9 +275,12 @@ export function parsePlatformRecord(value: unknown): PlatformRecord | undefined
if (Array.isArray(value.changes)) {
const changes = value.changes.flatMap((entry): NonNullable<PlatformRecord["changes"]> => isRecord(entry) && isString(entry.id, 220) && isString(entry.route, 2000)
&& CHANGE_GROUPS.includes(entry.group as ChangeGroup) && isString(entry.description, 2000)
? [{ id: entry.id, route: entry.route, group: entry.group as ChangeGroup, description: entry.description, ...(isString(entry.scheme, 20) ? { scheme: entry.scheme } : {}) }] : []);
? [{ id: entry.id, route: entry.route, group: entry.group as ChangeGroup, description: entry.description, ...(isString(entry.scheme, 20) ? { scheme: entry.scheme } : {}),
...(entry.offScreen === true && entry.group === "accessibility-only" ? { offScreen: true as const } : {}) }] : []);
if (changes.length <= 2000) record.changes = changes;
}
const navigation = parseNavigation(value.navigation);
if (navigation) record.navigation = navigation;
if (Array.isArray(value.context)) record.context = value.context.filter((entry): entry is string => isString(entry, 2000)).slice(0, 10);
if (Array.isArray(value.diagnostics)) {
record.diagnostics = value.diagnostics.flatMap((entry): NonNullable<PlatformRecord["diagnostics"]> => isRecord(entry) && isString(entry.route, 2000) && isString(entry.code, 200)
@@ -244,11 +289,35 @@ export function parsePlatformRecord(value: unknown): PlatformRecord | undefined
}
if (isCount(value.notCompared) && value.notCompared > 0) record.notCompared = value.notCompared;
if (isCount(value.diagnosticCount)) record.diagnosticCount = value.diagnosticCount;
if (Array.isArray(value.sessionFailures)) {
const failures = value.sessionFailures.flatMap((entry): NonNullable<PlatformRecord["sessionFailures"]> => isRecord(entry) && isSide(entry.side)
&& (entry.scheme === "light" || entry.scheme === "dark") && isCount(entry.frames) ? [{ side: entry.side, scheme: entry.scheme, frames: entry.frames }] : []).slice(0, 4);
if (failures.length) record.sessionFailures = failures;
}
if (isRecord(value.failure) && isString(value.failure.step, 200) && isString(value.failure.message)) record.failure = { step: value.failure.step, message: value.failure.message };
if (value.superseded === true) record.superseded = true;
return record;
}
function parseNavigation(value: unknown): CommentNavigation | undefined {
if (!isRecord(value) || !Array.isArray(value.nodes) || !Array.isArray(value.edges) || value.nodes.length > MAX_NAVIGATION_NODES
|| value.edges.length > MAX_NAVIGATION_EDGES) return undefined;
const nodes: CommentNavigation["nodes"] = [];
for (const node of value.nodes) {
if (!isRecord(node) || !isString(node.route, 200) || !optional(node.group, (group): group is NavigationGroup => NAVIGATION_GROUPS.includes(group as NavigationGroup))
|| !(node.tab === undefined || node.tab === true)) return undefined;
nodes.push({ route: node.route, ...(node.group ? { group: node.group } : {}), ...(node.tab ? { tab: true } : {}) });
}
const edges: CommentNavigation["edges"] = [];
const isNode = (index: unknown): index is number => isCount(index) && index < nodes.length;
for (const edge of value.edges) {
if (!isRecord(edge) || !isNode(edge.from) || !isNode(edge.to) || !Array.isArray(edge.kinds) || !edge.kinds.length
|| !edge.kinds.every(kind => NAVIGATION_KINDS.includes(kind as NavigationKind)) || !(edge.base === undefined || edge.base === true)) return undefined;
edges.push({ from: edge.from, to: edge.to, kinds: [...new Set(edge.kinds as NavigationKind[])], ...(edge.base ? { base: true } : {}) });
}
return nodes.length && edges.length ? { nodes, edges } : undefined;
}
export function parseHistory(value: unknown): HistoryEntry[] {
if (!Array.isArray(value)) return [];
const entries: HistoryEntry[] = [];
@@ -273,12 +342,18 @@ export function parseHistory(value: unknown): HistoryEntry[] {
export function readCommentRecords(body: string | undefined): CommentRecords {
const records: CommentRecords = { current: {}, history: [] };
if (!body) return records;
for (const match of body.matchAll(RECORD_LINE)) {
const record = parsePlatformRecord(decodePayload(match[2]!));
const target = record && recordTarget(record);
if (!record || target !== match[1] || !isTargetId(target)) continue;
if (records.current[target] || Object.keys(records.current).length < MAX_COMMENT_TARGETS) records.current[target] = record;
}
const read = (pattern: RegExp, into: Partial<Record<CommentTarget, PlatformRecord>>) => {
for (const match of body.matchAll(pattern)) {
const record = parsePlatformRecord(decodePayload(match[2]!));
const target = record && recordTarget(record);
if (!record || target !== match[1] || !isTargetId(target)) continue;
if (into[target] || Object.keys(into).length < MAX_COMMENT_TARGETS) into[target] = record;
}
};
read(RECORD_LINE, records.current);
const previous: Partial<Record<CommentTarget, PlatformRecord>> = {};
read(PREVIOUS_LINE, previous);
if (Object.keys(previous).length) records.previous = previous;
const history = body.match(HISTORY_LINE);
if (history) records.history = parseHistory(decodePayload(history[1]!));
return records;
@@ -332,14 +407,43 @@ export function foldRecord(existing: CommentRecords, incoming: PlatformRecord):
const record = withKnownTarget(incoming, existing);
const target = recordTarget(record);
const currentRun = current[0];
if (!currentRun && record.superseded) return { current: {}, history: pushHistory(existing.history, historyEntryOf([record])) };
const previous = existing.previous ?? {};
if (!currentRun && record.superseded) return { current: {}, previous, history: pushHistory(existing.history, historyEntryOf([record])) };
if (!currentRun || currentRun.runId === record.runId) {
return { current: { ...existing.current, [target]: record }, history: existing.history };
const next = { ...existing.current, [target]: record };
// A rerun's new attempt replaces a ready record with one that is not; the ready one stays readable meanwhile.
const replaced = existing.current[target];
const rerun = replaced && isReady(replaced) && !isReady(record) && Number(record.runAttempt) > Number(replaced.runAttempt);
return { current: next, previous: trimPrevious(rerun ? { ...ofRun(previous, record.runId), [target]: replaced } : previous, next), history: existing.history };
}
if (runOrder(currentRun.runId) > runOrder(record.runId) || record.superseded) {
return { current: existing.current, history: pushHistory(existing.history, historyEntryOf([{ ...record, superseded: true }])) };
return { current: existing.current, previous, history: pushHistory(existing.history, historyEntryOf([{ ...record, superseded: true }])) };
}
return { current: { [target]: record }, history: pushHistory(existing.history, historyEntryOf(current)) };
// The review of the run being replaced stays readable until the new run has a ready target. It replaces older
// previous records rather than joining them: what the comment shows as one review comes from one run. A run without a
// ready target has no review, so the one before it stays.
const kept = current.some(isReady) ? Object.fromEntries(current.filter(isFinished).map(item => [recordTarget(item), item])) : previous;
const next = { [target]: record };
return { current: next, previous: trimPrevious(kept, next), history: pushHistory(existing.history, historyEntryOf(current)) };
}
const isFinished = (record: PlatformRecord) => record.phase === "ready" || record.phase === "failed";
const isReady = (record: PlatformRecord) => record.phase === "ready";
function ofRun(records: Partial<Record<CommentTarget, PlatformRecord>>, runId: string): Partial<Record<CommentTarget, PlatformRecord>> {
return Object.fromEntries(Object.entries(records).filter(([, record]) => record?.runId === runId));
}
/**
* A target's previous record goes once its current one is ready; all of them go once every current record finished
* and one is ready. A failure is no review: the previous one stays up beside it.
*/
function trimPrevious(previous: Partial<Record<CommentTarget, PlatformRecord>>, current: Partial<Record<CommentTarget, PlatformRecord>>): Partial<Record<CommentTarget, PlatformRecord>> {
const records = Object.values(current).filter((record): record is PlatformRecord => Boolean(record));
if (records.every(isFinished) && records.some(isReady)) return {};
const trimmed = { ...previous };
for (const record of records) if (isReady(record)) delete trimmed[recordTarget(record)];
return trimmed;
}
/**
@@ -349,7 +453,7 @@ export function foldRecord(existing: CommentRecords, incoming: PlatformRecord):
function withKnownTarget(record: PlatformRecord, existing: CommentRecords): PlatformRecord {
const target = recordTarget(record);
if (!record.target || record.label !== undefined || record.order !== undefined) return record;
const known = existing.current[target];
const known = existing.current[target] ?? existing.previous?.[target];
if (known) return { ...record, platform: known.platform, ...(known.label !== undefined ? { label: known.label } : {}), ...(known.order !== undefined ? { order: known.order } : {}) };
const item = existing.history.find(entry => entry.platforms[target])?.platforms[target];
return item ? { ...record, ...(item.label !== undefined ? { label: item.label } : {}), ...(item.order !== undefined ? { order: item.order } : {}) } : record;
+282 -126
View File
@@ -1,14 +1,16 @@
import { createCipheriv, createHash, randomBytes } from "node:crypto";
import { createHash, randomBytes } from "node:crypto";
import { spawnSync } from "node:child_process";
import { appendFile, cp, lstat, mkdir, mkdtemp, readdir, readFile, realpath, rm, writeFile } from "node:fs/promises";
import { tmpdir } from "node:os";
import { join, relative, resolve, sep } from "node:path";
import { Uint8ArrayReader, Uint8ArrayWriter, ZipWriter, configure } from "@zip.js/zip.js";
import { MAX_METADATA_BYTES } from "./agent-api.js";
import { loadReport, MAX_METADATA_BYTES } from "./agent-api.js";
import { MAX_ASSET_BYTES, manifestBody, PREVIEW_ASSET } from "./preview-manifest.js";
import { publishCiReportComment, readCiReportInput, type CiReportInput } from "./artifact.js";
import { retryGitHubLookup } from "./github-lookup.js";
import { readZipDirectory, readZipEntry } from "./zip-reader.js";
import { reviewChangeCount, reviewDevice } from "./pr-comment-body.js";
import { isTargetId, type ExplorePlatform } from "../config/explore-config.js";
import { isTargetId, warnIgnoredExpiry, type ExplorePlatform } from "../config/explore-config.js";
import type { RunTarget } from "../types.js";
import { targetParameter } from "./viewer-links.js";
import type { ArtifactReport } from "../contracts/artifact-report.js";
@@ -21,21 +23,23 @@ import { TOOL_ROOT, packageBin } from "../shared/package-paths.js";
/**
* `ci:deploy`: publish one token-gated EAS Hosting deployment for a packaged review.
*
* The deployment is the prebuilt preview export of viewer-host (routes `/p/[token]`, see viewer-host/README.md)
* plus the report ZIP encrypted with a key that only the deployment's Worker environment holds. The Worker
* serves the Viewer and the decrypted report at `/p/<token>` and answers 404 to everything else; after
* `expiresAt` (if any) the token page says the link expired. Deployments are never deleted automatically. The report job already published the GitHub artifact links, so a failed deployment leaves
* the comment on the download-and-drop flow.
* The deployment is the prebuilt preview export of viewer-host (routes `/p/[token]`, see viewer-host/README.md) plus
* the report as static files under the link's token: each target's manifest at `p/<token>/targets/<target>.json` and
* every screenshot at `p/<token>/assets/<sha256>.<ext>`. Nothing is encrypted; the 160-bit token in every path is what
* keeps them private. The Worker serves the Viewer at `/p/<token>` and answers 404 to everything else. A deployment
* lasts as long as EAS Hosting keeps it (it deletes deployments without an alias after 30 days on Free, 90 on paid
* plans); Explore sets no expiry of its own. The report job already published the GitHub artifact links, so a failed
* deployment leaves the comment on the download-and-drop flow.
*/
export type CommandRunner = (command: string, args: readonly string[], options: { cwd: string; env: NodeJS.ProcessEnv })
=> { status: number | null; stdout: string; stderr: string };
export type PreviewReportInput = {
/** Set by the matrix workflows; the preview then serves this report at report.zip?platform=<platform>. */
/** Set by the matrix workflows; the preview then serves this report at manifest.json?platform=<platform>. */
platform?: ExplorePlatform;
/**
* The capture target, set by the matrix workflows (wins over `platform`): the preview serves the report as
* `<target id>.zip.enc` at report.zip?target=<id> (?platform=<id> for a target named after its platform).
* `targets/<target id>.json`, selected by manifest.json?target=<id> (?platform=<id> for a target named after its platform).
*/
target?: RunTarget;
inputPath: string;
@@ -45,13 +49,13 @@ export type PreviewReportInput = {
};
export type PreviewDeployOptions = (PreviewReportInput | {
/** One report per platform from the same run, sealed into one deployment behind one token. */
/** One report per platform from the same run, deployed together behind one token. */
reports: PreviewReportInput[];
}) & {
environment?: NodeJS.ProcessEnv;
transport?: typeof fetch;
run?: CommandRunner;
/** Pause between `eas deploy` attempts; tests replace it so retries take no real time. */
/** Pause between `eas deploy` attempts and between GitHub lookups that may succeed later; tests replace it so retries take no real time. */
sleep?: (milliseconds: number) => Promise<void>;
/** The preview export (client/ and server/); defaults to the one shipped in this package. */
exportPath?: string;
@@ -81,7 +85,6 @@ export type PreviewDeployResult = {
};
export const PREVIEW_EXPORT_PATH = join(TOOL_ROOT, "dist", "hosting-preview");
const CIPHER_AAD = "expo-explore-preview/1";
const BASE32 = "abcdefghijklmnopqrstuvwxyz234567";
/** The largest report ZIP a hosted preview serves. */
export const MAX_REPORT_BYTES = 256 * 1024 * 1024;
@@ -93,7 +96,7 @@ export function platformTarget(platform: ExplorePlatform): RunTarget {
return { id: platform, platform, label: PLATFORM_LABELS[platform] };
}
/** The report.zip query that selects a target (see targetParameter): `?platform=ios` for configs without targets, as before. */
/** The manifest.json query that selects a target (see targetParameter): `?platform=ios` for configs without targets, as before. */
function targetQuery(target: RunTarget): string {
const [key, value] = targetParameter(target.id);
return `?${key}=${value}`;
@@ -141,11 +144,21 @@ const DEPLOY_ATTEMPTS = 3;
const ALIAS_LIMIT = /reached the limit of \d+ aliases[^\n]*/;
const DEPLOY_RETRY_DELAYS_MS = [5_000, 15_000] as const;
/** The tail of what a command printed, stderr first, without the EXPO_TOKEN should eas ever echo it. */
function commandOutput(result: ReturnType<CommandRunner>, token: string | undefined): string {
/**
* EAS Hosting serves static files with its own headers, so the screenshots and manifests cannot carry the Worker's
* `noindex`; this keeps crawlers from fetching them. The page and API stay crawlable, so a crawler still sees their `noindex`.
*/
export const PREVIEW_ROBOTS = "User-agent: *\nDisallow: /p/*/assets/\nDisallow: /p/*/targets/\n";
/**
* The tail of what a command printed, stderr first, without the secrets: the EXPO_TOKEN should eas ever echo it, and the
* link's token, which every static file's path carries (eas names the file whose upload failed).
*/
function commandOutput(result: ReturnType<CommandRunner>, secrets: Array<string | undefined>): string {
const lines = `${result.stderr}\n${result.stdout}`.split("\n").map(line => line.trimEnd()).filter(Boolean);
const tail = lines.slice(-30).join("\n");
return token ? tail.split(token).join("***") : tail;
let tail = lines.slice(-30).join("\n");
for (const secret of secrets) if (secret) tail = tail.split(secret).join("***");
return tail;
}
/** `eas whoami` has no JSON output; its account list is one `• name (Role: …)` line per account. */
@@ -252,13 +265,6 @@ export async function readReportZip(zip: Uint8Array): Promise<ReportZipSummary>
}
/** iv (12 bytes) || AES-256-GCM ciphertext || tag, authenticated with the preview AAD; what the Worker decrypts. */
export function sealReport(plain: Uint8Array, key: Uint8Array, iv: Uint8Array): Uint8Array {
const cipher = createCipheriv("aes-256-gcm", key, iv);
cipher.setAAD(Buffer.from(CIPHER_AAD));
const body = Buffer.concat([cipher.update(plain), cipher.final(), cipher.getAuthTag()]);
return new Uint8Array(Buffer.concat([iv, body]));
}
/** Fails fast when the Viewer export that hosted reports need is not installed, so a run does not explore for minutes before learning it cannot upload. */
export async function assertHostingExport(exportPath: string = PREVIEW_EXPORT_PATH): Promise<void> {
await assertExport(resolve(exportPath));
@@ -272,12 +278,13 @@ async function assertExport(exportPath: string): Promise<void> {
}
}
type SealedReport = { entry: PreviewReportInput; input: CiReportInput; target?: RunTarget; zip: Uint8Array; sha256: string; file: string };
/** A report to deploy; `file` names its manifest, `targets/<file>.json`: the target id, or `report` for a single unlisted one. */
type DeployedReport = { entry: PreviewReportInput; input: CiReportInput; target?: RunTarget; zip: Uint8Array; sha256: string; file: string };
function previewError(code: string, message: string, nextAction: string): CiError { return new CiError({ code, stage: "preflight", message, nextAction }); }
/**
* A device label for the platform switch, safe in the sealed .env: letters, digits and a few separators only. `eas
* A device label for the platform switch, safe in the deployment's .env: letters, digits and a few separators only. `eas
* deploy` reads that file with variable expansion, so a `$` (even single-quoted) would pull the deploying process's
* environment into the label, and a `#` or a backslash would cut or corrupt the JSON around it.
*/
@@ -291,14 +298,25 @@ function listingLabel(label: string): string {
/** One EXPLORE_PREVIEW_REPORTS entry, keys in the order the Worker echoes them in reports.json. */
export type PreviewListingEntry = { target: string; platform: ExplorePlatform; label?: string; path: string; changes?: number; device?: string };
function listingEntry(target: RunTarget, path: string, changes?: number, device?: string): PreviewListingEntry {
/** A listing entry before the deployment's token, and so its manifest's path, is known. */
type ListingTarget = Omit<PreviewListingEntry, "path">;
function listingEntry(target: RunTarget, changes?: number, device?: string): ListingTarget {
const label = listingLabel(target.label), shown = device ? listingDevice(device) : "";
return { target: target.id, platform: target.platform, ...(label ? { label } : {}), path, ...(changes === undefined ? {} : { changes }), ...(shown ? { device: shown } : {}) };
return { target: target.id, platform: target.platform, ...(label ? { label } : {}), ...(changes === undefined ? {} : { changes }), ...(shown ? { device: shown } : {}) };
}
/** The entry with its manifest's path, where the Worker reads it from (`path` after `label`, as reports.json leaves it out). */
function listedAt(entry: ListingTarget, path: string): PreviewListingEntry {
const { changes, device, ...head } = entry;
return { ...head, path, ...(changes === undefined ? {} : { changes }), ...(device === undefined ? {} : { device }) };
}
/** Where a deployment serves a target's manifest. */
export function manifestPath(token: string, file: string): string {
return `/p/${token}/targets/${file}.json`;
}
/** The worker's EXPLORE_PREVIEW_REPORTS entries, in the order the workflow lists targets; also what reports.json must echo. */
export function previewReportListing(reports: ReadonlyArray<Pick<SealedReport, "input" | "file"> & { target: RunTarget }>, reportDirectory: string): PreviewListingEntry[] {
return reports.map(report => listingEntry(report.target, `/_report/${reportDirectory}/${report.file}`, reviewChangeCount(report.input.commentState), reviewDevice(report.input.commentState)));
/** The worker's EXPLORE_PREVIEW_REPORTS entries without their paths, in the order the workflow lists targets. */
export function previewReportListing(reports: ReadonlyArray<Pick<DeployedReport, "input"> & { target: RunTarget }>): ListingTarget[] {
return reports.map(report => listingEntry(report.target, reviewChangeCount(report.input.commentState), reviewDevice(report.input.commentState)));
}
export async function deployPreview(options: PreviewDeployOptions): Promise<PreviewDeployResult> {
@@ -324,6 +342,8 @@ export async function deployPreview(options: PreviewDeployOptions): Promise<Prev
const first = inputs[0]!.input; const review = first.publicReview; const pr = review.pullRequest;
if (!first.preview) throw new CiError({ code: "PREVIEW_NOT_CONFIGURED", stage: "preflight",
message: "This review did not request a hosted preview", nextAction: "Set report.preview.hosting: eas in .expo-explore/config.yaml, or remove the deploy job." });
// `expiresAfterDays` is what a capture from before Explore stopped setting an expiry recorded.
if (first.preview.expiresAfterIgnored || "expiresAfterDays" in first.preview) warnIgnoredExpiry("report.preview.expiresAfter", environment);
if (!environment.EXPO_TOKEN?.trim()) throw new CiError({ code: "EXPO_TOKEN_MISSING", stage: "preflight",
message: "EXPO_TOKEN is required to deploy a hosted preview", nextAction: "Pass the EXPO_TOKEN secret to the deploy job; it must be able to create projects in the app's Expo account." });
if (!environment.GITHUB_TOKEN?.trim()) throw new CiError({ code: "GITHUB_TOKEN_MISSING", stage: "preflight",
@@ -339,27 +359,25 @@ export async function deployPreview(options: PreviewDeployOptions): Promise<Prev
const exportPath = resolve(options.exportPath ?? PREVIEW_EXPORT_PATH);
await assertExport(exportPath);
const reports: SealedReport[] = [];
const reports: DeployedReport[] = [];
for (const { entry, input, target } of inputs) {
let zip: Uint8Array;
try { zip = await zipReportDirectory(entry.reportPath); }
catch (error) { throw toCiError(error, { code: "REPORT_ZIP_FAILED", stage: "export",
nextAction: "Pass the ci:report output directory downloaded from this run's explore-review artifact." }); }
reports.push({ entry, input, ...(target ? { target } : {}), zip, sha256: createHash("sha256").update(zip).digest("hex"), file: multi && target ? `${target.id}.zip.enc` : "report.zip.enc" });
reports.push({ entry, input, ...(target ? { target } : {}), zip, sha256: createHash("sha256").update(zip).digest("hex"), file: multi && target ? target.id : "report" });
}
const reportDirectory = base32(random(10));
const expiresAt = first.preview.expiresAfterDays === null ? null : now() + first.preview.expiresAfterDays * 86_400_000;
const project = previewProjectSlug(pr.repository, first.preview.project);
const listing = multi ? previewReportListing(reports.map(report => ({ ...report, target: report.target! })), reportDirectory) : undefined;
let hosted: Awaited<ReturnType<typeof deploySealedReports>>;
const listing = multi ? previewReportListing(reports.map(report => ({ ...report, target: report.target! }))) : undefined;
let hosted: Awaited<ReturnType<typeof deployReports>>;
try {
hosted = await deploySealedReports({
hosted = await deployReports({
reports: reports.map(report => ({ zip: report.zip, sha256: report.sha256, file: report.file, query: multi && report.target ? targetQuery(report.target) : "" })),
reportDirectory, ...(listing ? { listing } : {}), project, ...(first.preview.owner ? { owner: first.preview.owner } : {}),
...(listing ? { listing } : {}), project, ...(first.preview.owner ? { owner: first.preview.owner } : {}),
ownerHint: "Set report.preview.owner in .expo-explore/config.yaml to the account that should hold the preview project.",
projectHint: "Give EXPO_TOKEN's account permission to create projects, or set report.preview.project to an existing project it can deploy to.",
expiresAt, exportPath, environment, run, transport, now, random, ...(options.sleep ? { sleep: options.sleep } : {}),
exportPath, environment, run, transport, now, random, ...(options.sleep ? { sleep: options.sleep } : {}),
// The runner masks secrets in job summaries too, so a masked token would reach a comparison's summary as ***.
// Its link appears nowhere else (not in the log), as a PR's appears only in its comment, for the same readers.
...(options.onSecret && pr.origin !== "compare" ? { onSecret: options.onSecret } : {}),
@@ -369,22 +387,22 @@ export async function deployPreview(options: PreviewDeployOptions): Promise<Prev
if (pr.origin !== "compare") for (const report of reports) {
try {
await publishCiReportComment({ inputPath: report.entry.inputPath, artifactUrl: report.entry.artifactUrl, environment, transport,
...commentTarget(report), previewFailed: true });
...commentTarget(report), previewFailed: true, ...(options.sleep ? { sleep: options.sleep } : {}) });
} catch { /* the deployment failure is the one to report */ }
}
throw error;
}
const previewUrl = hosted.previewUrl;
// A comparison has no pull request: its link goes to the run's summary, which the same repository readers see.
if (pr.origin === "compare") await writeCompareSummary(environment, pr, previewUrl, expiresAt, reports.map(report => report.input));
if (pr.origin === "compare") await writeCompareSummary(environment, pr, previewUrl, reports.map(report => report.input));
{
let headSha = pr.head.sha;
if (pr.origin !== "compare") for (const report of reports) {
const published = await publishCiReportComment({ inputPath: report.entry.inputPath, artifactUrl: report.entry.artifactUrl, environment, transport,
...commentTarget(report), preview: { url: previewUrl, expiresAt } });
...commentTarget(report), preview: { url: previewUrl, expiresAt: null }, ...(options.sleep ? { sleep: options.sleep } : {}) });
headSha = published.headSha;
}
return { previewUrl, expiresAt: expiresAt === null ? null : new Date(expiresAt).toISOString(), deploymentId: hosted.deploymentId, deploymentUrl: hosted.deploymentUrl,
return { previewUrl, expiresAt: null, deploymentId: hosted.deploymentId, deploymentUrl: hosted.deploymentUrl,
project, owner: hosted.owner, reportSha256: reports[0]!.sha256, reportBytes: reports[0]!.zip.byteLength,
reports: reports.map(report => ({ ...(report.entry.platform ? { platform: report.entry.platform } : {}), ...(report.target ? { target: report.target } : {}), artifactUrl: report.entry.artifactUrl,
reportSha256: report.sha256, reportBytes: report.zip.byteLength })), headSha,
@@ -393,13 +411,13 @@ export async function deployPreview(options: PreviewDeployOptions): Promise<Prev
}
/** Which comment section and artifact a report is: its target, or (single-report runs) none. */
function commentTarget(report: SealedReport): { target?: string; platform?: ExplorePlatform } {
function commentTarget(report: DeployedReport): { target?: string; platform?: ExplorePlatform } {
if (report.target) return { target: report.target.id, platform: report.target.platform };
return report.entry.platform ? { platform: report.entry.platform } : {};
}
/** The comparison's job summary: what was compared, the hosted link, and each target's change count. */
export function renderCompareSummary(comparison: CompareReviewPresentation, previewUrl: string, expiresAt: number | null, inputs: CiReportInput[]): string {
export function renderCompareSummary(comparison: CompareReviewPresentation, previewUrl: string, inputs: CiReportInput[]): string {
const platforms = inputs.map(input => {
const changes = reviewChangeCount(input.commentState);
const device = reviewDevice(input.commentState);
@@ -407,19 +425,20 @@ export function renderCompareSummary(comparison: CompareReviewPresentation, prev
return `- ${label}${device ? ` (${device})` : ""}: ${changes === undefined ? "no review" : `${changes} ${changes === 1 ? "change" : "changes"}`}`;
});
return [`### Explore: ${compareTitleMarkdown(comparison)}`, "",
`[Open the report](${previewUrl}) · [${comparison.base.sha.slice(0, 8)}..${comparison.head.sha.slice(0, 8)}](${comparison.url})`
+ (expiresAt === null ? "" : ` · link expires ${new Date(expiresAt).toISOString().slice(0, 10)}`), "",
`[Open the report](${previewUrl}) · [${comparison.base.sha.slice(0, 8)}..${comparison.head.sha.slice(0, 8)}](${comparison.url})`, "",
...platforms, ""].join("\n");
}
async function writeCompareSummary(environment: NodeJS.ProcessEnv, comparison: CompareReviewPresentation, previewUrl: string, expiresAt: number | null, inputs: CiReportInput[]) {
async function writeCompareSummary(environment: NodeJS.ProcessEnv, comparison: CompareReviewPresentation, previewUrl: string, inputs: CiReportInput[]) {
const path = environment.GITHUB_STEP_SUMMARY;
if (!path) return;
await appendFile(path, renderCompareSummary(comparison, previewUrl, expiresAt, inputs));
await appendFile(path, renderCompareSummary(comparison, previewUrl, inputs));
}
type HostedReportInput = { zip: Uint8Array; sha256: string; file: string; query: string };
export type HostedDeployment = { previewUrl: string; expiresAt: string | null; deploymentId: string; deploymentUrl: string; aliasUrl?: string; project: string; owner: string };
export type HostedDeployment = { previewUrl: string; expiresAt: string | null; deploymentId: string; deploymentUrl: string; aliasUrl?: string;
/** The link is the alias, which still served an earlier deployment when the wait for it ran out. */
aliasBehind?: true; project: string; owner: string };
/** EAS Hosting alias names: lowercase slugs, as `eas deploy --alias` normalises them. */
export const HOSTING_ALIAS = /^[a-z0-9][a-z0-9-]{0,38}[a-z0-9]$/;
@@ -427,24 +446,73 @@ export const HOSTING_ALIAS = /^[a-z0-9][a-z0-9-]{0,38}[a-z0-9]$/;
export const HOSTED_TOKEN = /^[a-z2-7]{32}$/;
/**
* Seal the report ZIPs, deploy the exported Viewer around them to EAS Hosting, and verify the live bytes.
* Shared by the PR preview (`ci:deploy`) and standalone exploration reports; it knows nothing about GitHub.
* Each report's manifest at `p/<token>/targets/<file>.json` and every asset once at `p/<token>/assets/<sha256>.<ext>`
* (targets that share a screenshot share its file), under the export's client directory. Every asset's bytes must hash
* to its name. `deployment` stamps every manifest, so a check can tell this deployment from one an alias still serves.
* Returns, per report, the manifest written and one asset to fetch back.
*/
async function deploySealedReports(options: {
reports: HostedReportInput[]; reportDirectory: string; listing?: ReturnType<typeof previewReportListing>; project: string; owner?: string;
ownerHint: string; projectHint: string; expiresAt: number | null; exportPath: string; environment: NodeJS.ProcessEnv; run: typeof defaultRun;
async function writeReportFiles(client: string, token: string, reports: HostedReportInput[], deployment: string): Promise<Array<{ manifest: string; asset?: string }>> {
await writeFile(join(client, "robots.txt"), PREVIEW_ROBOTS);
const root = join(client, "p", token);
await mkdir(join(root, "targets"), { recursive: true });
await mkdir(join(root, "assets"), { recursive: true });
// Each asset's size as written: a target that shares the file must list the same size in its manifest.
const assets = new Map<string, number>();
const written: Array<{ manifest: string; asset?: string }> = [];
for (const report of reports) {
const loaded = await loadReport(report.zip);
if (!loaded) throw deployError("REPORT_ZIP_FAILED", "The report ZIP is not a report the hosted Viewer can open.", "Pass the report directory or ZIP that Explore wrote.");
const manifest = manifestBody(loaded, deployment);
await writeFile(join(root, "targets", `${report.file}.json`), manifest);
let first: string | undefined;
for (const [name, entry] of loaded.entries) {
const sha = PREVIEW_ASSET.exec(name)?.[1];
if (!sha) continue;
first ??= name;
const bad = () => deployError("REPORT_ZIP_FAILED", `The report's ${name} is unreadable, larger than ${MAX_ASSET_BYTES / 1024 / 1024} MiB, or does not match its name.`,
"Capture the report again; its screenshots are named by their SHA-256.");
if (assets.has(name)) {
if (assets.get(name) !== entry.size) throw bad();
continue;
}
const bytes = await loaded.read(name, MAX_ASSET_BYTES);
if (!bytes || bytes.byteLength !== entry.size || createHash("sha256").update(bytes).digest("hex") !== sha) throw bad();
await writeFile(join(root, name), bytes);
assets.set(name, bytes.byteLength);
}
written.push({ manifest, ...(first ? { asset: first } : {}) });
}
return written;
}
/**
* Write the report ZIPs as static files under the token, deploy the exported Viewer around them to EAS Hosting, and
* verify what the Viewer opens. Shared by the PR preview (`ci:deploy`) and standalone reports; it knows nothing about GitHub.
*/
async function deployReports(options: {
/** With `listing`, one entry per report in the same order. */
reports: HostedReportInput[]; listing?: ListingTarget[]; project: string; owner?: string;
ownerHint: string; projectHint: string; exportPath: string; environment: NodeJS.ProcessEnv; run: typeof defaultRun;
transport: typeof fetch; now: () => number; random: (bytes: number) => Uint8Array; onSecret?: (secret: string) => void; verifyTimeoutMs: number;
/** `eas deploy --alias`: a fixed host that always serves the newest deployment, for links that must not change between runs. */
alias?: string;
/** A caller-chosen `/p/<token>`; with `alias`, the hosted link is identical from one run to the next. */
token?: string;
/** How long the alias has to serve this deployment once the deployment does; `verifyTimeoutMs` by default. */
aliasWaitMs?: number;
/**
* Link the alias even when it still serves an earlier deployment after `aliasWaitMs`, for a link whose readers are
* told it can lag, rather than the deployment's own host.
*/
linkBehindAlias?: boolean;
sleep?: (milliseconds: number) => Promise<void>;
}): Promise<HostedDeployment> {
const { reports, reportDirectory, listing, project, environment, run, transport, now, random, expiresAt } = options;
const token = options.token ?? base32(random(20)); const key = random(32);
const { reports, project, environment, run, transport, now, random } = options;
const token = options.token ?? base32(random(20));
options.onSecret?.(token);
// `eas deploy` reads the sealed .env through @expo/env, where any variable already in its own environment wins,
// even an empty one: an inherited EXPLORE_PREVIEW_* would replace the sealed value and every route would 404.
const listing = options.listing?.map((entry, index) => listedAt(entry, manifestPath(token, reports[index]!.file)));
// `eas deploy` reads the deployment's .env through @expo/env, where any variable already in its own environment wins,
// even an empty one: an inherited EXPLORE_PREVIEW_* would replace the deployment's value and every route would 404.
const easEnvironment = { ...Object.fromEntries(Object.entries(environment).filter(([name]) => !name.startsWith("EXPLORE_PREVIEW_"))),
CI: "1", EXPO_NO_TELEMETRY: "1" };
// Run the pinned CLI through this Node binary: the deploy job has no node on PATH for spawned shebang scripts to rely on.
@@ -454,14 +522,10 @@ async function deploySealedReports(options: {
const work = await mkdtemp(join(tmpdir(), "explore-preview-"));
try {
await cp(options.exportPath, join(work, "dist"), { recursive: true });
const sealedDirectory = join(work, "dist", "client", "_report", reportDirectory);
await mkdir(sealedDirectory, { recursive: true });
for (const report of reports) await writeFile(join(sealedDirectory, report.file), sealReport(report.zip, key, random(12)));
const written = await writeReportFiles(join(work, "dist", "client"), token, reports, base32(random(10)));
await writeFile(join(work, ".env"), [
`EXPLORE_PREVIEW_TOKEN=${token}`,
`EXPLORE_PREVIEW_KEY=${Buffer.from(key).toString("base64")}`,
...(expiresAt === null ? [] : [`EXPLORE_PREVIEW_EXPIRES_AT=${new Date(expiresAt).toISOString()}`]),
listing ? `EXPLORE_PREVIEW_REPORTS=${JSON.stringify(listing)}` : `EXPLORE_PREVIEW_REPORT=/_report/${reportDirectory}/${reports[0]!.file}`,
listing ? `EXPLORE_PREVIEW_REPORTS=${JSON.stringify(listing)}` : `EXPLORE_PREVIEW_REPORT=${manifestPath(token, reports[0]!.file)}`,
"",
].join("\n"), { mode: 0o600 });
@@ -502,23 +566,24 @@ async function deploySealedReports(options: {
if (aliasLimit) {
throw deployError("PREVIEW_ALIAS_LIMIT", `EAS Hosting could not assign the alias ${options.alias}: ${aliasLimit[0]}`,
`Delete the aliases of closed pull requests on @${owner}/${project} (eas deploy:alias:delete pr-<n> --non-interactive) or raise the account's plan, then rerun the deploy job.`,
{ retryable: false, details: { project, owner, alias: options.alias, output: commandOutput(deploy, environment.EXPO_TOKEN) } });
{ retryable: false, details: { project, owner, alias: options.alias, output: commandOutput(deploy, [environment.EXPO_TOKEN, token]) } });
}
process.stderr.write(`eas deploy attempt ${attempt} of ${DEPLOY_ATTEMPTS} failed (exit ${deploy.status}):\n${commandOutput(deploy, environment.EXPO_TOKEN)}\n`);
process.stderr.write(`eas deploy attempt ${attempt} of ${DEPLOY_ATTEMPTS} failed (exit ${deploy.status}):\n${commandOutput(deploy, [environment.EXPO_TOKEN, token])}\n`);
if (attempt < DEPLOY_ATTEMPTS) await sleep(DEPLOY_RETRY_DELAYS_MS[attempt - 1] ?? DEPLOY_RETRY_DELAYS_MS.at(-1)!);
}
if (!deploymentUrl) {
throw deployError("PREVIEW_DEPLOY_FAILED", `EAS Hosting deployment failed ${DEPLOY_ATTEMPTS} times: ${lastLine(deploy.stderr || deploy.stdout)}`,
"Check the EAS Hosting status for the preview project and rerun the deploy job; the eas output of every attempt is in the job log.",
{ retryable: true, details: { project, owner, attempts: DEPLOY_ATTEMPTS, output: commandOutput(deploy, environment.EXPO_TOKEN) } });
{ retryable: true, details: { project, owner, attempts: DEPLOY_ATTEMPTS, output: commandOutput(deploy, [environment.EXPO_TOKEN, token]) } });
}
const aliasUrl = options.alias ? hostedOrigin(deployment.aliases?.find(candidate => candidate.name === options.alias)?.url) : undefined;
if (options.alias && !aliasUrl) {
throw deployError("PREVIEW_DEPLOY_FAILED", `EAS Hosting deployed ${deploymentUrl.origin} but did not report the alias ${options.alias}.`,
"Check the deployment's aliases in the EAS Hosting dashboard and rerun.", { retryable: true, details: { project, owner, alias: options.alias } });
}
const verify = (origin: string, alias = false) => verifyDeployment({ origin, token, transport, now, timeoutMs: options.verifyTimeoutMs, alias,
reports: reports.map(report => ({ query: report.query, sha256: report.sha256, bytes: report.zip.byteLength })),
const verify = (origin: string, alias = false) => verifyDeployment({ origin, token, transport, now, alias,
timeoutMs: alias ? options.aliasWaitMs ?? options.verifyTimeoutMs : options.verifyTimeoutMs,
reports: reports.map((report, index) => ({ query: report.query, ...written[index]! })),
...(listing ? { listing: JSON.stringify(listing.map(({ path: _path, ...rest }) => rest)) } : {}) });
const seconds = (milliseconds: number) => `${Math.round(milliseconds / 1000)} s after eas deploy returned`;
const served = await verify(deploymentUrl.origin);
@@ -527,7 +592,9 @@ async function deploySealedReports(options: {
// taken longer than the whole budget (every verification failure of 2026-10-14 was the alias, with the deployment
// serving in 1 s). The deployment is verified and immutable, so when the alias is still behind the pull request
// links the deployment instead of failing the job; the alias catches up on its own and the next push relinks it.
// With `linkBehindAlias` it links the alias anyway, and its readers are told the link can lag.
let linkOrigin = deploymentUrl.origin;
let aliasBehind = false;
if (aliasUrl) {
try {
const alias = await verify(aliasUrl.origin, true);
@@ -535,12 +602,16 @@ async function deploySealedReports(options: {
linkOrigin = aliasUrl.origin;
} catch (error) {
if ((error as { code?: string }).code !== "PREVIEW_ALIAS_BEHIND") throw error;
process.stderr.write(`::warning::${(error as Error).message} The pull request links the deployment ${deploymentUrl.origin} instead.\n`);
aliasBehind = Boolean(options.linkBehindAlias);
if (aliasBehind) linkOrigin = aliasUrl.origin;
process.stderr.write(`::warning::${(error as Error).message} ${aliasBehind
? `The pull request links it anyway, until it catches up, beside the deployment ${deploymentUrl.origin}.`
: `The pull request links the deployment ${deploymentUrl.origin} instead.`}\n`);
}
}
const previewUrl = `${linkOrigin}/p/${token}`;
return { previewUrl, expiresAt: expiresAt === null ? null : new Date(expiresAt).toISOString(), deploymentId: deployment.identifier!, deploymentUrl: deploymentUrl.origin,
...(aliasUrl ? { aliasUrl: aliasUrl.origin } : {}), project, owner };
return { previewUrl, expiresAt: null, deploymentId: deployment.identifier!, deploymentUrl: deploymentUrl.origin,
...(aliasUrl ? { aliasUrl: aliasUrl.origin } : {}), ...(aliasBehind ? { aliasBehind: true as const } : {}), project, owner };
} finally {
await rm(work, { recursive: true, force: true });
}
@@ -554,7 +625,6 @@ export type ExplorationDeployOptions = {
project: string;
owner?: string;
/** Days until the hosted link stops serving the report; null never expires. */
expiresAfterDays: number | null;
exportPath?: string;
environment?: NodeJS.ProcessEnv;
run?: typeof defaultRun;
@@ -577,7 +647,7 @@ export async function deployExplorationReport(options: ExplorationDeployOptions)
message: "An EAS login is required to host the report", nextAction: "Run `eas login` or set EXPO_TOKEN, then rerun with --upload." });
if (!/^[a-z0-9][a-z0-9-]{0,38}[a-z0-9]$/.test(options.project)) throw previewError("PREVIEW_PROJECT_INVALID", `Invalid EAS project slug: ${options.project}`, "Pass --project as a lowercase slug such as expo-explore.");
if (options.alias !== undefined && !HOSTING_ALIAS.test(options.alias)) throw previewError("PREVIEW_ALIAS_INVALID", `Invalid EAS Hosting alias: ${options.alias}`, "Pass --alias as a lowercase slug such as bluesky.");
if (options.token !== undefined && !HOSTED_TOKEN.test(options.token)) throw previewError("PREVIEW_TOKEN_INVALID", "EXPLORE_PREVIEW_TOKEN must be 32 characters of a-z and 2-7.", "Generate one with `openssl rand 20 | base32 | tr A-Z a-z` and set EXPLORE_PREVIEW_TOKEN, or unset it for a random token.");
if (options.token !== undefined && !HOSTED_TOKEN.test(options.token)) throw previewError("PREVIEW_TOKEN_INVALID", "EXPLORE_PREVIEW_TOKEN must be 32 characters of a-z and 2-7.", "Generate one with `LC_ALL=C tr -dc a-z2-7 </dev/urandom | head -c 32` and set EXPLORE_PREVIEW_TOKEN, or unset it for a random token.");
const exportPath = resolve(options.exportPath ?? PREVIEW_EXPORT_PATH);
await assertExport(exportPath);
let zip: Uint8Array;
@@ -586,21 +656,20 @@ export async function deployExplorationReport(options: ExplorationDeployOptions)
const random = options.random ?? (bytes => new Uint8Array(randomBytes(bytes)));
const now = options.now ?? Date.now;
const sha256 = createHash("sha256").update(zip).digest("hex");
const reportDirectory = base32(random(10));
const target = options.platform ? platformTarget(options.platform) : undefined;
const file = target ? `${target.id}.zip.enc` : "report.zip.enc";
const hosted = await deploySealedReports({ reports: [{ zip, sha256, file, query: target ? targetQuery(target) : "" }], reportDirectory,
...(target ? { listing: [listingEntry(target, `/_report/${reportDirectory}/${file}`)] } : {}), project: options.project,
const file = target ? target.id : "report";
const hosted = await deployReports({ reports: [{ zip, sha256, file, query: target ? targetQuery(target) : "" }],
...(target ? { listing: [listingEntry(target)] } : {}), project: options.project,
...(options.owner ? { owner: options.owner } : {}), ownerHint: "Pass --owner with the Expo account that should host the report.",
projectHint: "Give this EAS login permission to create projects under the owner, or pass --project with an existing slug.",
expiresAt: options.expiresAfterDays === null ? null : now() + options.expiresAfterDays * 86_400_000, exportPath, environment, run: options.run ?? defaultRun, ...(options.sleep ? { sleep: options.sleep } : {}),
exportPath, environment, run: options.run ?? defaultRun, ...(options.sleep ? { sleep: options.sleep } : {}),
transport: options.transport ?? fetch, now, random, ...(options.onSecret ? { onSecret: options.onSecret } : {}), verifyTimeoutMs: options.verifyTimeoutMs ?? 180_000,
...(options.alias ? { alias: options.alias } : {}), ...(options.token ? { token: options.token } : {}) });
return { ...hosted, reportSha256: sha256, reportBytes: zip.byteLength };
}
export type FixturePreviewOptions = {
/** Report ZIPs as a hosted preview serves them from report.zip, one per target (`platform` alone: the platform's target), with what the switch says of each. */
/** Report ZIPs as Explore writes them, one per target (`platform` alone: the platform's target), with what the switch says of each. */
reports: Array<{ platform: ExplorePlatform; target?: RunTarget; zipPath: string; changes?: number; device?: string }>;
project: string;
owner?: string;
@@ -616,8 +685,20 @@ export type FixturePreviewOptions = {
random?: (bytes: number) => Uint8Array;
onSecret?: (secret: string) => void;
verifyTimeoutMs?: number;
/** How long the alias has to serve the new deployment before the link is the alias anyway; FIXTURE_ALIAS_WAIT_MS by default. */
aliasWaitMs?: number;
};
/**
* How long a fixture preview waits for its alias. An edge of EAS Hosting that served an alias keeps serving the
* deployment it named for about five minutes (307–314 s, measured 2026-10-19 redeploying back to back from one machine),
* so a deploy whose check reaches an edge that served the alias in the last five minutes, to the previous deploy's check
* or to a reviewer, finds it behind; otherwise it has moved within seconds. Waiting the whole verification budget for
* that held a third of the pull requests' deploy jobs for three minutes, only to link the deployment instead. The pull
* request's comment says when the alias is behind.
*/
export const FIXTURE_ALIAS_WAIT_MS = 20_000;
/**
* Host fixed reports behind a fixed alias with this checkout's Viewer: the Explore repository's own pull requests
* preview their Viewer changes on real data. No GitHub token, no expiry; each run replaces what the alias serves.
@@ -632,23 +713,45 @@ export async function deployFixturePreview(options: FixturePreviewOptions): Prom
const exportPath = resolve(options.exportPath ?? PREVIEW_EXPORT_PATH);
await assertExport(exportPath);
const random = options.random ?? (bytes => new Uint8Array(randomBytes(bytes)));
const reportDirectory = base32(random(10));
const reports = await Promise.all(options.reports.map(async report => {
const zip = new Uint8Array(await readFile(report.zipPath));
const target = report.target ?? platformTarget(report.platform);
return { ...report, target, zip, sha256: createHash("sha256").update(zip).digest("hex"), file: `${target.id}.zip.enc` };
return { ...report, target, zip, sha256: createHash("sha256").update(zip).digest("hex"), file: target.id };
}));
const listing = reports.map(report => listingEntry(report.target, `/_report/${reportDirectory}/${report.file}`, report.changes, report.device));
return deploySealedReports({
const listing = reports.map(report => listingEntry(report.target, report.changes, report.device));
return deployReports({
reports: reports.map(report => ({ zip: report.zip, sha256: report.sha256, file: report.file, query: targetQuery(report.target) })),
reportDirectory, listing, project: options.project, ...(options.owner ? { owner: options.owner } : {}),
listing, project: options.project, ...(options.owner ? { owner: options.owner } : {}),
ownerHint: "Pass --owner with the Expo account that holds the preview project.",
projectHint: "Give EXPO_TOKEN's account permission to create projects, or pass --project with an existing slug.",
expiresAt: null, exportPath, environment, run: options.run ?? defaultRun, ...(options.sleep ? { sleep: options.sleep } : {}), transport: options.transport ?? fetch, now: options.now ?? Date.now, random,
exportPath, environment, run: options.run ?? defaultRun, ...(options.sleep ? { sleep: options.sleep } : {}), transport: options.transport ?? fetch, now: options.now ?? Date.now, random,
...(options.onSecret ? { onSecret: options.onSecret } : {}), verifyTimeoutMs: options.verifyTimeoutMs ?? 180_000, alias: options.alias,
aliasWaitMs: options.aliasWaitMs ?? FIXTURE_ALIAS_WAIT_MS, linkBehindAlias: true,
...(options.token ? { token: options.token } : {}) });
}
/**
* The reports a fixture directory holds (scripts/refresh-viewer-preview-fixtures.sh writes it): its reports.json, as a
* hosted preview's /reports.json lists them, with `<target>.zip` beside it for each entry. An entry without a target,
* from fixtures refreshed before targets, is its platform's target with `<platform>.zip`.
*/
export function fixtureReports(listing: unknown, directory: string): FixturePreviewOptions["reports"] {
const invalid = () => previewError("PREVIEW_REPORTS_INVALID", "The fixtures' reports.json does not list valid reports",
"Refresh the fixtures with scripts/refresh-viewer-preview-fixtures.sh.");
if (!Array.isArray(listing)) throw invalid();
return listing.map((entry: unknown) => {
if (typeof entry !== "object" || entry === null) throw invalid();
const { platform, target: id, label, changes, device } = entry as Record<string, unknown>;
if (platform !== "ios" && platform !== "android") throw invalid();
const target = id === undefined ? undefined : readTarget({ id, platform, label });
if (id !== undefined && !target) throw invalid();
return {
platform, ...(target ? { target } : {}), zipPath: join(directory, `${target?.id ?? platform}.zip`),
...(typeof changes === "number" ? { changes } : {}), ...(typeof device === "string" ? { device } : {}),
};
});
}
/**
* Report ZIPs on this machine behind one token (`capture:review --upload`, `report:upload`), one per target: the same
* sealed deployment as a PR preview, from ZIPs instead of GitHub artifacts, with the local EAS login or EXPO_TOKEN.
@@ -656,7 +759,7 @@ export async function deployFixturePreview(options: FixturePreviewOptions): Prom
export async function deployLocalReports(options: {
/** One per target; an entry without `target` is its platform's target, as before targets. */
reports: Array<{ platform: ExplorePlatform; target?: RunTarget; zip: Uint8Array; changes?: number; device?: string }>;
project: string; owner?: string; expiresAfterDays: number | null;
project: string; owner?: string;
/** What to change when the owner or project is not usable; defaults name report.preview in config.yaml. */
ownerHint?: string; projectHint?: string;
exportPath?: string; environment?: NodeJS.ProcessEnv; run?: CommandRunner; sleep?: (milliseconds: number) => Promise<void>;
@@ -668,19 +771,18 @@ export async function deployLocalReports(options: {
await assertExport(exportPath);
const now = options.now ?? Date.now;
const random = options.random ?? (bytes => new Uint8Array(randomBytes(bytes)));
const reportDirectory = base32(random(10));
// Always listed, even alone: the Worker takes an unlisted report for iOS, so an Android one would be mislabelled.
const reports = options.reports.map(report => {
const target = report.target ?? platformTarget(report.platform);
return { ...report, target, sha256: createHash("sha256").update(report.zip).digest("hex"), file: `${target.id}.zip.enc` };
return { ...report, target, sha256: createHash("sha256").update(report.zip).digest("hex"), file: target.id };
});
const listing = reports.map(report => listingEntry(report.target, `/_report/${reportDirectory}/${report.file}`, report.changes, report.device));
return deploySealedReports({
const listing = reports.map(report => listingEntry(report.target, report.changes, report.device));
return deployReports({
reports: reports.map(report => ({ zip: report.zip, sha256: report.sha256, file: report.file, query: targetQuery(report.target) })),
reportDirectory, listing, project: options.project, ...(options.owner ? { owner: options.owner } : {}),
listing, project: options.project, ...(options.owner ? { owner: options.owner } : {}),
ownerHint: options.ownerHint ?? "Set report.preview.owner in .expo-explore/config.yaml to the account that should hold the hosted reports.",
projectHint: options.projectHint ?? "Make sure your Expo account can create projects, or set report.preview.project to an existing project you can deploy to.",
expiresAt: options.expiresAfterDays === null ? null : now() + options.expiresAfterDays * 86_400_000, exportPath, environment,
exportPath, environment,
run: options.run ?? defaultRun, ...(options.sleep ? { sleep: options.sleep } : {}), transport: options.transport ?? fetch, now, random,
verifyTimeoutMs: options.verifyTimeoutMs ?? 180_000 });
}
@@ -688,23 +790,43 @@ export async function deployLocalReports(options: {
/** explore-(capture|review)-<run>-<attempt>-<target id>; the id starts with a letter, so the run and attempt end unambiguously. */
const ARTIFACT_DIRECTORY = /^explore-(capture|review)-\d+-\d+-([a-z][a-z0-9-]{0,31})$/;
/**
* The target a lone artifact holds. actions/download-artifact puts each artifact a pattern matches in a directory named
* after it, except when the pattern matches exactly one, which it extracts into the download path itself: then only the
* report names its target (its platform for configs without targets).
*/
async function unnamedArtifactTarget(root: string, kind: "capture" | "review"): Promise<string | undefined> {
const file = join(root, kind === "capture" ? "report-input.json" : "review.json");
if (!(await lstat(file).catch(() => undefined))?.isFile()) return undefined;
const runtime = kind === "capture" ? (await readCiReportInput(file)).publicReview.runtime
: (JSON.parse(await readFile(file, "utf8")) as Partial<ArtifactReport>).review?.runtime;
const id = readTarget(runtime?.target)?.id ?? runtime?.platform;
return isTargetId(id) ? id : undefined;
}
/**
* The fan-in deploy job's inputs: actions/download-artifact puts each explore-capture-<run>-<attempt>-<target>
* directory under --inputs and each explore-review-… directory under --reports (the target is the platform for
* configs without targets). Pairs them by target, in the order the run's review lists its targets, and looks up each
* report artifact's URL on the run, since matrix jobs cannot hand their artifact URLs forward.
* report artifact's URL on the run, since matrix jobs cannot hand their artifact URLs forward. A target whose capture
* failed uploaded no capture artifact, and one whose report job failed no report artifact: the preview serves the
* targets that have both, and the review workflow's failure job names the others on the comment.
*/
export async function collectPlatformReports(options: { inputsPath: string; reportsPath: string; environment?: NodeJS.ProcessEnv; transport?: typeof fetch }): Promise<PreviewReportInput[]> {
export async function collectPlatformReports(options: { inputsPath: string; reportsPath: string; environment?: NodeJS.ProcessEnv; transport?: typeof fetch;
/** Pause between artifact lookups that may succeed later; tests replace it so retries take no real time. */
sleep?: (milliseconds: number) => Promise<void> }): Promise<PreviewReportInput[]> {
const environment = options.environment ?? process.env; const transport = options.transport ?? fetch;
const directories = async (root: string, kind: "capture" | "review") => {
const found = new Map<string, string>();
for (const name of (await readdir(root)).sort()) {
// A pattern that matched nothing leaves no directory behind.
for (const name of (await readdir(root).catch(() => [] as string[])).sort()) {
const match = ARTIFACT_DIRECTORY.exec(name);
if (!match || match[1] !== kind || !isTargetId(match[2])) continue;
if (found.has(match[2]!)) throw previewError("PREVIEW_REPORTS_INVALID", `Several ${kind} artifacts for ${match[2]} were downloaded`,
"Download only this run attempt's artifacts into the --inputs and --reports directories.");
found.set(match[2]!, join(root, name));
}
if (!found.size) { const id = await unnamedArtifactTarget(root, kind); if (id) found.set(id, root); }
return found;
};
const inputs = await directories(options.inputsPath, "capture"); const reports = await directories(options.reportsPath, "review");
@@ -715,8 +837,9 @@ export async function collectPlatformReports(options: { inputsPath: string; repo
const entries: PreviewReportInput[] = [];
for (const [id, inputDirectory] of inputs) {
const reportPath = reports.get(id);
if (!reportPath) throw previewError("PREVIEW_REPORTS_INCOMPLETE", `The ${id} capture has no report artifact to deploy`,
"Rerun the workflow; the report job for that target did not upload its explore-review artifact.");
// In brentvatne/euxy#172 (review run 39348956443) one target's failed job kept every other target's report off the
// PR; a target without a report is left out of the preview instead.
if (!reportPath) { process.stderr.write(`The ${id} capture has no report artifact (its report job failed), so the preview leaves it out.\n`); continue; }
const inputPath = join(inputDirectory, "report-input.json");
const runtime = (await readCiReportInput(inputPath)).publicReview.runtime;
// A report from before targets names only its platform, and its artifact is named after it.
@@ -726,6 +849,8 @@ export async function collectPlatformReports(options: { inputsPath: string; repo
const target = named ?? (id === "ios" || id === "android" ? platformTarget(id) : { id, platform: runtime.platform ?? "ios", label: id });
entries.push({ ...(id === target.platform ? { platform: target.platform } : {}), target, inputPath, reportPath, artifactUrl: "" });
}
if (!entries.length) throw previewError("PREVIEW_REPORTS_INCOMPLETE", "No target's capture has a report artifact to deploy",
"Rerun the workflow; the report jobs did not upload their explore-review artifacts.");
// Config order, as the review lists its targets (or, before targets, its platforms); directory order otherwise.
const listed = (await readCiReportInput(entries[0]!.inputPath)).publicReview.runtime;
const order: string[] = listed.targets?.map(target => target.id) ?? listed.platforms ?? [];
@@ -740,21 +865,29 @@ export async function collectPlatformReports(options: { inputsPath: string; repo
for (const entry of entries) {
const name = `explore-review-${review.runId}-${review.runAttempt}-${entry.target!.id}`;
const url = `https://api.github.com/repos/${repository}/actions/runs/${review.runId}/artifacts?per_page=100&name=${encodeURIComponent(name)}`;
const response = await transport(url, { headers: { accept: "application/vnd.github+json", authorization: `Bearer ${environment.GITHUB_TOKEN ?? ""}`,
"x-github-api-version": "2022-11-28" }, signal: AbortSignal.timeout(60_000) });
const body = response.ok ? await response.json() as { artifacts?: Array<{ id?: unknown; name?: unknown; expired?: unknown }> } : undefined;
const artifact = body?.artifacts?.find(candidate => candidate.name === name && candidate.expired !== true);
if (!artifact || typeof artifact.id !== "number" || !Number.isSafeInteger(artifact.id) || artifact.id <= 0) {
throw deployError("REPORT_ARTIFACT_MISSING", `The run has no artifact named ${name} (HTTP ${response.status}).`,
"Rerun the workflow so every target's report job uploads its artifact before the deploy job runs.", { retryable: true });
}
// The run's artifact list can lag an upload, as the artifact lookup of a report job's publication does, so an
// empty list, a 404, a 5xx or no answer is looked up again; a bad token or a missing permission fails at once.
const artifact = await retryGitHubLookup(async () => {
const response = await transport(url, { headers: { accept: "application/vnd.github+json", authorization: `Bearer ${environment.GITHUB_TOKEN ?? ""}`,
"x-github-api-version": "2022-11-28" }, signal: AbortSignal.timeout(60_000) });
const body = response.ok ? await response.json() as { artifacts?: Array<{ id?: unknown; name?: unknown; expired?: unknown }> } : undefined;
const found = body?.artifacts?.find(candidate => candidate.name === name && candidate.expired !== true);
if (!found || typeof found.id !== "number" || !Number.isSafeInteger(found.id) || found.id <= 0) {
throw deployError("REPORT_ARTIFACT_MISSING", `The run has no artifact named ${name} (HTTP ${response.status}).`,
"Rerun the workflow so every target's report job uploads its artifact before the deploy job runs.", { retryable: true, details: { status: response.status } });
}
return found;
}, error => {
const status = error instanceof CiError ? error.details?.status : undefined;
return !(error instanceof CiError) || typeof status === "number" && (status === 200 || status === 404 || status >= 500);
}, options.sleep);
entry.artifactUrl = `https://github.com/${repository}/actions/runs/${review.runId}/artifacts/${artifact.id}`;
}
return entries;
}
/** The deployment must serve the Viewer and the exact reports at the token, and nothing anywhere else. */
async function verifyDeployment(check: { origin: string; token: string; reports: Array<{ query: string; sha256: string; bytes: number }>; listing?: string;
async function verifyDeployment(check: { origin: string; token: string; reports: Array<{ query: string; manifest: string; asset?: string }>; listing?: string;
transport: typeof fetch; now: () => number; timeoutMs: number; alias?: boolean }): Promise<{ liveAfterMs: number }> {
const get = async (path: string, accept = "*/*") => {
try {
@@ -765,16 +898,46 @@ async function verifyDeployment(check: { origin: string; token: string; reports:
"Check EAS Hosting availability and rerun the deploy job.", { retryable: true, cause: error });
}
};
// What the token serves differs from what was deployed: a listing or report that does not match, or undefined.
const mismatch = async (): Promise<CiError | undefined> => {
// What the Viewer opens: each target's manifest, exactly as written (its deployment stamp tells this deployment from
// the one an alias still points at, even for an unchanged report), and one of its screenshots, a static file.
for (const expected of check.reports) {
const manifest = await get(`/p/${check.token}/manifest.json${expected.query}`);
const body = manifest.status === 200 ? await manifest.text() : "";
const served = expected.asset ? await get(`/p/${check.token}/${expected.asset}`) : undefined;
const assetBytes = served?.status === 200 ? new Uint8Array(await served.arrayBuffer()) : new Uint8Array();
const sha = expected.asset ? PREVIEW_ASSET.exec(expected.asset)?.[1] : undefined;
if (body !== expected.manifest || (sha && createHash("sha256").update(assetBytes).digest("hex") !== sha)) {
return deployError("PREVIEW_VERIFY_FAILED", "The hosted preview did not serve the report's manifest and screenshots, which the Viewer opens.",
"Rerun the deploy job; if it keeps failing, inspect the deployment in the EAS Hosting dashboard.",
{ details: { status: manifest.status, ...(served ? { assetStatus: served.status } : {}), report: expected.query } });
}
}
if (check.listing !== undefined) {
const list = await get(`/p/${check.token}/reports.json`);
const body = list.status === 200 ? await list.text() : "";
if (list.status !== 200 || body !== check.listing) return deployError("PREVIEW_VERIFY_FAILED", "The hosted preview did not list the targets that were deployed.",
"Rerun the deploy job; the deployment's report listing did not match.", { details: { status: list.status } });
}
return undefined;
};
const started = check.now();
let liveAfterMs = 0;
for (;;) {
// A browser's request: the token page negotiates, and anything that does not ask for HTML gets the Markdown index.
const shell = await get(`/p/${check.token}`, "text/html");
if (shell.status === 200 && (shell.headers.get("content-type") ?? "").startsWith("text/html")) { liveAfterMs = check.now() - started; break; }
const serving = shell.status === 200 && (shell.headers.get("content-type") ?? "").startsWith("text/html");
// An alias with a fixed token (the example report) serves the token page from the deployment it still points at, so
// it has moved only once it serves this deployment's listing and reports as well.
const stale = serving && check.alias ? await mismatch() : undefined;
if (serving && !stale) { liveAfterMs = check.now() - started; break; }
const waitedMs = check.now() - started;
if (waitedMs > check.timeoutMs) {
// An alias that does not serve the token page yet still points at an earlier deployment; the deployment itself not serving is a failed deployment.
if (check.alias) throw deployError("PREVIEW_ALIAS_BEHIND", `The alias ${check.origin} still answers HTTP ${shell.status} ${Math.round(waitedMs / 1000)} s after eas deploy returned.`,
if (check.alias) throw deployError("PREVIEW_ALIAS_BEHIND", stale
? `The alias ${check.origin} still serves an earlier deployment ${Math.round(waitedMs / 1000)} s after eas deploy returned.`
: `The alias ${check.origin} still answers HTTP ${shell.status} ${Math.round(waitedMs / 1000)} s after eas deploy returned.`,
"Nothing to do: the alias catches up on its own, and the next push verifies it again.", { retryable: true, details: { origin: check.origin, status: shell.status, waitedMs } });
throw deployError("PREVIEW_VERIFY_FAILED", `The hosted preview did not become available (HTTP ${shell.status}).`,
"Rerun the deploy job; if the deployment keeps failing verification, inspect it in the EAS Hosting dashboard.", { retryable: true, details: { origin: check.origin, status: shell.status, waitedMs } });
@@ -785,23 +948,16 @@ async function verifyDeployment(check: { origin: string; token: string; reports:
if (api.status !== 200 || !(api.headers.get("content-type") ?? "").startsWith("application/json")) {
throw deployError("PREVIEW_VERIFY_FAILED", `The hosted preview did not answer the agent API (HTTP ${api.status}).`, "Rerun the deploy job; if the deployment keeps failing verification, inspect it in the EAS Hosting dashboard.", { retryable: true });
}
for (const expected of check.reports) {
const report = await get(`/p/${check.token}/report.zip${expected.query}`);
const bytes = report.status === 200 ? new Uint8Array(await report.arrayBuffer()) : new Uint8Array();
if (report.status !== 200 || bytes.byteLength !== expected.bytes || createHash("sha256").update(bytes).digest("hex") !== expected.sha256) {
throw deployError("PREVIEW_VERIFY_FAILED", "The hosted preview did not serve the exact report that was deployed.",
"Rerun the deploy job; the deployment's sealed key or ciphertext did not match.", { details: { status: report.status, report: expected.query } });
}
}
if (check.listing !== undefined) {
const list = await get(`/p/${check.token}/reports.json`);
const body = list.status === 200 ? await list.text() : "";
if (list.status !== 200 || body !== check.listing) throw deployError("PREVIEW_VERIFY_FAILED", "The hosted preview did not list the targets that were deployed.",
"Rerun the deploy job; the deployment's report listing did not match.", { details: { status: list.status } });
const robots = await get("/robots.txt");
if (robots.status !== 200 || await robots.text() !== PREVIEW_ROBOTS) {
throw deployError("PREVIEW_VERIFY_FAILED", `The hosted preview did not serve its robots.txt (HTTP ${robots.status}), which keeps crawlers away from the screenshots.`,
"Rerun the deploy job; if it keeps failing, inspect the deployment in the EAS Hosting dashboard.", { retryable: true });
}
// The deployment itself serves exactly what was deployed or it failed; the alias was checked while it was polled.
if (!check.alias) { const failure = await mismatch(); if (failure) throw failure; }
const other = check.token.endsWith("a") ? `${check.token.slice(0, -1)}b` : `${check.token.slice(0, -1)}a`;
// Errors name each probe by its shape: the near-miss token differs from the real one in its last character only.
for (const [path, shape] of [["/", "/"], [`/p/${other}`, "/p/<wrong token>"], [`/p/${other}/report.zip`, "/p/<wrong token>/report.zip"],
for (const [path, shape] of [["/", "/"], [`/p/${other}`, "/p/<wrong token>"], [`/p/${other}/manifest.json`, "/p/<wrong token>/manifest.json"],
[`/p/${check.token}/viewer/index.html`, "/p/<token>/viewer/index.html"]] as const) {
const response = await get(path);
if (response.status !== 404) throw deployError("PREVIEW_VERIFY_FAILED", `The hosted preview answered HTTP ${response.status} without a valid token at ${shape}.`,
+75
View File
@@ -0,0 +1,75 @@
/**
* What a hosted or local report page serves the Viewer instead of the whole ZIP: the manifest (the report's metadata
* file verbatim plus every asset's size) and each asset on request. `ci:deploy` writes them as a hosted preview's
* static files, the local report server answers them from the ZIP, and the preview Worker reads the manifest back.
*/
import { reportFromJson, type LoadedReport } from "./agent-api.js";
/** The manifest format the Viewer accepts (viewer/src/utils/artifact-report.ts). */
export const PREVIEW_MANIFEST_FORMAT = "explore.preview-manifest/1";
/** `assets/<sha256>.<ext>`, the only asset names a report page serves. */
export const PREVIEW_ASSET = /^assets\/([a-f0-9]{64})\.(png|jpg|mp4)$/;
export const ASSET_TYPES: Record<string, string> = { png: "image/png", jpg: "image/jpeg", mp4: "video/mp4" };
export const MAX_ASSET_BYTES = 32 * 1024 * 1024;
/**
* The metadata file verbatim plus the size of every asset, so the Viewer opens the report before any screenshot.
* Directory records are not assets. `deployment` names the hosted deployment that wrote it, so a check can tell a new
* deployment of an unchanged report from the one an alias still points at.
*/
export function manifestBody(loaded: LoadedReport, deployment?: string): string {
const entries: Record<string, number> = {};
for (const [name, entry] of loaded.entries) if (name.startsWith("assets/") && !name.endsWith("/")) entries[name] = entry.size;
const stamp = deployment ? `"deployment":${JSON.stringify(deployment)},` : "";
return `{"format":"${PREVIEW_MANIFEST_FORMAT}",${stamp}"file":"${loaded.file}","entries":${JSON.stringify(entries)},"report":${new TextDecoder().decode(loaded.raw)}}`;
}
/** The largest metadata file the Viewer opens (`ARTIFACT_LIMITS.metadata` in viewer/src/utils/artifact-report.ts). */
export const MAX_PREVIEW_METADATA_BYTES = 8 * 1024 * 1024;
/** A manifest is the metadata file plus one size per asset; the Viewer refuses a larger one. */
export const MAX_MANIFEST_BYTES = MAX_PREVIEW_METADATA_BYTES + 2 * 1024 * 1024;
const REPORT_KEY = ',"report":';
/**
* The report a manifest describes, reading its assets through `readAsset`; undefined when it is not a manifest of a report.
* The metadata is parsed once and kept as the manifest's own bytes. `manifestBody` writes `report` last, so its bytes are
* the text between `,"report":` and the closing brace; any other layout is parsed whole and re-serialized.
*/
export function reportFromManifest(bytes: Uint8Array, readAsset: LoadedReport["read"]): LoadedReport | undefined {
if (bytes.byteLength > MAX_MANIFEST_BYTES) return undefined;
const text = new TextDecoder().decode(bytes);
let header: unknown, report: unknown, raw: Uint8Array | undefined;
const at = text.indexOf(REPORT_KEY);
// A string cannot hold `,"report":` (its quotes are escaped); a nested key can, and then the header before it is not JSON on its own.
if (at !== -1 && text.endsWith("}") && /^[\x20-\x7e]*$/.test(text.slice(0, at))) {
try {
header = JSON.parse(`${text.slice(0, at)}}`);
report = JSON.parse(text.slice(at + REPORT_KEY.length, -1));
raw = bytes.subarray(at + REPORT_KEY.length, bytes.byteLength - 1);
} catch { header = undefined; }
}
if (raw === undefined) {
try { header = JSON.parse(text); } catch { return undefined; }
report = (header as { report?: unknown } | null)?.report;
raw = new TextEncoder().encode(JSON.stringify(report));
}
if (!header || typeof header !== "object") return undefined;
const { format, file, entries } = header as { format?: unknown; file?: unknown; entries?: unknown };
if (format !== PREVIEW_MANIFEST_FORMAT || (file !== "review.json" && file !== "exploration.json")) return undefined;
if (!entries || typeof entries !== "object" || Array.isArray(entries) || !report || typeof report !== "object") return undefined;
const sizes = new Map<string, { size: number }>();
for (const [name, size] of Object.entries(entries)) {
if (!PREVIEW_ASSET.test(name) || !Number.isSafeInteger(size) || (size as number) < 0) return undefined;
sizes.set(name, { size: size as number });
}
if (raw.byteLength > MAX_PREVIEW_METADATA_BYTES) return undefined;
return reportFromJson(file, raw, report, sizes, (name, limit) => (sizes.has(name) ? readAsset(name, limit) : Promise.resolve(undefined)));
}
/** One asset of the report with its content type; undefined when the report has no such asset. */
export async function previewAsset(loaded: LoadedReport, path: string): Promise<{ bytes: Uint8Array; type: string } | undefined> {
const match = PREVIEW_ASSET.exec(path);
if (!match || !loaded.entries.has(path)) return undefined;
const bytes = await loaded.read(path, MAX_ASSET_BYTES);
return bytes ? { bytes, type: ASSET_TYPES[match[2]!]! } : undefined;
}
+1 -1
View File
@@ -15,7 +15,7 @@ export type ViewerDeepLinks = {
surface: (surfaceId: string) => string;
};
/** A hosted preview: the Viewer and its decrypted report served from one token-gated EAS Hosting deployment. */
/** A hosted preview: the Viewer and its report served from one token-gated EAS Hosting deployment. */
export type PreviewLinks = ViewerDeepLinks & {
/** Epoch milliseconds after which the deployment's Worker stops serving the report; null when the link does not expire. */
expiresAt: number | null;
+1 -1
View File
@@ -1,5 +1,5 @@
/**
* Reads the sealed report archive once it is decrypted: a ZIP written by `zipReportDirectory`
* Reads a report ZIP (deploy, the local report server, the agent API): a ZIP written by `zipReportDirectory`
* (review.json or exploration.json deflated, assets/<sha256>.png stored). Only the central directory
* is trusted for sizes and offsets, every offset is bounds-checked, and entries are inflated on demand.
*/
+18 -3
View File
@@ -1,10 +1,10 @@
import type { RunTarget } from "../types.js";
import { appearancePairs, bundleColorSchemes, routeIdOfScreen } from "../capture/appearance.js";
import { appearancePairs, bundleColorSchemes, failedPassOutcome, routeIdOfScreen } from "../capture/appearance.js";
import { createHash } from "node:crypto";
import { lstat, readFile, realpath } from "node:fs/promises";
import { isAbsolute, relative, resolve, sep } from "node:path";
import { assertDecodedPng } from "../shared/png-evidence.js";
import { assertAgentAcceptance, validateCapturePlan, type AppIconImage, type BundleLaunchCapture, type CaptureBundle, type CaptureOutcome, type CapturedImage, type LaunchFilmstrip, type LaunchFrame, type PublicLaunchSide, type PublicReview, homeIndicatorBandPercent, launchSurfaceId, statusBarBandPercent } from "../contracts/capture.js";
import { assertAgentAcceptance, validateCapturePlan, type AppIconImage, type BundleLaunchCapture, type CaptureBundle, type CaptureOutcome, type CapturedImage, type LaunchFilmstrip, type LaunchFrame, type PublicLaunchSide, type PublicReview, homeIndicatorBandPercent, launchSurfaceId, statusBarBandPercent, systemRegions, type SystemRegion } from "../contracts/capture.js";
import { COLOR_SCHEMES } from "../capture/appearance.js";
import { launchDirectory } from "../capture/launch-capture.js";
import { APP_ICON_FILE, APP_ICON_MAX_SIZE } from "../capture/app-icon.js";
@@ -95,9 +95,11 @@ function secondaryPassOutcome(bundle: CaptureBundle, selected: ReadonlySet<strin
const { routeId, scheme } = routeIdOfScreen(id);
const own = bundle.outcomes[id];
const verified = own?.status === "captured" && own.landing === "verified";
return scheme !== undefined && selected.has(routeId) && verified !== (bundle.outcomes[routeId]?.status === "captured");
return scheme !== undefined && selected.has(routeId)
&& (verified !== (bundle.outcomes[routeId]?.status === "captured") || failedPassOutcome(bundle, id));
}
export async function readCaptureBundle(directory: string): Promise<CaptureBundle> {
const bundle = JSON.parse(await readFile(resolve(await realpath(directory), "bundle.json"), "utf8")) as CaptureBundle;
if (!bundle || bundle.contract !== "explore.bundle" || typeof bundle.gitCommit !== "string"
@@ -110,6 +112,11 @@ export async function readCaptureBundle(directory: string): Promise<CaptureBundl
throw new Error("Capture bundle completion timestamp is invalid");
}
if (bundle.appIcon !== undefined) assertAppIconRecord(bundle.appIcon);
if (bundle.sessionFailures !== undefined && (!Array.isArray(bundle.sessionFailures) || bundle.sessionFailures.some(failure => !failure
|| !["light", "dark"].includes(failure.scheme) || !failure.error || typeof failure.error.code !== "string" || typeof failure.error.message !== "string"
|| typeof failure.error.stage !== "string" || typeof failure.error.nextAction !== "string" || typeof failure.error.retryable !== "boolean"))) {
throw new Error("Capture bundle session failures are invalid");
}
if (bundle.launch !== undefined) assertLaunchRecords(bundle.launch);
const selected = new Set(bundle.selectedSurfaceIds);
if (selected.size !== bundle.selectedSurfaceIds.length
@@ -263,11 +270,17 @@ export type CaptureReviewOptions = {
/** The target this report captures, and every target config.yaml captures (this one among them), when more than one. */
target?: RunTarget;
targets?: RunTarget[];
/**
* The system-drawn areas the verdict left out of each screenshot; absent, the platform's bands, as the assessment's
* default. A bundle does not record a foldable's pose, so its panel is not guessed here.
*/
systemRegions?: SystemRegion[];
};
/** Build the validated review independently of where its metadata and images will be stored. */
async function constructCaptureReview(options: CaptureReviewOptions, store: ScreenshotStore) {
const plan = validateCapturePlan(options.plan);
const excluded = new Set(plan.excludedSurfaceIds ?? []);
const pr = projectReviewPresentation(options.pullRequest);
if (pr.head.sha !== plan.headSha || pr.base.sha !== plan.baseSha) throw new Error("PR identity differs from capture plan");
const run = `${identifier(options.runId)}-${identifier(options.runAttempt)}`;
@@ -301,6 +314,7 @@ async function constructCaptureReview(options: CaptureReviewOptions, store: Scre
runtime: { projectId: head.bundle.easProjectId ?? "", buildId: head.bundle.easBuildId ?? "", gitCommit: plan.headSha,
platform: head.bundle.platform ?? "ios", ...(head.bundle.controller === "argent" ? { controller: "argent" as const } : {}), statusBarBandPercent: statusBarBandPercent(head.bundle.platform),
...(homeIndicatorBandPercent(head.bundle.platform) > 0 ? { homeIndicatorBandPercent: homeIndicatorBandPercent(head.bundle.platform) } : {}),
systemRegions: (options.systemRegions ?? systemRegions({ platform: head.bundle.platform })).map((area) => ({ ...area })),
...(options.platforms && options.platforms.length > 1 ? { platforms: [...options.platforms] } : {}),
...(options.target ? { target: { ...options.target } } : {}),
...(options.targets && options.targets.length > 1 ? { targets: options.targets.map(target => ({ ...target })) } : {}),
@@ -341,6 +355,7 @@ async function constructCaptureReview(options: CaptureReviewOptions, store: Scre
return { id, status, base: headOnly ? null : withId(s.baseSurface), head: withId(s.headSurface),
...(colorSchemes.length > 1 ? { colorScheme: scheme } : {}),
...(s.sourceChange ? { sourceChange: s.sourceChange } : {}),
...(excluded.has(s.id) ? { excluded: true as const } : {}),
...(impactKinds(s).length ? { sourceImpact: impactKinds(s) } : {}),
...(!headOnly && surfaceSourceScope(s) ? { sourceScope: surfaceSourceScope(s)! } : {}),
...(baseOutcome ? { baseOutcome } : {}),
+278 -66
View File
@@ -1,6 +1,6 @@
import { unattributedIssue } from "../capture/health.js";
import { appearancePairs, bundleColorSchemes } from "../capture/appearance.js";
import { homeIndicatorBandPercent, launchSurfaceId, statusBarBandPercent } from "../contracts/capture.js";
import { appearancePairs, bundleColorSchemes, failedPassOutcome, type ColorScheme } from "../capture/appearance.js";
import { launchSurfaceId, systemRegions as defaultSystemRegions, type SystemRegion } from "../contracts/capture.js";
import { readFileSync, statSync } from "node:fs";
import { stat } from "node:fs/promises";
import { isAbsolute, resolve } from "node:path";
@@ -26,8 +26,10 @@ import {
type VisualEvidenceAssessment,
} from "./visual-evidence.js";
import { AGENT_DEVICE_VERSION } from "../shared/agent-device-version.js";
import { STATUS_BAR_BAND_PERCENT } from "../shared/png-evidence.js";
import { measureRegionShifts } from "./region-shift.js";
import { STATUS_BAR_BAND_PERCENT, decodePng, differentPixelsInRect, maxColorDistance, type DecodedFrame } from "../shared/png-evidence.js";
import type { RegionMove, RegionShift } from "./region-shift.js";
import { measureRegionMovesInWorker } from "./region-shift-worker.js";
import { availableParallelism } from "node:os";
export const DEFAULT_VISUAL_THRESHOLD = 0.1;
export const DEFAULT_EVIDENCE_TIME_BUDGET_MS = 10 * 60_000;
@@ -77,8 +79,16 @@ export type PublicEvidenceSummary = {
comparison: { planned: number; eligible: number; skipped: number };
/** Every unavailable or unverified requested side, plus operational comparison errors. */
diagnostics: PublicEvidenceDiagnostic[];
/**
* Each appearance pass of a side that failed while its primary one captured (the bundle's `sessionFailures`), with how
* many of its frames the failure left unavailable: the comment names it once, not once per frame.
*/
sessionFailures?: PublicSessionFailure[];
};
export type PublicSessionFailure = { side: "base" | "head"; scheme: ColorScheme; frames: number;
code: string; stage: string; message: string; nextAction: string; retryable: boolean };
const MAX_SUMMARY_ERRORS = 5;
function recordError(errors: string[], surfaceId: string, error: unknown): void {
@@ -142,16 +152,61 @@ function visualVerdict(assessment: VisualEvidenceAssessment, threshold: number):
};
}
type PublishedRegion = NonNullable<PublicVisualVerdict["regions"]>[number];
/** A region as published: the fields a review carries, without the comparison's own bookkeeping. */
function publishedRegion(region: VisualDiffRegion): PublishedRegion {
return { rect: region.rect, normalizedRect: region.normalizedRect, differentPixels: region.differentPixels,
shareOfDiffPercentage: region.shareOfDiffPercentage, dominantChange: region.dominantChange, location: region.location };
}
/**
* The regions to publish from every region the comparison found, given what moved: each region with its small
* shift, if one explains it, and each large move as one region in place of the regions it explains (where the
* content is now, their pixels and share, its location the region the content arrived in), largest share first.
* `allRegionsMoved` when every region, before any are left out, moved.
*/
export function publishMeasuredRegions(all: readonly VisualDiffRegion[], shifts: ReadonlyArray<RegionShift | undefined>,
moves: readonly RegionMove[], width: number, height: number): Pick<PublicVisualVerdict, "regions" | "allRegionsMoved"> {
const merged = new Set(moves.flatMap(move => move.members));
const percent = (value: number, total: number) => Math.round((value / total) * 10_000) / 100;
const regions: PublishedRegion[] = [
...all.flatMap((region, index) => {
if (merged.has(index)) return [];
const shift = shifts[index];
return [{ ...publishedRegion(region), ...(shift ? { shift } : {}) }];
}),
...moves.map(move => {
const members = move.members.map(index => all[index]!);
return {
rect: move.rect,
normalizedRect: { x: percent(move.rect.x, width), y: percent(move.rect.y, height), width: percent(move.rect.width, width), height: percent(move.rect.height, height) },
differentPixels: members.reduce((sum, region) => sum + region.differentPixels, 0),
// Each member's share is rounded, so their sum can pass 100; a share is at most the whole difference.
shareOfDiffPercentage: Math.min(100, Math.round(members.reduce((sum, region) => sum + region.shareOfDiffPercentage, 0) * 100) / 100),
dominantChange: "mixed", location: members[0]!.location, shift: move.shift,
};
}),
];
return {
regions: regions.sort((left, right) => right.shareOfDiffPercentage - left.shareOfDiffPercentage).slice(0, MAX_REGIONS),
...(regions.length > 0 && regions.every(region => region.shift) ? { allRegionsMoved: true as const } : {}),
};
}
/**
* The published regions with how far their content moved, where a move explains it: a layout that shifted by a
* device pixel reads as moved, not as changed. Every region the comparison found is measured, not only the ones
* published, so `allRegionsMoved` can say nothing else on the screen changed. A measurement that fails leaves the
* regions as they were.
* device pixel reads as moved, not as changed, and an element that moved further (a card recentred on a wider
* screen) reads as one moved region instead of the fragments the comparison split it into. Every region the
* comparison found is measured, not only the ones published, so `allRegionsMoved` can say nothing else on the
* screen changed. A measurement that fails leaves the regions as they were.
*/
async function withRegionShifts(basePath: string, headPath: string, all: readonly VisualDiffRegion[], published: NonNullable<PublicVisualVerdict["regions"]>):
Promise<Pick<PublicVisualVerdict, "regions" | "allRegionsMoved">> {
try {
const shifts = await measureRegionShifts(basePath, headPath, all.map(region => region.rect));
const { shifts, moves, width, height } = await measureRegionMovesInWorker(basePath, headPath,
all.map(region => ({ ...region.rect, dominantChange: region.dominantChange })));
if (moves.length) return publishMeasuredRegions(all, shifts, moves, width, height);
const shiftOf = new Map(all.map((region, index) => [region.rect, shifts[index]] as const));
return {
regions: published.map(region => {
@@ -171,31 +226,103 @@ export function ignoreStatusBar(measurement: ScreenshotDiffMeasurement, bandPerc
return { measurement: bands.measurement, ignored: bands.ignoredTop };
}
/**
* Leaves out of a measurement every diff region that lies entirely inside the system's bands: the status bar at the
* top and, on iOS, the home indicator at the bottom. Both are drawn by the OS and differ between captures without
* the app having changed. A region that crosses into the app's area is kept whole.
*/
export function ignoreSystemBands(measurement: ScreenshotDiffMeasurement, bands: { top: number; bottom: number }):
/** Both frames a measurement compared, decoded on demand: only a region that crosses a band's edge needs them. */
export type SystemBandFrames = { baseline: DecodedFrame; current: DecodedFrame; distance: number };
export type SystemBandFrameLoader = () => SystemBandFrames | undefined;
/** Reads the measurement's two screenshots when a band's edge has to be resolved pixel by pixel. */
export function systemBandFrameLoader(measurement: ScreenshotDiffMeasurement, threshold: number): SystemBandFrameLoader {
return () => {
try {
const baseline = decodePng(readFileSync(measurement.baselinePath)); const current = decodePng(readFileSync(measurement.currentPath));
if (!baseline || !current || baseline.width !== current.width || baseline.height !== current.height) return undefined;
return { baseline, current, distance: maxColorDistance(threshold) };
} catch { return undefined; }
};
}
/** `ignoreSystemRegions` for the plain case: a status-bar band across the top and a home-indicator band along the bottom. */
export function ignoreSystemBands(measurement: ScreenshotDiffMeasurement, bands: { top: number; bottom: number }, frames?: SystemBandFrameLoader):
{ measurement: ScreenshotDiffMeasurement; ignoredTop: number; ignoredBottom: number } {
if (measurement.match || measurement.regions.length === 0) return { measurement, ignoredTop: 0, ignoredBottom: 0 };
const inTop = (region: VisualDiffRegion) => bands.top > 0 && region.normalizedRect.y + region.normalizedRect.height <= bands.top;
const inBottom = (region: VisualDiffRegion) => bands.bottom > 0 && region.normalizedRect.y >= 100 - bands.bottom;
const inBand = (region: VisualDiffRegion) => inTop(region) || inBottom(region);
const ignoredRegions = measurement.regions.filter(inBand);
if (ignoredRegions.length === 0) return { measurement, ignoredTop: 0, ignoredBottom: 0 };
const ignoredPixels = ignoredRegions.reduce((total, region) => total + region.differentPixels, 0);
const regions: SystemRegion[] = [
...(bands.top > 0 ? [{ kind: "status-bar" as const, x: 0, y: 0, width: 100, height: bands.top }] : []),
...(bands.bottom > 0 ? [{ kind: "home-indicator" as const, x: 0, y: 100 - bands.bottom, width: 100, height: bands.bottom }] : []),
];
const result = ignoreSystemRegions(measurement, regions, frames);
return { measurement: result.measurement, ignoredTop: result.ignoredStatusBar, ignoredBottom: result.ignoredHomeIndicator };
}
/**
* Leaves out of a measurement every diff region that lies inside a system-drawn area: the status bar (a band across
* the top of a phone, a pill in the rail of a foldable) and, on iOS, the home indicator at the bottom. Both differ
* between captures without the app having changed. A region that crosses the edge of a band across the whole width
* is clipped to the part outside it when the frames can be read: its pixels inside the band are dropped, and it stays
* only for what differs in the app's area. Without frames, and across a pill's edge, such a region is kept whole.
* (A region's pixel count is its own cluster's, while its rect can overlap other regions, so what it keeps is counted
* outside the band and clamped to its own count.)
*/
export function ignoreSystemRegions(measurement: ScreenshotDiffMeasurement, system: readonly SystemRegion[], frames?: SystemBandFrameLoader):
{ measurement: ScreenshotDiffMeasurement; ignoredStatusBar: number; ignoredHomeIndicator: number } {
if (measurement.match || measurement.regions.length === 0 || system.length === 0) return { measurement, ignoredStatusBar: 0, ignoredHomeIndicator: 0 };
// agent-device rounds each of x, y, width and height to 0.01%, so an edge it reports can be up to 0.01% past the real one.
const slack = 0.015;
const inside = (region: VisualDiffRegion, area: SystemRegion) => {
const rect = region.normalizedRect;
return rect.x >= area.x - slack && rect.y >= area.y - slack
&& rect.x + rect.width <= area.x + area.width + slack && rect.y + rect.height <= area.y + area.height + slack;
};
const across = system.filter((area) => area.x <= slack && area.x + area.width >= 100 - slack);
const topBand = across.find((area) => area.y <= slack);
const bottomBand = across.find((area) => area !== topBand && area.y + area.height >= 100 - slack);
const crossesTop = (region: VisualDiffRegion) => topBand !== undefined && region.normalizedRect.y < topBand.y + topBand.height - slack;
const crossesBottom = (region: VisualDiffRegion) => bottomBand !== undefined && region.normalizedRect.y + region.normalizedRect.height > bottomBand.y + slack;
let loaded: SystemBandFrames | undefined | null = null;
const load = () => { if (loaded === null) loaded = frames?.() ?? undefined; return loaded; };
const ignored: Record<SystemRegion["kind"], number> = { "status-bar": 0, "home-indicator": 0 };
let ignoredPixels = 0;
const remaining: VisualDiffRegion[] = [];
for (const region of measurement.regions) {
const area = system.find((candidate) => inside(region, candidate));
if (area) { ignored[area.kind] += 1; ignoredPixels += region.differentPixels; continue; }
const top = crossesTop(region) ? topBand : undefined; const bottom = crossesBottom(region) ? bottomBand : undefined;
const decoded = top || bottom ? load() : undefined;
if (!decoded) { remaining.push(region); continue; }
const { height, width } = decoded.baseline;
const rect = { ...region.rect };
if (top) {
const bandRows = Math.floor((height * (top.y + top.height)) / 100);
const rows = Math.max(0, Math.min(rect.y + rect.height, bandRows) - rect.y); rect.y += rows; rect.height -= rows;
}
if (bottom) rect.height = Math.min(rect.y + rect.height, height - Math.floor((height * bottom.height) / 100)) - rect.y;
// Counted in what is left of the rect, not in what the band takes: the rect can enclose another region's pixels,
// which can only add to what is kept (up to the region's own count), never take from it.
const differentPixels = rect.height > 0 ? Math.min(region.differentPixels, differentPixelsInRect(decoded.baseline, decoded.current, rect, decoded.distance)) : 0;
const inBand = region.differentPixels - differentPixels;
if (differentPixels === 0) {
ignored[(top ?? bottom)!.kind] += 1;
ignoredPixels += region.differentPixels;
continue;
}
// Its pixels inside the band were left out too, so the band is marked as not compared, as for a dropped region.
if (inBand > 0) { if (top) ignored[top.kind] += 1; if (bottom) ignored[bottom.kind] += 1; }
ignoredPixels += inBand;
const round = (value: number) => Math.round(value * 100) / 100;
remaining.push({ ...region, rect, differentPixels,
normalizedRect: { x: round((rect.x / width) * 100), y: round((rect.y / height) * 100), width: round((rect.width / width) * 100), height: round((rect.height / height) * 100) } });
}
if (ignoredPixels === 0) return { measurement, ignoredStatusBar: 0, ignoredHomeIndicator: 0 };
const differentPixels = Math.max(0, measurement.differentPixels - ignoredPixels);
const remaining = measurement.regions.filter((region) => !inBand(region));
const regions = remaining.map((region) => ({ ...region,
shareOfDiffPercentage: differentPixels > 0 ? Math.round((region.differentPixels / differentPixels) * 10000) / 100 : 0 }));
return {
ignoredTop: ignoredRegions.filter(inTop).length,
ignoredBottom: ignoredRegions.filter((region) => !inTop(region)).length,
ignoredStatusBar: ignored["status-bar"],
ignoredHomeIndicator: ignored["home-indicator"],
measurement: {
...measurement,
differentPixels,
mismatchPercentage: measurement.totalPixels > 0 ? Math.round((differentPixels / measurement.totalPixels) * 10000) / 100 : 0,
match: differentPixels === 0 || remaining.length === 0,
regions: remaining,
match: differentPixels === 0 || regions.length === 0,
regions,
},
};
}
@@ -289,6 +416,40 @@ function impactFor(surface: PlanSurface, presence: ComparatorSurfaceImpact["pres
};
}
/** The screenshot pairs `assessPublicEvidence` compares: both sides captured and landed, in the loop's order. */
function comparablePairs(options: { plan: SourceCapturePlan; base: PublicEvidenceSide; head: PublicEvidenceSide }, colorSchemes: readonly ColorScheme[],
baseColorSchemes: readonly ColorScheme[]): Array<{ id: string; basePath: string; headPath: string }> {
const pairs: Array<{ id: string; basePath: string; headPath: string }> = [];
for (const surface of options.plan.surfaces) {
if (surface.baseSurface?.kind !== "screen" || surface.headSurface?.kind !== "screen" || (!surface.baseCapture && !surface.headCapture)) continue;
for (const { id, baseId } of appearancePairs(surface.id, colorSchemes, baseColorSchemes)) {
if (baseId === null) continue;
const basePath = localImagePath(options.base, baseId);
const headPath = localImagePath(options.head, id);
if (!basePath || !headPath || captureDiagnostic(surface, "base", options.base, baseId) || captureDiagnostic(surface, "head", options.head, id)) continue;
pairs.push({ id, basePath, headPath });
}
}
return pairs;
}
/** Runs tasks at most `size` at a time, each started in the order it was queued. */
function concurrencyLimit(size: number): <T>(task: () => Promise<T>) => Promise<T> {
let active = 0;
const queue: Array<() => void> = [];
const next = () => { if (active < size) queue.shift()?.(); };
return task => new Promise((resolve, reject) => {
queue.push(() => {
active += 1;
task().then(resolve, reject).finally(() => { active -= 1; next(); });
});
next();
});
}
/** The limit the reviews in this process share when none is given: a local review compares its targets at once. */
let processLimit: ReturnType<typeof concurrencyLimit> | undefined;
function localImagePath(side: PublicEvidenceSide, id: string): string | undefined {
const frame = side.bundle.images[id];
return frame ? localEvidencePath(side, frame.path) : undefined;
@@ -309,19 +470,21 @@ function localEvidencePath(side: PublicEvidenceSide, relativePath: string): stri
*/
export function decideReview(scope: SurfaceSourceScope, visual: PublicVisualVerdict, summary?: WhatChangedSummary): PublicSurfaceReview {
const facts = summary?.facts.map((fact) => ({ category: fact.category, templateId: fact.templateId, bullet: truncate(fact.bullet) })) ?? [];
const differs = visual.verdict === "modified" || facts.length > 0;
// Text that also differs between two captures of the same build is runtime state: listed, never what makes a change.
const changes = facts.filter((fact) => fact.templateId !== "visible-text.unstable-within-build");
const differs = visual.verdict === "modified" || changes.length > 0;
const inconclusive = !differs && (visual.verdict === "needs-review" || Boolean(summary?.needsReview));
const issue = summary?.reviewIssues[0]?.reason;
const evidenceText = facts[0]?.bullet ?? visualDescription(visual);
const evidenceText = changes[0]?.bullet ?? visualDescription(visual);
if (scope === "none") {
if (!differs) return { status: "unchanged", description: "No visible change" };
// The comparator's generic visual fact reads as a styling change; a variant names what differs.
const state = facts.find((fact) => fact.category !== "visual")?.bullet
const state = [...changes, ...facts].find((fact) => fact.category !== "visual")?.bullet
?? `${percentage(visual.mismatchPercentage)}${visual.regions?.[0]?.location ? ` · ${visual.regions[0].location}` : ""}`;
return { status: "variant", description: truncate(`Runtime state differs: ${state}`), ...(facts.length ? { facts } : {}) };
}
if (differs) {
const description = !facts.length && issue ? `${visualDescription(visual)} · ${issue}` : evidenceText;
const description = !changes.length && issue ? `${visualDescription(visual)} · ${issue}` : evidenceText;
return { status: "modified", description: truncate(description), ...(facts.length ? { facts } : {}) };
}
if (inconclusive) {
@@ -369,6 +532,8 @@ function captureDiagnostic(surface: PlanSurface, side: "base" | "head", evidence
export async function assessPublicEvidence(options: {
/** The captured platform; its status bar band is left out of the visual verdict. Absent: iOS. */
platform?: "ios" | "android";
/** The system-drawn areas left out of the visual verdict; absent, the platform's bands. */
systemRegions?: SystemRegion[];
plan: SourceCapturePlan;
base: PublicEvidenceSide;
head: PublicEvidenceSide;
@@ -378,8 +543,11 @@ export async function assessPublicEvidence(options: {
timeBudgetMs?: number;
now?: () => number;
measure?: (baselinePath: string, currentPath: string) => Promise<ScreenshotDiffMeasurement>;
/** How many screenshot pairs compare at once. Absent: one per available CPU, shared by every review in this process. */
concurrency?: number;
}): Promise<PublicEvidenceResult> {
const threshold = options.threshold ?? DEFAULT_VISUAL_THRESHOLD;
const system = options.systemRegions ?? defaultSystemRegions({ platform: options.platform });
const now = options.now ?? Date.now;
const deadline = now() + (options.timeBudgetMs ?? DEFAULT_EVIDENCE_TIME_BUDGET_MS);
const surfaces: Record<string, PublicSurfaceAssessment> = {};
@@ -395,6 +563,34 @@ export async function assessPublicEvidence(options: {
const measure = options.measure ?? (async (baselinePath: string, currentPath: string) => (
await diffScreenshots({ baselinePath, currentPath, command: options.agentDeviceCommand, threshold })
));
type ComparedPair = { visual: PublicVisualVerdict; assessment: VisualEvidenceAssessment };
/** One pair's measurement, verdict and region moves: the work that runs ahead of the surface loop. */
const comparePair = async (basePath: string, headPath: string): Promise<ComparedPair> => {
const raw = await measure(basePath, headPath);
const { measurement, ignoredStatusBar: ignoredTop, ignoredHomeIndicator: ignoredBottom } = ignoreSystemRegions(raw, system, systemBandFrameLoader(raw, threshold));
const assessment = assessVisualMeasurements({
baselineSamples: 1,
currentSamples: 1,
requiredSamples: 1,
baselineWithin: [],
currentWithin: [],
crossBuild: [measurement],
representative: measurement,
});
const visual = visualVerdict(assessment, threshold);
// Measuring moves is extra: once the comparison budget is spent, regions go out unmeasured.
if (visual.verdict === "modified" && visual.regions?.length && now() <= deadline) {
Object.assign(visual, await withRegionShifts(basePath, headPath, measurement.regions, visual.regions));
}
if (ignoredTop > 0) visual.ignoredStatusBarRegions = ignoredTop;
if (ignoredBottom > 0) visual.ignoredHomeIndicatorRegions = ignoredBottom;
if (ignoredTop > 0 || ignoredBottom > 0) visual.systemRegions = system.map((area) => ({ ...area }));
return { visual, assessment };
};
/** A pair compared if the budget has not expired when its turn to start comes; undefined when it has. */
const compareInBudget = async (basePath: string, headPath: string): Promise<ComparedPair | undefined> => (
now() > deadline ? undefined : await comparePair(basePath, headPath)
);
const visualCounts = { compared: 0, modified: 0, unchanged: 0, needsReview: 0, failed: 0, errors: [] as string[] };
const accessibilityCounts = { compared: 0, described: 0, failed: 0, skipped: 0, errors: [] as string[] };
const reviewCounts = { modified: 0, needsReview: 0, variants: 0, unchanged: 0 };
@@ -411,6 +607,20 @@ export async function assessPublicEvidence(options: {
const colorSchemes = bundleColorSchemes(options.head.bundle);
const baseColorSchemes = bundleColorSchemes(options.base.bundle);
// The pairs the loop below compares start now, a few at a time: each is a comparator process and a pixel search
// that take seconds on a large screenshot, so they overlap instead of queueing behind one another. The loop reads
// them in its own order, so the result does not depend on which finishes first. A pair whose turn comes after the
// budget expired is not started; one started in time is waited for and used.
const ahead = new Map<string, Promise<ComparedPair | undefined>>();
if (visualAvailable) {
const limit = options.concurrency !== undefined ? concurrencyLimit(Math.max(1, Math.floor(options.concurrency)))
: processLimit ??= concurrencyLimit(availableParallelism());
for (const { id, basePath, headPath } of comparablePairs(options, colorSchemes, baseColorSchemes)) {
const started = limit(() => compareInBudget(basePath, headPath));
started.catch(() => undefined);
ahead.set(id, started);
}
}
for (const surface of options.plan.surfaces) {
for (const side of ["base", "head"] as const) {
if (!surface[`${side}Capture`] || surface[`${side}Surface`]?.kind !== "screen") continue;
@@ -439,7 +649,19 @@ export async function assessPublicEvidence(options: {
if (surface.baseSurface?.kind !== "screen" || surface.headSurface?.kind !== "screen"
|| (!surface.baseCapture && !surface.headCapture)) continue;
// Every appearance the capture rendered is compared like-for-like: `<id>@<scheme>` frames against each other.
for (const { id, scheme, baseId } of appearancePairs(surface.id, colorSchemes, baseColorSchemes)) {
const pairs = appearancePairs(surface.id, colorSchemes, baseColorSchemes);
// The screen's other appearances, rendered by both builds and verified on both: a label that varies between a
// build's own appearances is runtime state. An unverified capture may show another screen, so it is no evidence.
const baseRoute = surface.baseSurface, headRoute = surface.headSurface;
const siblings = (current: string) => pairs.flatMap((pair) => {
if (pair.id === current || pair.baseId === null || captureDiagnostic(surface, "base", options.base, pair.baseId)
|| captureDiagnostic(surface, "head", options.head, pair.id)) return [];
try {
return [{ base: matchedSide("base", options.base, { ...baseRoute, id: pair.baseId }, options.profile),
head: matchedSide("head", options.head, { ...headRoute, id: pair.id }, options.profile) }];
} catch { return []; }
});
for (const { id, scheme, baseId } of pairs) {
// An appearance the base never rendered has nothing to compare against: the head frame is new evidence.
if (baseId === null) {
if (localImagePath(options.head, id)) {
@@ -464,47 +686,29 @@ export async function assessPublicEvidence(options: {
continue;
}
let visual: PublicVisualVerdict;
let assessment: VisualEvidenceAssessment | undefined;
if (now() > deadline) {
let compared: ComparedPair | undefined;
try {
compared = await (ahead.get(id) ?? compareInBudget(basePath, headPath));
} catch (error) {
visualCounts.failed += 1;
const diagnostic = comparisonError(id, error, { code: "VISUAL_COMPARISON_FAILED", stage: "compare",
message: "Screenshot comparison failed.", nextAction: "Check the screenshot files and the agent-device comparison diagnostics, then rerun comparison.", retryable: true });
recordError(visualCounts.errors, id, diagnostic.message);
continue;
}
if (!compared) {
visualCounts.failed += 1;
const diagnostic = comparisonError(id, undefined, { code: "VISUAL_COMPARISON_BUDGET_EXHAUSTED", stage: "compare",
message: "The time budget expired before screenshot comparison.",
nextAction: "Increase the evidence comparison time budget or reduce the number of selected routes, then rerun comparison.", retryable: true });
recordError(visualCounts.errors, id, diagnostic.message);
continue;
} else {
try {
const raw = await measure(basePath, headPath);
const { measurement, ignoredTop, ignoredBottom } = ignoreSystemBands(raw, { top: statusBarBandPercent(options.platform), bottom: homeIndicatorBandPercent(options.platform) });
assessment = assessVisualMeasurements({
baselineSamples: 1,
currentSamples: 1,
requiredSamples: 1,
baselineWithin: [],
currentWithin: [],
crossBuild: [measurement],
representative: measurement,
});
visual = visualVerdict(assessment, threshold);
// Measuring moves is extra: once the comparison budget is spent, regions go out unmeasured.
if (visual.verdict === "modified" && visual.regions?.length && now() <= deadline) {
Object.assign(visual, await withRegionShifts(basePath, headPath, measurement.regions, visual.regions));
}
if (ignoredTop > 0) visual.ignoredStatusBarRegions = ignoredTop;
if (ignoredBottom > 0) visual.ignoredHomeIndicatorRegions = ignoredBottom;
visualCounts.compared += 1;
if (visual.verdict === "modified") visualCounts.modified += 1;
else if (visual.verdict === "unchanged") visualCounts.unchanged += 1;
else visualCounts.needsReview += 1;
} catch (error) {
visualCounts.failed += 1;
const diagnostic = comparisonError(id, error, { code: "VISUAL_COMPARISON_FAILED", stage: "compare",
message: "Screenshot comparison failed.", nextAction: "Check the screenshot files and the agent-device comparison diagnostics, then rerun comparison.", retryable: true });
recordError(visualCounts.errors, id, diagnostic.message);
continue;
}
}
const { visual, assessment } = compared;
visualCounts.compared += 1;
if (visual.verdict === "modified") visualCounts.modified += 1;
else if (visual.verdict === "unchanged") visualCounts.unchanged += 1;
else visualCounts.needsReview += 1;
const scope = surfaceSourceScope(surface) ?? "none";
let review: PublicSurfaceReview | undefined;
@@ -515,10 +719,12 @@ export async function assessPublicEvidence(options: {
base: matchedSide("base", options.base, { ...surface.baseSurface, id: baseId }, options.profile),
head: matchedSide("head", options.head, { ...surface.headSurface, id }, options.profile),
visualAssessment: assessment,
siblings: siblings(id),
});
const summary = summarizeReviewerChanges({ changes: result.reviewerChanges });
accessibilityCounts.compared += 1;
review = decideReview(scope, visual, summary);
if (result.offScreen && review.status === "modified" && visual.verdict === "unchanged") review = { ...review, offScreen: true };
if (review.facts?.length) accessibilityCounts.described += 1;
} catch (error) {
accessibilityCounts.failed += 1;
@@ -544,11 +750,13 @@ export async function assessPublicEvidence(options: {
const basePath = localEvidencePath(options.base, baseLaunch.splash.path); const headPath = localEvidencePath(options.head, headLaunch.splash.path);
if (!basePath || !headPath) continue;
try {
const { measurement, ignoredTop, ignoredBottom } = ignoreSystemBands(await measure(basePath, headPath), { top: statusBarBandPercent(options.platform), bottom: homeIndicatorBandPercent(options.platform) });
const raw = await measure(basePath, headPath);
const { measurement, ignoredStatusBar: ignoredTop, ignoredHomeIndicator: ignoredBottom } = ignoreSystemRegions(raw, system, systemBandFrameLoader(raw, threshold));
const visual = visualVerdict(assessVisualMeasurements({ baselineSamples: 1, currentSamples: 1, requiredSamples: 1,
baselineWithin: [], currentWithin: [], crossBuild: [measurement], representative: measurement }), threshold);
if (ignoredTop > 0) visual.ignoredStatusBarRegions = ignoredTop;
if (ignoredBottom > 0) visual.ignoredHomeIndicatorRegions = ignoredBottom;
if (ignoredTop > 0 || ignoredBottom > 0) visual.systemRegions = system.map((area) => ({ ...area }));
surfaces[id] = { visual };
} catch (error) {
comparisonError(id, error, { code: "VISUAL_COMPARISON_FAILED", stage: "compare",
@@ -556,6 +764,9 @@ export async function assessPublicEvidence(options: {
}
}
const sessionFailures = (["base", "head"] as const).flatMap(side => (options[side].bundle.sessionFailures ?? []).map(({ scheme, error }): PublicSessionFailure => ({
side, scheme, frames: Object.keys(options[side].bundle.outcomes).filter(id => id.endsWith(`@${scheme}`) && failedPassOutcome(options[side].bundle, id)
&& options[side].bundle.outcomes[id]!.error?.message === error.message).length, ...error })));
return {
surfaces,
summary: {
@@ -563,6 +774,7 @@ export async function assessPublicEvidence(options: {
accessibility: accessibilityAvailable ? accessibilityCounts : null,
review: reviewCounts,
capture, comparison, diagnostics,
...(sessionFailures.length ? { sessionFailures } : {}),
},
};
}
+166 -8
View File
@@ -17,7 +17,7 @@ import type {
ObservedElement,
Screen,
} from "../types.js";
import type { VisualEvidenceAssessment } from "./visual-evidence.js";
import type { VisualDiffRegion, VisualEvidenceAssessment } from "./visual-evidence.js";
import { CAPTURE_OBSERVER_IDENTIFIER, parseCaptureDestination } from "../capture/observer.js";
export type StableReviewerElementAnnotation = {
@@ -70,6 +70,11 @@ export type CommonSurfaceComparatorInput = {
visualAssessment: VisualEvidenceAssessment;
/** Explicit stable runtime scope; semantic meaning is never inferred from diff pixels. */
visualScope?: VisualReviewerChange["scope"];
/**
* The screen's other appearances, captured and verified on both builds. A label that varies between each build's own
* captures, in the words that differ between the builds, is runtime state, and its difference is reported as such.
*/
siblings?: readonly { base: MatchedSurfaceSide; head: MatchedSurfaceSide }[];
};
export type EvidenceComparatorInput =
@@ -88,6 +93,8 @@ export type EvidenceComparatorResult = {
visualAssessment?: VisualEvidenceAssessment;
/** Feed directly to summarizeReviewerChanges. */
reviewerChanges: ReviewerChangeInput[];
/** The screenshots match, and every element the changes name lies at least partly past the edge of the screen. */
offScreen?: true;
};
type SupportedInput = Extract<ReviewerChangeInput, { assessment: "supported" }>;
@@ -139,6 +146,7 @@ const CHANGE_PRIORITY: Record<SupportedReviewerChange["kind"], number> = {
"element-added": 20,
"element-removed": 20,
"visible-text-changed": 30,
"unstable-text": 35,
"action-target-changed": 40,
"accessibility-label-changed": 50,
"accessibility-role-changed": 50,
@@ -611,6 +619,54 @@ function outermost(elements: readonly ObservedElement[]): ObservedElement[] {
});
}
/**
* The screen in its accessibility tree's units: the element at the origin with the screenshot's shape (either way round,
* since a frame turned to show an opened iPhone Duo upright keeps the tree on the panel's portrait axes), the largest if
* several. Both controllers report the app's root that way. Without the screenshot's size there is no telling which
* element is the screen.
*/
function viewportOf(side: ResolvedSide): ElementBounds | undefined {
const image = side.map.artifacts.find(artifact => artifact.id === side.observation.screenshotArtifactId);
if (!image?.width || !image.height) return undefined;
const { width, height } = image;
const shaped = (bounds: ElementBounds) => Math.abs(bounds.width * height - bounds.height * width) <= 0.01 * width * height
|| Math.abs(bounds.width * width - bounds.height * height) <= 0.01 * width * height;
return side.observation.elements.flatMap(element => boundsOf(element) ?? [])
.filter(bounds => Math.abs(bounds.x) < 1 && Math.abs(bounds.y) < 1 && bounds.width > 0 && bounds.height > 0 && shaped(bounds))
.sort((a, b) => b.width * b.height - a.width * a.height)[0];
}
/**
* Whether the screenshots cannot show the changes because the elements they name lie past the edge of the screen: each
* change names at least one element, and more than half of each named element's area is beyond the screen on its side.
* Both controllers leave an element wholly outside the screen out of the tree (the Argent bridge as agent-device does),
* so a change past the edge can only show as an element straddling it, as Tempo's footnote below the fold of its sheet
* on the closed iPhone Duo in brentvatne/euxy#172 (review run 39362642343) would. Mostly outside, not merely crossing
* an edge: a control nudged a point or two past the edge is still on the screen, and its change stays "accessibility
* only". A change whose elements have no bounds never qualifies.
*/
function changesOffScreen(facts: readonly SupportedInput[], base: ResolvedSide, head: ResolvedSide): boolean {
const changes = facts.filter(fact => !["unstable-text", "no-visible-change", "visual-changed"].includes(fact.change.kind));
if (!changes.length) return false;
const viewports = { base: viewportOf(base), head: viewportOf(head) };
const outside = (reference: string) => {
const [, side, id] = /^(base|head):element:(.+)$/.exec(reference) ?? [];
if (side !== "base" && side !== "head") return undefined;
const viewport = viewports[side];
const element = (side === "base" ? base : head).observation.elements.find(candidate => candidate.elementId === id);
const bounds = element && boundsOf(element);
if (!viewport || !bounds) return false;
const visibleWidth = Math.max(0, Math.min(bounds.x + bounds.width, viewport.x + viewport.width) - Math.max(bounds.x, viewport.x));
const visibleHeight = Math.max(0, Math.min(bounds.y + bounds.height, viewport.y + viewport.height) - Math.max(bounds.y, viewport.y));
const area = bounds.width * bounds.height;
return area > 0 && visibleWidth * visibleHeight < area / 2;
};
return changes.every(fact => {
const named = fact.evidenceRefs.map(outside).filter((value): value is boolean => value !== undefined);
return named.length > 0 && named.every(Boolean);
});
}
function specificByLabel(elements: readonly ObservedElement[]): ObservedElement[] {
const byLabel = new Map<string, ObservedElement>();
const unlabelled: ObservedElement[] = [];
@@ -772,10 +828,76 @@ function actionTarget(side: ResolvedSide, element: ObservedElement): ActionTarge
return target ? { kind: "target", target } : { kind: "none", evidenceRefs: [] };
}
type ResolvedSibling = { base: ResolvedSide; head: ResolvedSide };
function resolveSiblings(input: CommonSurfaceComparatorInput, requiredSamples: 1 | 3): ResolvedSibling[] {
const resolve = (side: "base" | "head", sibling: MatchedSurfaceSide) => {
const screenId = sibling.map.observations.find((candidate) => candidate.id === sibling.observationId)?.screenId;
const resolution = screenId === undefined ? undefined : resolveSide(side, sibling, { ...input.impact, id: screenId }, requiredSamples);
// A sibling is evidence only when it resolves as cleanly as a comparison would.
return resolution?.kind === "resolved" && resolution.side.issues.length === 0 ? resolution.side : undefined;
};
return (input.siblings ?? []).flatMap((sibling) => {
const base = resolve("base", sibling.base);
const head = resolve("head", sibling.head);
return base && head ? [{ base, head }] : [];
});
}
function sameBounds(left: ElementBounds | undefined, right: ElementBounds | undefined): boolean {
return !!left && !!right && left.x === right.x && left.y === right.y && left.width === right.width && left.height === right.height;
}
/** The element at the same place in another capture: the one of its role with its identifier or, without one, its bounds. */
function counterpart(element: ObservedElement, other: ResolvedSide): ObservedElement | undefined {
const identifier = compact(element.identifier ?? "");
const bounds = boundsOf(element);
const candidates = other.elements.filter((candidate) => candidate.role === element.role && (identifier
? compact(candidate.identifier ?? "") === identifier
: !compact(candidate.identifier ?? "") && sameBounds(bounds, boundsOf(candidate))));
return candidates.length === 1 ? candidates[0] : undefined;
}
/** The word positions at which two labels differ, or undefined when they have different numbers of words. */
function changedWords(left: string, right: string): number[] | undefined {
const leftWords = left.split(" ");
const rightWords = right.split(" ");
return leftWords.length === rightWords.length
? leftWords.flatMap((word, index) => word === rightWords[index] ? [] : [index])
: undefined;
}
/**
* What the head showed in its other appearance, when a label's difference between the builds is runtime state: each
* build shows the label differently in its own two appearances, the other appearance differs between the builds too,
* and every word that differs between the builds is one that varies within a build. A label that follows the
* appearance ("Switch to dark mode") does not vary within a build that broke it, or keeps a word that changed, so
* its change is still a change.
*/
function unstableWithin(base: ObservedElement, head: ObservedElement, siblings: readonly ResolvedSibling[]): string | undefined {
const baseLabel = compact(base.label);
const headLabel = compact(head.label);
for (const sibling of siblings) {
const baseOther = compact(counterpart(base, sibling.base)?.label ?? "");
const headOther = compact(counterpart(head, sibling.head)?.label ?? "");
if (!baseLabel || !headLabel || !baseOther || !headOther) continue;
const withinBase = changedWords(baseLabel, baseOther);
const withinHead = changedWords(headLabel, headOther);
const across = changedWords(baseLabel, headLabel);
const acrossOther = changedWords(baseOther, headOther);
if (!withinBase?.length || !withinHead?.length || !across?.length || !acrossOther?.length) continue;
const varying = new Set([...withinBase, ...withinHead]);
if ([...across, ...acrossOther].every((index) => varying.has(index))) return headOther;
}
return undefined;
}
function comparePair(
pair: ElementPair,
base: ResolvedSide,
head: ResolvedSide,
/** What the head showed in its other appearance when this element's label is runtime state. */
within?: string,
): { facts: SupportedInput[]; issues: IssueInput[] } {
const facts: SupportedInput[] = [];
const issues: IssueInput[] = [];
@@ -784,6 +906,10 @@ function comparePair(
const headLabel = compact(pair.head.label);
const baseAnnotation = base.annotationsByElementId.get(pair.base.elementId);
const headAnnotation = head.annotationsByElementId.get(pair.head.elementId);
// Runtime state replaces a label change only where the label comparison itself holds: provenance still decides first.
const labelChange = (change: SupportedReviewerChange): SupportedReviewerChange => within
? { kind: "unstable-text", subjectRole: reviewerRole(pair.head.role), before: baseLabel, after: headLabel, within }
: change;
if (baseLabel !== headLabel) {
if (baseAnnotation && headAnnotation) {
@@ -793,14 +919,14 @@ function comparePair(
} else if (!baseLabel || !headLabel) {
issues.push(inconclusive("Visible-title comparison is missing before or after text.", refs));
} else {
facts.push(supported({
facts.push(supported(labelChange({
kind: "visible-text-changed",
subjectRole: subject.subjectRole,
subjectLabel: subject.subjectLabel,
textKind: pair.head.role === "button" ? "title" : "label",
before: baseLabel,
after: headLabel,
}, refs));
}), refs));
}
} else if (baseAnnotation || headAnnotation) {
issues.push(inconclusive(
@@ -810,12 +936,12 @@ function comparePair(
} else if (!baseLabel || !headLabel) {
issues.push(inconclusive("Accessibility label comparison is missing before or after text.", refs));
} else {
facts.push(supported({
facts.push(supported(labelChange({
kind: "accessibility-label-changed",
subjectRole: reviewerRole(pair.head.role),
before: baseLabel,
after: headLabel,
}, refs));
}), refs));
}
}
if (pair.base.role !== pair.head.role) {
@@ -925,11 +1051,20 @@ function visualFact(
fact: supported({ kind: "visual-changed", scope: { kind: "landmark", label } }, refs),
};
}
const near = nearEdge(input.visualAssessment.representative?.regions ?? []);
return {
fact: supported({ kind: "visual-changed", scope: { kind: "main-content" } }, refs),
fact: supported({ kind: "visual-changed", scope: { kind: "main-content", ...(near ? { near } : {}) } }, refs),
};
}
/** Which edge every diff region hugs, when all of them sit in the outer 12 % of the frame: not "the main content". */
function nearEdge(regions: readonly VisualDiffRegion[]): "top" | "bottom" | undefined {
if (regions.length === 0) return undefined;
if (regions.every((region) => region.normalizedRect.y + region.normalizedRect.height <= 12)) return "top";
if (regions.every((region) => region.normalizedRect.y >= 88)) return "bottom";
return undefined;
}
function comparisonForAdded(head: RuntimeEvidence): RuntimeEvidenceComparison {
return {
contract: RUNTIME_EVIDENCE_COMPARISON_CONTRACT,
@@ -1101,6 +1236,27 @@ function commonResult(input: CommonSurfaceComparatorInput): EvidenceComparatorRe
gone = base.elements.filter(element => content(element, headWords));
arrived = head.elements.filter(element => content(element, baseWords));
}
// Text without an identifier is matched by its label, so a label that changed on its own leaves as one element and
// arrives as another: the one of its role at the same place on both sides is runtime state when its label is.
const siblings = mixedBackends ? [] : resolveSiblings(input, input.impact.baseSamples);
const runtimeText = { base: new Set<string>(), head: new Set<string>() };
for (const removed of siblings.length ? specificByLabel(gone) : []) {
const added = counterpart(removed, head);
if (!added || !arrived.includes(added) || counterpart(added, base) !== removed) continue;
const within = unstableWithin(removed, added, siblings);
if (!within) continue;
runtimeText.base.add(compact(removed.label));
runtimeText.head.add(compact(added.label));
facts.push(supported({
kind: "unstable-text",
subjectRole: reviewerRole(added.role),
before: compact(removed.label),
after: compact(added.label),
within,
}, [elementRef("base", removed), elementRef("head", added)]));
}
gone = gone.filter((element) => !runtimeText.base.has(compact(element.label)));
arrived = arrived.filter((element) => !runtimeText.head.has(compact(element.label)));
for (const element of outermost(specificByLabel(gone))) {
const label = compact(element.label);
if (!label) {
@@ -1132,7 +1288,7 @@ function commonResult(input: CommonSurfaceComparatorInput): EvidenceComparatorRe
}
}
for (const pair of pairing.pairs) {
const comparison = comparePair(pair, base, head);
const comparison = comparePair(pair, base, head, unstableWithin(pair.base, pair.head, siblings));
facts.push(...comparison.facts);
issues.push(...comparison.issues);
}
@@ -1145,7 +1301,7 @@ function commonResult(input: CommonSurfaceComparatorInput): EvidenceComparatorRe
const visual = visualFact(input, base, head);
// Pixels always move when an element appears or its text changes. The generic styling line
// only informs when nothing in the accessibility tree accounts for the difference.
const explained = facts.length > 0 && input.visualScope?.kind !== "landmark";
const explained = facts.some((fact) => fact.change.kind !== "unstable-text") && input.visualScope?.kind !== "landmark";
if (visual.fact && !explained) facts.push(visual.fact);
if (visual.issue) issues.push(visual.issue);
@@ -1170,6 +1326,7 @@ function commonResult(input: CommonSurfaceComparatorInput): EvidenceComparatorRe
const status: "modified" | "unchanged" = hasRuntimeDelta || reasons.length > 0
? "modified"
: "unchanged";
const offScreen = input.visualAssessment.status === "unchanged" && changesOffScreen(facts, base, head);
return {
surfaceId: input.impact.id,
status,
@@ -1182,6 +1339,7 @@ function commonResult(input: CommonSurfaceComparatorInput): EvidenceComparatorRe
),
visualAssessment: input.visualAssessment,
reviewerChanges,
...(offScreen ? { offScreen: true as const } : {}),
};
}
+33
View File
@@ -0,0 +1,33 @@
import { Worker, isMainThread, parentPort, workerData } from "node:worker_threads";
import { measureRegionMoves, type Rect } from "./region-shift.js";
type MeasuredRegions = Awaited<ReturnType<typeof measureRegionMoves>>;
type RegionMoveWork = { regionMoveWork: true; basePath: string; headPath: string; regions: ReadonlyArray<Rect & { dominantChange?: string }> };
type RegionMoveReply = { measured: MeasuredRegions } | { error: string };
/**
* Measures region shifts and large moves on a worker thread, so several screenshot pairs search at once while the
* main thread stays free: the search is CPU-bound and takes seconds on a tablet-sized screenshot. Falls back to the
* calling thread when no worker can start.
*/
export async function measureRegionMovesInWorker(basePath: string, headPath: string, regions: RegionMoveWork["regions"]): Promise<MeasuredRegions> {
const work: RegionMoveWork = { regionMoveWork: true, basePath, headPath, regions };
let worker: Worker;
try { worker = new Worker(new URL("./region-shift-worker.js", import.meta.url), { workerData: work }); }
catch { return measureRegionMoves(basePath, headPath, regions); }
return new Promise((resolve, reject) => {
let settled = false;
const finish = (outcome: () => void) => { if (!settled) { settled = true; outcome(); } };
worker.once("message", (reply: RegionMoveReply) => finish(() => "error" in reply ? reject(new Error(reply.error)) : resolve(reply.measured)));
// A worker that fails to load or dies before answering is a worker that did not start: the calling thread measures.
worker.once("error", () => finish(() => resolve(measureRegionMoves(basePath, headPath, regions))));
worker.once("exit", () => finish(() => resolve(measureRegionMoves(basePath, headPath, regions))));
});
}
if (!isMainThread && parentPort && (workerData as Partial<RegionMoveWork> | null)?.regionMoveWork === true) {
const port = parentPort;
const work = workerData as RegionMoveWork;
measureRegionMoves(work.basePath, work.headPath, work.regions)
.then(measured => port.postMessage({ measured } satisfies RegionMoveReply), (error: unknown) => port.postMessage({ error: (error as Error).message } satisfies RegionMoveReply));
}
+304 -8
View File
@@ -4,7 +4,7 @@ import type { DecodedFrame } from "../shared/png-evidence.js";
/** How far a region's content moved from Before to After, in screenshot pixels (right and down are positive). */
export type RegionShift = { x: number; y: number };
type Rect = { x: number; y: number; width: number; height: number };
export type Rect = { x: number; y: number; width: number; height: number };
/**
* The farthest a region is searched for, in screenshot pixels: layout drift, not a different screen. Drift adds up
@@ -63,13 +63,38 @@ function at(source: Image, x: number, y: number, channel?: number): number {
return value(x0, y0) * (1 - fx) * (1 - fy) + value(x1, y0) * fx * (1 - fy) + value(x0, y1) * (1 - fx) * fy + value(x1, y1) * fx * fy;
}
/** How badly After's region matches Before shifted by (sx, sy). */
function mismatch(base: Image, head: Image, rect: Rect, sx: number, sy: number): number {
/**
* How badly After's region matches Before shifted by (sx, sy): the summed luminance difference, row by row. A sum
* that passes `limit` stops early and returns a value above it: such a fit is neither the best, nor a tie, nor an
* ambiguity, so its exact error is never read. Whole-pixel shifts read the pixel directly (the bilinear weights are
* exactly 1 and 0 there); fractional ones interpolate like `at`.
*/
function mismatch(base: Image, head: Image, rect: Rect, sx: number, sy: number, limit = Infinity): number {
const { width, height, luminance } = base;
const maxX = width - 1, maxY = height - 1;
const headLuminance = head.luminance, headWidth = head.width;
const whole = Number.isInteger(sx) && Number.isInteger(sy);
let error = 0;
for (let y = rect.y; y < rect.y + rect.height; y += 1) {
for (let x = rect.x; x < rect.x + rect.width; x += 1) {
error += Math.abs(at(base, x - sx, y - sy) - head.luminance[y * head.width + x]!);
const headRow = y * headWidth;
if (whole) {
const row = Math.min(maxY, Math.max(0, y - sy)) * width;
for (let x = rect.x; x < rect.x + rect.width; x += 1) {
error += Math.abs(luminance[row + Math.min(maxX, Math.max(0, x - sx))]! - headLuminance[headRow + x]!);
}
} else {
const cy = Math.min(maxY, Math.max(0, y - sy));
const y0 = Math.floor(cy), y1 = Math.min(maxY, y0 + 1), fy = cy - y0;
const row0 = y0 * width, row1 = y1 * width;
for (let x = rect.x; x < rect.x + rect.width; x += 1) {
const cx = Math.min(maxX, Math.max(0, x - sx));
const x0 = Math.floor(cx), x1 = Math.min(maxX, x0 + 1), fx = cx - x0;
const sample = luminance[row0 + x0]! * (1 - fx) * (1 - fy) + luminance[row0 + x1]! * fx * (1 - fy)
+ luminance[row1 + x0]! * (1 - fx) * fy + luminance[row1 + x1]! * fx * fy;
error += Math.abs(sample - headLuminance[headRow + x]!);
}
}
if (error > limit) return error;
}
return error;
}
@@ -110,18 +135,29 @@ function regionShift(base: Image, head: Image, region: Rect): RegionShift | unde
const unshifted = mismatch(base, head, rect, 0, 0);
if (unshifted === 0) return undefined;
// Whole pixels first, every pixel counting: sampling every other row aliases on thin shapes (text, hairlines).
// A fit worse than the best so far by more than the ambiguity margin cannot become the best, tie with it, or make
// it ambiguous; its measurement stops as soon as that is known. The margin is taken from the worst fit the tie
// could choose, since a tie goes to the smallest move, not the least error.
const whole: Fit[] = [];
let least = unshifted;
for (let sy = -MAX_SHIFT_Y; sy <= MAX_SHIFT_Y; sy += 1) {
for (let sx = -MAX_SHIFT_X; sx <= MAX_SHIFT_X; sx += 1) whole.push({ sx, sy, error: mismatch(base, head, rect, sx, sy) });
for (let sx = -MAX_SHIFT_X; sx <= MAX_SHIFT_X; sx += 1) {
const error = sx === 0 && sy === 0 ? unshifted : mismatch(base, head, rect, sx, sy, (least * TIE + 1e-6) * AMBIGUOUS + 1e-6);
least = Math.min(least, error);
whole.push({ sx, sy, error });
}
}
const coarse = bestOf(whole);
if (whole.some(fit => Math.abs(fit.sy - coarse.sy) >= 2 && fit.error <= coarse.error * AMBIGUOUS + 1e-6)) return undefined;
// Then sixths of a pixel around it, from whole-number steps so a move up and a move down round alike.
const fine: Fit[] = [];
least = coarse.error;
for (let iy = -STEPS_PER_PIXEL; iy <= STEPS_PER_PIXEL; iy += 1) {
for (let ix = -STEPS_PER_PIXEL; ix <= STEPS_PER_PIXEL; ix += 1) {
const sx = coarse.sx + ix / STEPS_PER_PIXEL, sy = coarse.sy + iy / STEPS_PER_PIXEL;
fine.push({ sx, sy, error: ix === 0 && iy === 0 ? coarse.error : mismatch(base, head, rect, sx, sy) });
const error = ix === 0 && iy === 0 ? coarse.error : mismatch(base, head, rect, sx, sy, least * TIE + 1e-6);
least = Math.min(least, error);
fine.push({ sx, sy, error });
}
}
const best = bestOf(fine);
@@ -141,10 +177,270 @@ export function regionShifts(baseFrame: DecodedFrame, headFrame: DecodedFrame, r
return regions.map(region => regionShift(base, head, region));
}
/**
* A move too large for the drift search: content that left one area and arrived in another, such as a card that
* went from the left of a wide sheet to its centre. `rect` is where the moved content is in After, `shift` how far it
* went, and `members` the regions it explains (indices into the regions measured), the region it arrived in first.
*/
export type RegionMove = { rect: Rect; shift: RegionShift; members: number[] };
type Axis = "x" | "y";
/** How alike two regions' extents across the move must be to be one element's two ends: overlap and size. */
const PAIR_OVERLAP = 0.8;
const PAIR_SIZE = 0.25;
/** Rows (or columns) sampled for the brightness profiles that estimate a large move: every second one is plenty. */
const PROFILE_STEP = 2;
/** Another local best this close to the best makes a large move ambiguous: repeating content (a grid, a list). */
const AMBIGUOUS_LARGE = 1.05;
/** The share of the arrived region the move must explain, as for a small move. */
const MIN_EXPLAINED_LARGE = MIN_EXPLAINED;
/** After aligning by the move, at most this share of a region may still differ strongly (antialiased edges). */
const CHANGED_SHARE = 0.001;
/** How far across the move a large one may also drift, as layout below a moved element does; ties keep it in line. */
const ACROSS_OFFSETS = [0, -1, 1, -2, 2, -3, 3];
/**
* Blocks that give something drawn away, where every pixel differs by more than `channel` in some channel: a 2×2
* block of strongly different pixels (an icon, a badge, a mark), or a 4×4 block of moderately different ones (a
* tint, a low-contrast label, a grey badge). Antialiased edges and hairlines against another backdrop are thinner:
* the chevrons in euxy's Patterns rows (brentvatne/euxy#163) cross a curve that differs by up to 69 in 3×3 blocks
* but only 26 in 4×4 ones, while a grey "Off" label the move did not bring differs by 55 in 4×4 blocks.
*/
const SOLID_BLOCKS = [{ size: 2, channel: CHANGED_CHANNEL }, { size: 4, channel: 40 }];
const extent = (rect: Rect, axis: Axis) => axis === "x" ? { start: rect.x, size: rect.width } : { start: rect.y, size: rect.height };
const across = (axis: Axis): Axis => axis === "x" ? "y" : "x";
const shiftAlong = (axis: Axis, along: number, other = 0): RegionShift => axis === "x" ? { x: along, y: other } : { x: other, y: along };
/** Whether two regions could be one element's two ends of a move along `axis`: side by side, alike across it. */
function pairAlong(a: Rect, b: Rect, axis: Axis): boolean {
const ea = extent(a, axis), eb = extent(b, axis), ca = extent(a, across(axis)), cb = extent(b, across(axis));
const separate = ea.start + ea.size <= eb.start || eb.start + eb.size <= ea.start;
const overlap = Math.min(ca.start + ca.size, cb.start + cb.size) - Math.max(ca.start, cb.start);
const alike = Math.abs(ea.size - eb.size) <= PAIR_SIZE * Math.max(ea.size, eb.size);
return separate && alike && overlap >= PAIR_OVERLAP * Math.min(ca.size, cb.size);
}
/** Mean luminance along `axis` of a band across it: one value per column (axis x) or row (axis y). */
function profile(source: Image, axis: Axis, bandStart: number, bandEnd: number): Float32Array {
const length = axis === "x" ? source.width : source.height;
const values = new Float32Array(length);
for (let position = 0; position < length; position += 1) {
let sum = 0, count = 0;
for (let other = bandStart; other < bandEnd; other += PROFILE_STEP) {
sum += axis === "x" ? source.luminance[other * source.width + position]! : source.luminance[position * source.width + other]!;
count += 1;
}
values[position] = count ? sum / count : 0;
}
return values;
}
/**
* Whether a region holds something drawn: a solid block (SOLID_BLOCKS), or more than `allowed` strongly different
* pixels anywhere. `difference` is how much the pixel at (x, y) differs, in its most different channel.
*/
function drawnIn(rect: Rect, difference: (x: number, y: number) => number, allowed = Infinity): boolean {
// Per block: how many pixels differ in a row up to here, and for each column, how many rows have such a run.
const blocks = SOLID_BLOCKS.map(block => ({ ...block, run: 0, rows: new Uint32Array(rect.width) }));
let strong = 0;
for (let y = rect.y; y < rect.y + rect.height; y += 1) {
for (const block of blocks) block.run = 0;
for (let x = rect.x; x < rect.x + rect.width; x += 1) {
const value = difference(x, y), column = x - rect.x;
if (value > CHANGED_CHANNEL && ++strong > allowed) return true;
for (const block of blocks) {
block.run = value > block.channel ? block.run + 1 : 0;
block.rows[column] = block.run >= block.size ? block.rows[column]! + 1 : 0;
if (block.rows[column]! >= block.size) return true;
}
}
}
return false;
}
/**
* Whether, aligned by an integer move, some of the region still differs: content that changed, not only moved. A
* whole-pixel move copies the pixels, so only a sliver may differ strongly (antialiased edges against another
* backdrop, never more than CHANGED_SHARE of the region), and never a solid block. Pixels the move brings in from
* off screen differ.
*/
function changedAfterLargeMove(base: Image, head: Image, rect: Rect, sx: number, sy: number): boolean {
return drawnIn(rect, (x, y) => {
const bx = x - sx, by = y - sy;
if (bx < 0 || by < 0 || bx >= base.width || by >= base.height) return 255;
const h = (y * head.width + x) * 4, b = (by * base.width + bx) * 4;
return Math.max(Math.abs(base.frame.data[b]! - head.frame.data[h]!), Math.abs(base.frame.data[b + 1]! - head.frame.data[h + 1]!),
Math.abs(base.frame.data[b + 2]! - head.frame.data[h + 2]!));
}, Math.max(CHANGED_PIXELS, rect.width * rect.height * CHANGED_SHARE));
}
/** How much of a region's difference an integer move explains: 1 is all of it, 0 none. `unshifted` is the region's own. */
function explained(base: Image, head: Image, rect: Rect, shift: RegionShift, unshifted = mismatch(base, head, rect, 0, 0)): number {
return unshifted === 0 ? 0 : 1 - mismatch(base, head, rect, shift.x, shift.y) / unshifted;
}
/** Whether a region's content, moved by `shift`, is the same in After: explained, and nothing changed in it. */
function movedIntact(base: Image, head: Image, rect: Rect, shift: RegionShift): boolean {
return rect.width > 0 && rect.height > 0 && explained(base, head, rect, shift) >= MIN_EXPLAINED_LARGE
&& !changedAfterLargeMove(base, head, rect, shift.x, shift.y);
}
/**
* Whether a part of After where content left is plain backdrop: the colour around it (the median of the pixels just
* outside it, not in the moved content's `outline`), with nothing drawn on it. A part with nothing around it is not.
*/
function plain(head: Image, rect: Rect, outline: Rect): boolean {
if (rect.width <= 0 || rect.height <= 0) return true;
const around: number[][] = [[], [], []];
const add = (x: number, y: number) => {
if (x < 0 || y < 0 || x >= head.width || y >= head.height) return;
if (x >= outline.x && y >= outline.y && x < outline.x + outline.width && y < outline.y + outline.height) return;
for (let channel = 0; channel < 3; channel += 1) around[channel]!.push(head.frame.data[(y * head.width + x) * 4 + channel]!);
};
for (let x = rect.x - 1; x <= rect.x + rect.width; x += 1) { add(x, rect.y - 1); add(x, rect.y + rect.height); }
for (let y = rect.y; y < rect.y + rect.height; y += 1) { add(rect.x - 1, y); add(rect.x + rect.width, y); }
if (!around[0]!.length) return false;
const backdrop = around.map(values => values.sort((a, b) => a - b)[values.length >> 1]!);
return !drawnIn(rect, (x, y) => {
const offset = (y * head.width + x) * 4;
return Math.max(Math.abs(head.frame.data[offset]! - backdrop[0]!), Math.abs(head.frame.data[offset + 1]! - backdrop[1]!),
Math.abs(head.frame.data[offset + 2]! - backdrop[2]!));
});
}
/** The parts of `rect` before and after `outline` along `axis`: what is left of it outside the outline. */
function outside(rect: Rect, outline: Rect, axis: Axis): Rect[] {
const r = extent(rect, axis), o = extent(outline, axis);
const parts: Array<[number, number]> = [[r.start, Math.min(r.start + r.size, o.start)], [Math.max(r.start, o.start + o.size), r.start + r.size]];
return parts.filter(([from, to]) => to > from)
.map(([from, to]) => (axis === "x" ? { ...rect, x: from, width: to - from } : { ...rect, y: from, height: to - from }));
}
function clip(rect: Rect, width: number, height: number): Rect {
const x = Math.max(0, Math.floor(rect.x)), y = Math.max(0, Math.floor(rect.y));
return { x, y, width: Math.max(0, Math.min(width, Math.ceil(rect.x + rect.width)) - x), height: Math.max(0, Math.min(height, Math.ceil(rect.y + rect.height)) - y) };
}
function union(a: Rect, b: Rect): Rect {
const x = Math.min(a.x, b.x), y = Math.min(a.y, b.y);
return { x, y, width: Math.max(a.x + a.width, b.x + b.width) - x, height: Math.max(a.y + a.height, b.y + b.height) - y };
}
/**
* The large move a pair of regions shows, if one does: `left` is where content was (it differs because the content
* left), `arrived` where it is now, both already clipped to the screenshot. The pair only names the axis and how far
* to look; the distance is where After's brightness profile across the arrived region best matches Before's, and
* the pixels must agree once aligned, both where the content arrived and where the content that left went.
*/
function pairMove(base: Image, head: Image, left: Rect, arrived: Rect, axis: Axis): RegionShift | undefined {
const a = extent(arrived, axis), l = extent(left, axis), band = extent(arrived, across(axis)), lb = extent(left, across(axis));
const bandStart = Math.max(band.start, lb.start), bandEnd = Math.min(band.start + band.size, lb.start + lb.size);
if (bandEnd <= bandStart || a.size <= 0) return undefined;
const baseProfile = profile(base, axis, bandStart, bandEnd), headProfile = profile(head, axis, bandStart, bandEnd);
const length = baseProfile.length;
// Content moving toward the arrived region: the distance is positive when it lies after the region it left.
const sign = a.start > l.start ? 1 : -1;
const farthest = Math.abs(a.start + a.size - l.start) + Math.abs(l.start + l.size - a.start);
const errors: Array<{ distance: number; error: number }> = [];
for (let distance = (axis === "x" ? MAX_SHIFT_X : MAX_SHIFT_Y) + 1; distance <= farthest; distance += 1) {
const move = sign * distance;
let sum = 0, count = 0;
for (let position = a.start; position < a.start + a.size; position += 1) {
const from = position - move;
if (from < 0 || from >= length) continue;
sum += Math.abs(headProfile[position]! - baseProfile[from]!);
count += 1;
}
if (count >= a.size / 2) errors.push({ distance: move, error: sum / count });
}
if (errors.length < 3) return undefined;
const minima = errors.filter((fit, index) => index > 0 && index < errors.length - 1
&& fit.error <= errors[index - 1]!.error && fit.error <= errors[index + 1]!.error);
if (!minima.length) return undefined;
const best = minima.reduce((least, fit) => (fit.error < least.error ? fit : least));
if (minima.some(fit => fit !== best && Math.abs(fit.distance - best.distance) > 2 && fit.error <= best.error * AMBIGUOUS_LARGE + 1e-6)) return undefined;
// The pixels decide: across the move it may also drift a little, as content below a moved element does. A tie
// keeps the move in line (content the same all across it fits every drift equally).
let chosen: RegionShift | undefined, fit = -Infinity;
const unshifted = mismatch(base, head, arrived, 0, 0);
for (const other of ACROSS_OFFSETS) {
const shift = shiftAlong(axis, best.distance, other);
const share = explained(base, head, arrived, shift, unshifted);
if (share > fit) { fit = share; chosen = shift; }
}
if (!chosen || fit < MIN_EXPLAINED_LARGE || changedAfterLargeMove(base, head, arrived, chosen.x, chosen.y)) return undefined;
// The content that left must be intact where the move takes it: a deletion is not a move.
const landed = clip({ ...left, x: left.x + chosen.x, y: left.y + chosen.y }, head.width, head.height);
return movedIntact(base, head, landed, chosen) ? chosen : undefined;
}
/**
* Moves too large for the drift search, among the regions it left unexplained: two regions alike across an axis
* and side by side along it, one darker and one brighter, are the two ends of something that moved. Where the
* content left must be plain backdrop in After for that region to be part of the move; anything drawn there is a
* change of its own and stays one. A move that checks out also takes in any other unexplained region it explains
* (fragments of the same element where its old and new places overlap), and its outline is the moved content in After.
*/
export function regionMoves(baseFrame: DecodedFrame, headFrame: DecodedFrame,
regions: ReadonlyArray<Rect & { dominantChange?: string }>, shifts: ReadonlyArray<RegionShift | undefined>): RegionMove[] {
if (baseFrame.width !== headFrame.width || baseFrame.height !== headFrame.height) return [];
const base = image(baseFrame), head = image(headFrame);
const clipped = regions.map(region => clip(region, head.width, head.height));
const open = new Set(regions.map((_, index) => index).filter(index => !shifts[index] && clipped[index]!.width > 0 && clipped[index]!.height > 0));
const moves: RegionMove[] = [];
// Largest first: the element that moved the most content is the one a reader is looking at.
const order = [...open].sort((left, right) => clipped[right]!.width * clipped[right]!.height - clipped[left]!.width * clipped[left]!.height);
for (let first = 0; first < order.length; first += 1) {
for (let second = first + 1; second < order.length; second += 1) {
const i = order[first]!, j = order[second]!;
if (!open.has(i) || !open.has(j)) continue;
const opposite = new Set([regions[i]!.dominantChange, regions[j]!.dominantChange]);
if (!(opposite.has("brighter") && opposite.has("darker"))) continue;
const a = clipped[i]!, b = clipped[j]!;
for (const axis of ["x", "y"] as const) {
if (!pairAlong(a, b, axis)) continue;
// Either could be where the content arrived; the pixels decide which.
const forward = pairMove(base, head, a, b, axis);
const shift = forward ?? pairMove(base, head, b, a, axis);
if (!shift) continue;
const [leftIndex, arrivedIndex] = forward ? [i, j] : [j, i];
const left = clipped[leftIndex]!, arrived = clipped[arrivedIndex]!;
let rect = clip(union(arrived, { ...left, x: left.x + shift.x, y: left.y + shift.y }), head.width, head.height);
open.delete(i); open.delete(j);
// Where the content left: plain backdrop belongs to the move, anything drawn there is a change of its own.
const members = outside(left, rect, axis).every(part => plain(head, part, rect)) ? [arrivedIndex, leftIndex] : [arrivedIndex];
// Fragments of the same element inside its new place, where the old and new overlap.
for (const k of [...open]) {
const fragment = clipped[k]!;
const inside = fragment.x >= rect.x && fragment.y >= rect.y && fragment.x + fragment.width <= rect.x + rect.width && fragment.y + fragment.height <= rect.y + rect.height;
if (!inside || !movedIntact(base, head, fragment, shift)) continue;
members.push(k); open.delete(k); rect = union(rect, fragment);
}
moves.push({ rect, shift, members });
break;
}
}
}
return moves;
}
/** Reads both PNG screenshots and measures each region's shift. */
export async function measureRegionShifts(basePath: string, headPath: string, regions: readonly Rect[]): Promise<Array<RegionShift | undefined>> {
const [base, head] = await readFrames(basePath, headPath);
return regionShifts(base, head, regions);
}
/** Reads both PNG screenshots once and measures each region's shift, then the large moves the rest show. */
export async function measureRegionMoves(basePath: string, headPath: string, regions: ReadonlyArray<Rect & { dominantChange?: string }>):
Promise<{ shifts: Array<RegionShift | undefined>; moves: RegionMove[]; width: number; height: number }> {
const [base, head] = await readFrames(basePath, headPath);
const shifts = regionShifts(base, head, regions);
return { shifts, moves: regionMoves(base, head, regions, shifts), width: head.width, height: head.height };
}
async function readFrames(basePath: string, headPath: string): Promise<[DecodedFrame, DecodedFrame]> {
const { PNG } = createRequire(import.meta.url.replace(/\?.*$/, ""))("pngjs") as
{ PNG: { sync: { read(bytes: Buffer, options: { checkCRC: boolean }): DecodedFrame } } };
const [base, head] = await Promise.all([readFile(basePath), readFile(headPath)]);
return regionShifts(PNG.sync.read(base, { checkCRC: false }), PNG.sync.read(head, { checkCRC: false }), regions);
return [PNG.sync.read(base, { checkCRC: false }), PNG.sync.read(head, { checkCRC: false })];
}
+39 -2
View File
@@ -28,6 +28,7 @@ export type ReviewerChangeTemplateId =
| "element.removed"
| "visible-text.changed-short"
| "visible-text.changed-long"
| "visible-text.unstable-within-build"
| "interaction.target-changed"
| "accessibility.label-changed"
| "accessibility.role-changed"
@@ -95,6 +96,19 @@ export type VisibleTextReviewerChange = {
after: string;
};
/**
* A label that differs between the two builds but also between two captures of the same build (its light and dark
* appearances): a clock, a relative time, a random name. Runtime state, not a change.
*/
export type UnstableTextReviewerChange = {
kind: "unstable-text";
subjectRole: ReviewerSubjectRole;
before: string;
after: string;
/** What the same build showed in its other capture. */
within: string;
};
export type InteractionReviewerChange = {
kind: "action-target-changed";
subjectRole: ReviewerSubjectRole;
@@ -121,7 +135,7 @@ export type VisualReviewerChange = {
kind: "visual-changed";
scope:
| { kind: "landmark"; label: string }
| { kind: "main-content" };
| { kind: "main-content"; near?: "top" | "bottom" };
};
/**
@@ -134,6 +148,7 @@ export type SupportedReviewerChange =
| ScreenReviewerChange
| ElementReviewerChange
| VisibleTextReviewerChange
| UnstableTextReviewerChange
| InteractionReviewerChange
| AccessibilityReviewerChange
| VisualReviewerChange
@@ -387,6 +402,28 @@ function buildFact(change: SupportedReviewerChange, evidenceRefs: string[]): Can
},
};
}
case "unstable-text": {
const before = compact(change.before);
const after = compact(change.after);
const within = compact(change.within);
if (!before || !after || !within) return invalidChange("An unstable-text change requires all three labels.", evidenceRefs);
if (before === after) return invalidChange("An unstable-text change has identical labels.", evidenceRefs);
const seen = [...new Set([before, within, after])];
return {
kind: "fact",
fact: {
category: "visible-text",
templateId: "visible-text.unstable-within-build",
bullet: `Text varies between captures of the same build: ${seen.map(quote).join(" / ")}.`,
subjectRole: change.subjectRole,
subjectLabel: after,
before,
after,
priority: PRIORITY["visible-text"],
evidenceRefs,
},
};
}
case "accessibility-label-changed": {
const before = compact(change.before);
const after = compact(change.after);
@@ -466,7 +503,7 @@ function buildFact(change: SupportedReviewerChange, evidenceRefs: string[]): Can
fact: {
category: "visual",
templateId: "visual.main-content-changed",
bullet: "Updated visual styling in the main content area.",
bullet: change.scope.near ? `Updated visual styling near the ${change.scope.near} of the screen.` : "Updated visual styling in the main content area.",
subjectRole: "content",
subjectLabel: "Main content",
priority: PRIORITY.visual,
+28 -24
View File
@@ -1,4 +1,4 @@
import { spawnSync } from "node:child_process";
import { execFile } from "node:child_process";
import { lstat, readFile, realpath } from "node:fs/promises";
import { isAbsolute } from "node:path";
@@ -133,30 +133,34 @@ export async function diffScreenshots(options: {
|| !isAbsolute(options.command)
|| /[\u0000-\u001f\u007f]/.test(options.command)
) throw new Error("agent-device command must be an absolute path");
const result = spawnSync(
options.command,
[
"--json",
"diff",
"screenshot",
"--baseline",
baselinePath,
currentPath,
"--threshold",
String(threshold),
],
{
encoding: "utf8",
shell: false,
timeout: options.timeoutMs ?? 60_000,
env: {
PATH: process.env.PATH,
const result = await new Promise<{ error: (Error & { code?: unknown }) | null; stdout: string; stderr: string }>(resolve => {
execFile(
options.command,
[
"--json",
"diff",
"screenshot",
"--baseline",
baselinePath,
currentPath,
"--threshold",
String(threshold),
],
{
encoding: "utf8",
shell: false,
timeout: options.timeoutMs ?? 60_000,
env: {
PATH: process.env.PATH,
},
maxBuffer: 4 * 1024 * 1024,
},
maxBuffer: 4 * 1024 * 1024,
},
);
if (result.error) throw new Error(`Could not run screenshot diff: ${result.error.message}`);
if (result.status !== 0) {
(error, stdout, stderr) => resolve({ error, stdout, stderr }),
);
});
// execFile reports a non-zero exit as an error carrying the exit status; a spawn failure or timeout carries none.
if (result.error && typeof result.error.code !== "number") throw new Error(`Could not run screenshot diff: ${result.error.message}`);
if (result.error) {
const detail = (result.stderr || result.stdout || "unknown error").trim().slice(0, 500);
throw new Error(`Screenshot diff failed: ${detail}`);
}
+1 -1
View File
@@ -3,4 +3,4 @@
* controller (`--package-version`), the daemon health check, and the screenshot diff tool must
* all be this version; each caller checks it by exact equality.
*/
export const AGENT_DEVICE_VERSION = "0.21.28";
export const AGENT_DEVICE_VERSION = "0.21.29-nightly.20261018.74";
+12 -1
View File
@@ -55,6 +55,11 @@ export function ciErrorForCommand(error: unknown, command: string): CiError {
export type CiErrorRenderOptions = { secrets?: readonly string[] };
/** The values of every variable that names a secret (a token, secret, password or API key), to redact from what Explore writes. */
export function environmentSecrets(environment: NodeJS.ProcessEnv): string[] {
return Object.entries(environment).filter(([key]) => /(?:TOKEN|SECRET|PASSWORD|API_KEY)/i.test(key)).flatMap(([, value]) => value ? [value] : []);
}
function redact(value: string, secrets: readonly string[]): string {
let result = value;
// Controller failures may embed JSON or URLs inside their message rather than
@@ -148,6 +153,12 @@ export async function reportCiError(error: unknown, options: {
env?: NodeJS.ProcessEnv;
secrets?: readonly string[];
diagnosticsPath?: string;
/**
* The matrix target the failed command ran for, written into the diagnostics file. actions/download-artifact puts an
* artifact in a directory named after it only when a pattern matches several, so the failure job cannot always read
* the target from where the file lands.
*/
target?: string;
write?: (value: string) => void;
} = {}): Promise<CiErrorData> {
const env = options.env ?? process.env;
@@ -166,7 +177,7 @@ export async function reportCiError(error: unknown, options: {
if (env.GITHUB_STEP_SUMMARY) outputs.push({ label: "job summary", operation: appendFile(env.GITHUB_STEP_SUMMARY, `${renderCiErrorSummary(error, renderOptions)}\n`) });
if (options.diagnosticsPath) outputs.push({ label: "diagnostics file", operation: (async () => {
await mkdir(dirname(options.diagnosticsPath!), { recursive: true });
await writeFile(options.diagnosticsPath!, `${JSON.stringify(diagnostic, null, 2)}\n`);
await writeFile(options.diagnosticsPath!, `${JSON.stringify(options.target ? { ...diagnostic, target: options.target } : diagnostic, null, 2)}\n`);
})() });
const results = await Promise.allSettled(outputs.map(output => output.operation));
for (const [index, result] of results.entries()) {
+1 -1
View File
@@ -21,7 +21,7 @@ export type ControllerName = "agent-device" | "argent";
/** Each controller's npm release, the file Explore runs from it, and the integrity npm publishes for it. */
export const CONTROLLER_PACKAGES: Record<ControllerName, { name: string; version: string; bin: string; integrity: string }> = {
"agent-device": { name: "agent-device", version: AGENT_DEVICE_VERSION, bin: "bin/agent-device.mjs",
integrity: "sha512-uGEENVhPzSpIACrVlt16O1mloODWJyAemTCT0boDb4L006JXN2rG747u1TRoQJudOpdP/Pm2urxNRHeECLMLfg==" },
integrity: "sha512-7+ajWLr0J+8iLZ9q11ERbj6U14mC20umBK5BQQdw5aERA79yeXc3Ze6KmtvnILnWjEtoVkYQsXQasL8RL6Uatw==" },
argent: { name: "@swmansion/argent", version: ARGENT_VERSION, bin: "dist/cli.js",
integrity: "sha512-XXWlWZJlHKqnAvNx3RDhuiVFQOPqh5EDB76+mFAyqC2mutC6irzw03+M/6S/WQ+Dzz2IYOkEfOzQnOOdhrGgJg==" },
};
Loaded 100 of 241 files, more files were not shown because too many files have changed in this diff. Show more