<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
https://github.com/expo/workflow-orchestration/pull/851
# How
Added `RuntimeSettings` namespace in `build-tools`. It is expected to be called in the beginning with `loadAsync` to fetch data from the bucket.
# Test Plan
Once approved deploy to staging, run a few builds with different settings.
I want us to move off of killswitches in Redis coming from `workflow-orchestration` towards fetching killswitches from a simple JSON file.
I picked GCS over R2 because it lets us make objects fully public (no `Authorization` required) and I (Stanley) found no way to achieve the same thing with R2.
This pull request adds a script `scripts/edit-runtime-settings staging | production` that is going to let us edit the file.
* [eas-cli] Add eas update:embedded:list command
* [eas-cli] Put "All channels" first in the channel-filter prompt
* [eas-cli] Refresh graphql schema snapshot
* [build-tools] Disable apsd in iOS Simulator after boot
* [build-tools] Add unit test for disableApsdAsync
* [build-tools] Use { err } logging convention for apsd warning
* [build-tools] Test apsd disable in start_ios_simulator
<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
Use webrtc for better perf
# How
Use webrtc with serve sim for better perf
# Test Plan
Looks good locally
* [eas-cli] Add eas update:embedded:upload command
* Add tests for json output flag and mutation wrappers
* Apply prettier formatting to mutation test
* Invoke upload progress callback in test to cover no-op
* Use fs-extra so manifest read and bundle existence check can be awaited
* Drop misleading already-registered warn from upload
* Recognize EMBEDDED_UPDATE_ALREADY_EXISTS on upload and warn
* Drop already-exists CLI warn, server message is clear on its own
* Surface EMBEDDED_UPDATE_ALREADY_EXISTS as a single clean error on upload
* Flatten retry catch with early continue
<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
Try if h264 codec will improve stream perf for serve-sim
# How
Add `--codec h264` to the single `serve-sim` invocation in `startServeSimWithTunnelAsync` ([packages/build-tools/src/steps/utils/remoteDeviceRunSession.ts](https://github.com/expo/eas-cli/blob/main/packages/build-tools/src/steps/utils/remoteDeviceRunSession.ts)). All three callers — `start_agent_device_remote_session`, `start_argent_remote_session`, and `start_serve_sim_remote_session` — route through this helper, so the flag applies everywhere `serve-sim` is launched.
# Test Plan
- `yarn typecheck` on `build-tools` — only the two pre-existing `refreshAdHocProvisioningProfile` errors on `main`, unrelated to this change.
- Manual verification will happen once the step runs end-to-end on a turtle worker; the change is a single CLI flag added to an existing argument list.
### Motivation
- Ensure `EXPO_USE_PRECOMPILED_MODULES=0` in the build environment always wins and is not overwritten by `EAS_USE_PRECOMPILED_MODULES=1` to avoid surprising behavior for customers disabling precompiled modules.
- Make the precedence explicit and observable via logs so maintainers can reason about the opt-out path.
### Description
- Added an early guard in `resolvePrecompiledModulesPodInstallEnvAsync` to return early when `ctx.env.EXPO_USE_PRECOMPILED_MODULES === '0'` and log the opt-out message. (`packages/build-tools/src/ios/pod.ts`).
- Logged the presence of `EXPO_USE_PRECOMPILED_MODULES=0` with a clear message to explain why precompiled env vars are not injected. (`packages/build-tools/src/ios/pod.ts`).
- Extended the test helper to allow passing `EXPO_USE_PRECOMPILED_MODULES` via the build context environment and added a regression test that verifies `EXPO_USE_PRECOMPILED_MODULES=0` is preserved and that no base URL is set when both flags are present. (`packages/build-tools/src/ios/__tests__/pod.test.ts`).
### Testing
- Ran formatter: `npx oxfmt packages/build-tools/src/ios/pod.ts packages/build-tools/src/ios/__tests__/pod.test.ts` which completed successfully.
- Ran the unit test file: `cd packages/build-tools && yarn test src/ios/__tests__/pod.test.ts` which failed in this environment due to missing workspace module resolution for `@expo/eas-build-job` and `@expo/turtle-spawn` (test runner error, not a behavior regression).
------
[Codex Task](https://chatgpt.com/codex/cloud/tasks/task_e_6a180a1026848328b76bf4240a5872b1)
* [build-tools] Use ngrok tunnels for remote device run sessions
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* [build-tools] Use in-process @ngrok/ngrok SDK for our tunnels
Replace the spawned ngrok CLI (binary install + stdout URL parsing +
web_addr workaround) with the in-process @ngrok/ngrok SDK for the
agent-device and argent tunnels. The ngrok CLI is still installed for
serve-sim, which shells out to it for --tunnel-provider ngrok.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* [build-tools] Drop ngrok CLI install; serve-sim uses the SDK
serve-sim now embeds the @ngrok/ngrok SDK for --tunnel-provider ngrok, so
the ngrok agent no longer needs to be on PATH. No tunnel binary is
installed anywhere now: our own tunnels and serve-sim both run the agent
in-process. Removes ensureNgrokCliInstalledAsync and its callers.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
* [build-tools] Address review feedback
Simplify the "env var not set" SystemError messages (drop "created by the
API server" / the domain example), reword the fail-fast comments, and make
matchLabeledUrl take an options object.
Co-Authored-By: Claude Opus 4.7 (1M context) <[email protected]>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <[email protected]>
* [build-tools] Fix iOS Simulator readiness hang on Xcode 26.4
* [build-tools] Add changelog entry
* [build-tools] Use os.tmpdir() and node:fs for readiness screenshot
* [build-tools] Add unit test for Simulator readiness screenshot
* [eas-cli] eas go: use SDK version from current project when available
* [eas-cli] eas go: prompt for SDK version interactively
* Regenerate GQL types for ExpoGoBuild supportedSdkVersions
* Update packages/eas-cli/src/commands/go.ts
Co-authored-by: Stanisław Chmiela <[email protected]>
* Address review feedback on selectSdkVersionAsync
---------
Co-authored-by: Stanisław Chmiela <[email protected]>
## Summary
- adds `Datadog.log` for posting build-scoped log events to Datadog
- moves the Gradle cache restore log emission onto the shared Datadog helper
- replaces artifact dry-run Sentry reporting with Datadog logs for match, mismatch, and error outcomes
## Why
The absolute artifact glob validation should show both positive and negative dry-run outcomes. Sentry only surfaced mismatches/errors, while Datadog logs let us count successful matches as well.
## Validation
CI should pass. Going to review Datadog logs post-merge.
<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
Git-based integration builds call `eas build:internal` via `runEasBuildInternalAsync`. The refresh flag must reach the CLI and remain on the job object returned to the rest of the build.
# How
Forward `--refresh-ad-hoc-provisioning-profile` when set on the iOS job; retain `refresh_ad_hoc_provisioning_profile` in `validateEasBuildInternalResult`.
# Test Plan
`cd packages/build-tools && yarn test src/common/__tests__/easBuildInternal.test.ts`
<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
Workflow builds run `eas build:internal` on EAS workers to resolve credentials and emit the job JSON. The base branch adds `--refresh-ad-hoc-provisioning-profile` to eas build, but that flag was not available on `build:internal`, and the iOS job did not persist `refresh_ad_hoc_provisioning_profile` when the flag was set.
Without this wiring, git-based/workflow builds could not trigger ad-hoc profile refresh or pass the intent through to the job payload.
# How
- Add `--refresh-ad-hoc-provisioning-profile` to `eas build:internal` and pass it into `runBuildAndSubmitAsync` (same behavior as eas build).
- In `prepareJob.ts`, set `refresh_ad_hoc_provisioning_profile: true` on the iOS job when the credentials context has the flag enabled.
# Test Plan
`yarn typecheck`
<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
ENG-21094 needs a job field to carry ad-hoc provisioning profile refresh intent for git-based/workflow builds; #3716 added CLI behavior on eas build; this PR is schema-only.
# How
Updated `packages/eas-build-job/src/ios.ts` interface + Joi schema.
# Test Plan
`cd packages/eas-build-job && yarn test src/__tests__/ios.test.ts`
`yarn typecheck`
We now send build phases through `Datadog`!
---
## Summary
- remove the stale BuildPhaseStats type and BuildContext reporting callback
- drop the worker BUILD_PHASE_STATS websocket message and service forwarding path
- update tests to assert the remaining Datadog phase duration metrics path
## Why
Build phase metrics now flow from worker to www to Datadog, so the old worker to launcher to Datadog reporting path was dead code.
## Validation
- yarn run -T oxfmt --check packages/eas-build-job/src/common.ts packages/eas-build-job/src/index.ts packages/build-tools/src/context.ts packages/build-tools/src/builders/__tests__/android.test.ts packages/worker/src/context.ts packages/worker/src/service.ts packages/worker/src/external/turtle.ts packages/worker/src/__integration__/utils.ts packages/worker/src/__unit__/service.test.ts
- yarn jest --config jest/unit-config.ts src/builders/__tests__/android.test.ts src/__tests__/context.test.ts --runInBand --no-cache (from packages/build-tools)
- yarn jest --config jest.config.unit.ts src/__unit__/service.test.ts --runInBand (from packages/worker)
- yarn typecheck
## Summary
Updates artifact path resolution so absolute paths containing glob metacharacters are treated as glob patterns instead of literal paths.
This lets `eas/upload_artifact` handle paths like `/var/folders/.../T/maestro_xctestrunner_xcodebuild_output*`, while preserving the existing literal-path behavior for non-dynamic absolute paths.
## Why
`upload_artifact.path` already accepts paths or patterns, but absolute patterns were previously short-circuited through `fs.pathExists()` and never reached `fast-glob`.
## Test Plan
- `yarn workspace @expo/build-tools jest-unit packages/build-tools/src/utils/__tests__/artifacts.test.ts`
Note: the focused test was run in the existing local checkout with the same two-file patch. The clean PR worktree did not have its own install state.
## Summary
- Remove worker-side `resourceClass` parsing and types from runtime config.
- Remove `ResourceClassToPlatform`, since worker env setup now uses the job platform directly.
- Replace resource-class VM spec logging with OS-derived CPU model, vCPU count, and total RAM.
## Notes
- CPU model is normalized to `Apple M4 Pro` (from "Apple M4 Pro (Virtual)") for Apple chips and `Intel`/`AMD` for cloud CPU models.
## Validation
- `yarn oxfmt --check packages/worker/src/config.ts packages/worker/src/displayRuntimeInfo.ts packages/worker/src/external/turtle.ts packages/worker/src/env.ts packages/worker/src/__unit__/env.test.ts`
- `yarn workspace @expo/worker test:unit -- env.test.ts`
- `yarn workspace @expo/worker typecheck`
- verified `os.cpus().length` is 4 for 4 CPU tart VM and `os.totalmem() / 1024 / 1024 / 1024` is `8` for 8 GB RAM tart VM
## Summary
- Stop reporting project dependency metrics failures to the build logger.
- Keep the existing Sentry warning capture for this failure path.
## Context
This addresses user-visible logging seen in build https://expo.dev/accounts/ammar_tripiq/projects/tripiq-client/builds/8f01c353-187b-4368-b569-94484d04220b where missing package.json caused:
```
Failed to report project dependencies metrics
ENOENT: no such file or directory, open '/Users/expo/workingdir/build/package.json'
```
## Validation
- yarn oxfmt packages/worker/src/external/analytics.ts
Not sure if this is the right fix, but sounds right?
## Summary
Updates the iOS build resign Fastfile to pass `use_app_entitlements: true` to Fastlane's `resign` action.
This makes EAS preserve and merge the app's existing code-signing entitlements when re-signing an IPA with a new provisioning profile, instead of signing directly with the broader provisioning-profile entitlements. This matters for capabilities like associated domains, where the profile may contain `*` but the app signature needs concrete values such as `applinks:...` for universal links to keep working.
Context: https://exponent-internal.slack.com/archives/C09H6RFAX27/p1779446992844929
## Validation
- `yarn install`
- `yarn workspace @expo/template-file build`
- `yarn workspace @expo/build-tools jest-unit fastfile.test.ts`
- `yarn oxfmt packages/build-tools/src/templates/FastfileResign.ts packages/build-tools/src/ios/__tests__/__snapshots__/fastfile.test.ts.snap`
<!-- If this PR requires a changelog entry, add it by commenting the PR with the command `/changelog-entry [breaking-change|new-feature|bug-fix|chore] [message]`. -->
<!-- You can skip the changelog check by labeling the PR with "no changelog". -->
# Why
When a new device is added to an account, and a non-interactive development build is run (e.g. invoked by CI or by a workflow job), even without the `freeze-credentials` flag, the provisioning profile is not refreshed, and the internal build is not available for the newly registered device. See the Linear issue: https://linear.app/expo/issue/ENG-7041/allow-including-all-eas-registered-devices-automatically-in-non
# How
Added an opt-in ad-hoc provisioning profile refresh to non-interactive builds with flag `--refresh-ad-hoc-provisioning-profile`.
When the flag is present, we hit `www` GQL API to get the ASC API key (the submission key), and we invoke the ad-hoc profile sync with all devices selected. `packages/eas-cli/src/credentials/ios/actions/SetUpAdhocProvisioningProfile.ts` is where this logic happens.
The flag mustn't be present when `--freeze-credentials` is present, these are conflicting flags.
I'm leaning towards an opt-in flag instead of extending the default behavior because of the extra credentials configuration requirements and side effects of the provisioning profile refresh logic.
# Test Plan
Added/amended tests.
Manual verification:
1. Did `easd build --platform ios --profile development --non-interactive --refresh-ad-hoc-credential`, verified the provisioning profile matches the set of Apple devices on my account.
2. Changed the set of Apple devices on my account.
3. `easd build --platform ios --profile development --non-interactive --refresh-ad-hoc-credential` again, verified the re-generated provisioning profile matches the Apple devices on my account.