* [build-tools] Check all certificates in provisioning profile during verification
Apple provisioning profiles can contain multiple developer certificates
(e.g., when a team has multiple developers). The previous code only
checked DeveloperCertificates[0], causing false "profile and
distribution certificate don't match" errors when the signing
certificate was at a different index.
Co-Authored-By: Claude Opus 4.6 <[email protected]>
* [build-tools] Add CHANGELOG entry for multi-cert provisioning profile fix
Co-Authored-By: Claude Opus 4.6 <[email protected]>
---------
Co-authored-by: Claude Opus 4.6 <[email protected]>
# Why
Some environments that run Gradle do not have `en_US.UTF-8` available, which can lead to locale warnings and Gradle failures when Java falls back to a non-UTF-8 default charset.
In the affected environment, `locale -a` returned only:
```sh
C
C.utf8
POSIX
```
Setting `LC_ALL=C.UTF-8` fixed the Gradle issue. This matches the behavior described in [gradle/gradle#23391](https://github.com/gradle/gradle/issues/23391#issuecomment-1878979127), where Gradle/archive handling fails under a non-UTF-8 locale.
# How
Inline `LC_ALL: 'C.UTF-8'` at the two Android Gradle subprocess call sites in `@expo/build-tools`:
- `packages/build-tools/src/android/gradle.ts`
- `packages/build-tools/src/steps/utils/android/gradle.ts`
This keeps the change narrowly scoped to Gradle execution instead of changing the broader worker environment or unrelated iOS/macOS build paths.
# Test Plan
- Verified the affected environment does not support `en_US.UTF-8`:
```sh
locale -a
```
Output:
```sh
C
C.utf8
POSIX
```
- Confirmed that setting `LC_ALL=C.UTF-8` fixes the Gradle issue locally.
- Ran a targeted TypeScript check after the code change:
```sh
./node_modules/.bin/tsc -p packages/build-tools/tsconfig.json --noEmit
```
# Why
The Android builder currently only injects Gradle config during `PREPARE_CREDENTIALS`. That means `job.version.versionCode` and `job.version.versionName` are ignored for builds that do not provide Android build credentials, even though those version values are present on the job.
This ports the Android version injection change from `expo/eas-build#425`, now on top of the merged enum-backed `CONFIGURE_ANDROID_VERSION` build phase.
# How
- switch the Android builder from the legacy `eas-build.gradle` shim to the existing direct Gradle injection helpers
- keep credential injection gated on build credentials
- inject the Android version Gradle config whenever `job.version` includes `versionCode` or `versionName`, even if there are no credentials
- run that injection in `BuildPhase.CONFIGURE_ANDROID_VERSION`
- remove the now-unused legacy shim files
- warn when deprecated `eas-build.gradle` still exists, matching the previous conservative behavior
- add focused Android builder coverage for the no-credentials version case
- add helper coverage for the legacy warning path
# Test Plan
- `mise exec -- yarn workspace @expo/build-tools jest-unit src/steps/utils/android/__tests__/gradleConfig.test.ts src/builders/__tests__/android.test.ts`
- `mise exec -- yarn workspace @expo/build-tools test`
- `mise exec -- yarn workspace @expo/build-tools typecheck`
- `git diff --check`
# Why
Preparation for https://github.com/expo/eas-cli/pull/3483.
# How
- add `BuildPhase.CONFIGURE_ANDROID_VERSION`
- add display metadata for the new phase
- add website ID metadata for the new phase
# Test Plan
CI should pass.
## Why
Old task I asked Codex to do.
## How
- upgrade `eas-cli` from oclif v3 to v4 using the native v4 APIs
- migrate legacy array-style `static args = [...]` declarations to native v4 object-style `Args.*` definitions
- keep command typing fully native and type-safe, including `Flags.option(...).options` const assertions where needed for inference
- update tests and shared test helpers for the v4 config/hook behavior
Notes:
- argument order is preserved in oclif v4 because `static args = { ... }` is treated as an ordered object; positional args follow object insertion order
- the bin entrypoints now use `@oclif/core` `execute(...)`, which is the supported v4 boot path
- `execute(...)` already performs the old `run().then(flush).catch(handle)` flow internally, so the explicit `flush` and `handle` chaining was removed as redundant rather than as a behavior change
- `bin/dev` keeps the dev entrypoint behavior through `execute({ development: true, ... })`, which enables the development boot path and verbose debugging behavior in v4
## Test plan
- `./node_modules/.bin/tsc --project packages/eas-cli/tsconfig.build.json`
- `node packages/eas-cli/bin/run --help`
- `node packages/eas-cli/bin/dev --help`
- `npx -y @yarnpkg/[email protected].0 workspace eas-cli test --runInBand`
Instead of throwing an error when running `eas build -p android --non-interactive`
without existing credentials, automatically generate a keystore using the existing
`generateRandomKeystoreAsync()` function.
Co-authored-by: Claude Opus 4.5 <[email protected]>
# Why
Provide a way to override the new requirement for the `--environment` flag in `eas update`.
# How
Modify the check to skip when the `EAS_UPDATE_SKIP_ENVIRONMENT_CHECK` variable is set to a boolish value (1 or true).
If the command is run interactively, we now prompt the user for the environment, instead of throwing an error. This is skipped if the `--auto` flag is set. A missing environment flag already throws an error if `--non-interactive` is set.
Add the `CI` flag to the tests for whether we should skip the check in a CI environment.
# Test Plan
New unit tests added.
<!-- 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
We want to push logs to a logs service.
# How
Initially I wanted to set up a WebSocket connection between worker and the logs service to reduce the HTTP overhead, but I found having a persistent/reconnecting WebSocket connection is not trivial and HTTP should be ok.
Confirmed failure to upload is not logged in the GCS log.
# Test Plan
Tested manually.
https://github.com/user-attachments/assets/f80f0482-73f5-4dfb-855f-2361ed04aa4f
## Summary
- remove the duplicated `userFacingMessage` and `userFacingErrorCode` fields from the build error flow
- keep current runtime behavior and unknown-error semantics
- preserve the current worker protocol shape
Stanley: I don't want to have separate `userFacingMessage` and `message` / `userFacingErrorCode` and `errorCode`. It looks like the reason we had them separate was to log "UNKNOWN_ERROR" to the user if we encounter an internal issue like "NPM_CORRUPTED_PACKAGE". This pull request inverts the logic — we're going to have a `trackingCode` so we keep logging "NPM_CORRUPTED_PACKAGE" to Sentry and the default `errorCode` is the one we log to the user.
## Changes
- make `BuildError.format()` use canonical `message` and `errorCode`
- make `build-tools` store internal-only classification in `trackingCode` instead of parallel public fields
- make worker derive `internalErrorCode` from `trackingCode ?? errorCode`
- delete the obsolete `userFacing...` fields from `BuildError`
## Testing
- `packages/eas-build-job`: `npm exec -- corepack yarn test src/__tests__/errors.test.ts --runInBand`
- `packages/build-tools`: `npm exec -- corepack yarn jest-unit src/buildErrors/__tests__/detectError.test.ts --runInBand`
- `packages/worker`: `npm exec -- corepack yarn test:integration src/__integration__/stateSync.test.ts`
## Why
`submit_ios_from_linux` UNKNOWN_ERROR analysis showed App Store Connect responses that clearly indicate the same version-train failure class but with mixed codes (`90062`, `90186`, `90478`).
Current handling recognizes only `90062` and `90186`, so runs that include `90478` fall back to `UNKNOWN_ERROR`.
## How
- extend `isClosedVersionTrainError` to include ASC code `90478` while keeping strict `.every` matching
- broaden `EAS_UPLOAD_TO_ASC_CLOSED_VERSION_TRAIN` user-facing message to cover both closed train and lower-than-approved version cases
- update unit test to include `90478` in positive closed-train detection
## Examples
Existing `EAS_UPLOAD_TO_ASC_CLOSED_VERSION_TRAIN` runs:
- https://expo.dev/uuid/019cbe83-ac3b-756f-b7ef-447bae5ef4dc
- https://expo.dev/uuid/019cbe66-4f9b-7d95-a35e-06886b646434
Would-be closed-train runs currently persisted as `UNKNOWN_ERROR` (log errors include `90478` with `90062`/`90186`):
- https://expo.dev/uuid/019cbe71-62fa-7fee-9e77-bdc051cc6e8b
- https://expo.dev/uuid/019cbe44-5c75-7211-af66-35fe96a86903
## Test Plan
- `yarn --cwd packages/build-tools jest-unit src/steps/functions/__tests__/uploadToAsc.test.ts`
- `yarn --cwd packages/build-tools typecheck`
- `yarn fmt`
## Why
Uploads to App Store Connect can fail with code `90683` when required `*UsageDescription` privacy keys are missing in `Info.plist`.
Without dedicated handling this ends up as a generic upload failure, even though ASC already returns structured details that can guide the user.
## How
- add ASC error recognition for `90683` in `uploadToAsc` failed-state handling
- throw a specific `UserFacingError` with code `EAS_UPLOAD_TO_ASC_MISSING_PURPOSE_STRING`
- parse ASC error descriptions to extract and list missing `*UsageDescription` keys
- provide concrete remediation for both CNG (`ios.infoPlist`) and non-CNG (`Info.plist`) workflows
- set `docsUrl` on the user-facing error to Expo permissions docs
- add unit tests for detection and missing-key parsing helpers
## Test Plan
- `yarn --cwd packages/build-tools jest-unit src/steps/functions/__tests__/uploadToAsc.test.ts`
- `yarn --cwd packages/build-tools typecheck`
- `yarn fmt`
## Why
Some App Store Connect uploads currently end up as generic failures even when ASC returns explicit bundle identifier mismatch validation ([`90054`](https://expo.dev/accounts/dobrokruh/projects/dobrokruh/job-runs/019cb865-ef97-71d1-9137-f6ae4c05d94d)/[`90055`](https://expo.dev/accounts/central-studio/projects/hfc-fuels/job-runs/019cb882-1550-7801-9b4b-9647e1bce364)). We want a user-facing error that explains the mismatch and shows both identifiers so users can quickly determine whether they selected the wrong app or the wrong build profile/build.
## How
- add `isInvalidBundleIdentifierError` detector in `uploadToAsc` for ASC error codes `90054`/`90055`
- when matched, throw `UserFacingError` with code `EAS_UPLOAD_TO_ASC_INVALID_BUNDLE_ID`
- include both identifiers in message:
- IPA bundle identifier (best-effort read via `readIpaInfoAsync`)
- selected ASC app bundle identifier
- provide remediation covering both paths:
- wrong selected app identifier in submit profile
- wrong build selected / rebuild with different profile
- use `asyncResult` for best-effort IPA introspection (no nested catch needed)
- add unit tests for invalid bundle-id detector
## Test Plan
- `yarn --cwd packages/build-tools jest-unit src/steps/functions/__tests__/uploadToAsc.test.ts`
- `yarn --cwd packages/build-tools 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
Older (default) Android system image package does not support `-z`.
```
Forward stdin/stdout to a file or network connection.
-4 Force IPv4
-6 Force IPv6
-L Listen for multiple incoming connections (server mode)
-U Use a UNIX domain socket
-W SECONDS timeout for more data on an idle connection
-f Use FILENAME (ala /dev/ttyS0) instead of network
-l Listen for one incoming connection
-p Local port number
-q Quit SECONDS after EOF on stdin, even if stdout hasn't closed yet
-s Local source address
-t Allocate tty (must come before -l or -L)
-u Use UDP
-w SECONDS timeout to establish connection
```
# How
Replaced `-z` with a closed stdin.
# Test Plan
System tests should work.
Deployed worker from this branch at https://github.com/expo/eas-cli/actions/runs/22666833774/job/65700594158.
Confirmed works at https://staging.expo.dev/accounts/expo-services/projects/worker-system-tests/workflows/019cb88e-f710-74d3-87ba-3c17bfe45c00.
## Why
Generic/custom workflow steps can throw `UserFacingError`, but worker-level error handling currently maps any non-`BuildError` to `UnknownBuildError` / `UnknownCustomBuildError`. That causes actionable step errors to be persisted and exposed as `UNKNOWN_ERROR`.
## How
- add a `toBuildError` conversion in worker service catch path
- preserve existing `BuildError` values
- convert `UserFacingError` into `BuildError` with the same code/message/docsUrl
- keep unknown fallbacks only for truly unstructured errors
- add integration coverage in `stateSync` to assert generic job `UserFacingError` is exposed in `externalBuildError` and `internalErrorCode`
We should make a bigger refactor to error handling, but it's more involved — https://github.com/expo/eas-cli/pull/3447.
## Test Plan
CI should pass.
```
PASS src/__integration__/stateSync.test.ts
State sync mechanism
query new worker
✓ should return 'new' state (210 ms)
query worker after dispatch
✓ should return 'in-progress' state (170 ms)
query worker after build
✓ should return 'success' state (2062 ms)
query worker after error
✓ should return 'error' state with artifacts from error (88 ms)
✓ should expose user-facing error from generic job (60 ms)
```
Co-authored-by: Copilot <[email protected]>
https://github.com/user-attachments/assets/37835e4d-fca0-48a3-87e7-8c26e0c80a32
## Summary
- When running `eas go` while not logged in, an `ora` spinner ("Logging in to Expo...") was started before `getContextAsync` was called
- `getContextAsync` triggers interactive login prompts when unauthenticated, but the active spinner animation conflicts with readline input display, causing garbled/invisible prompts
- Fix: remove the spinner before `getContextAsync` and replace with `Log.withTick(...)` after it completes
## Test plan
- [ ] Run `eas go` while not logged in and verify login prompts are visible and interactive
- [ ] Run `eas go` while already logged in and verify "Logged in as ..." tick appears as expected
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-authored-by: Claude Sonnet 4.6 <[email protected]>
## Why
App Store Connect build upload failures were often surfaced as generic errors, which made remediation unclear for common cases like duplicate build numbers. We also needed stricter and safer response parsing so JSON parsing failures don’t leak raw parse errors.
## How
- Added typed `AscApiRequestError` in `AscApiClient` for structured ASC error payloads only.
- Kept non-structured ASC failures as regular `Error`.
- Hardened response parsing in `AscApiClient`:
- parse response text with `asyncResult` + `JSON.parse`,
- convert invalid success JSON into a controlled `Malformed JSON response...` error.
- Added `AscApiUtils.createBuildUploadAsync` and switched `uploadToAsc` to call it.
- Added duplicate build-number recognition for `/v1/buildUploads` failures when all ASC error codes are `ENTITY_ERROR.ATTRIBUTE.INVALID.DUPLICATE`.
- Added specific `UserFacingError` (`EAS_UPLOAD_TO_ASC_VERSION_DUPLICATE`) with concrete remediation:
- increment `ios.buildNumber` (or set `"autoIncrement": true`),
- rebuild and resubmit,
- include docs URL.
- Inlined closed-train constants in `uploadToAsc` as requested.
## Test Plan
- `yarn --cwd packages/build-tools jest-unit src/steps/utils/ios/__tests__/AscApiClient.test.ts`
- `yarn --cwd packages/build-tools jest-unit src/steps/utils/ios/__tests__/AscApiUtils.test.ts`
- `yarn --cwd packages/build-tools jest-unit src/steps/functions/__tests__/uploadToAsc.test.ts`
All commands above pass on this branch.
# Why
This adjusts the change in #3418 that added a requirement for the `--environment` flag in `eas update` for SDK 55 and greater.
The requirement should not have applied to the `update:configure` subcommand. This PR removes the requirement there.
# How
Remove the check in `configure.ts`, and adjust flag descriptions.
# Test Plan
Existing tests should pass.
# Why
When yarn moved to version 4, they removed the `--silent` flag. Because of this change, any use of yarn to call scripts and get output will also get yarn debugging garbage, and there is no way to suppress it.
The release scripts are mostly ok with this, except for calling the `next-version` script. The output of this script has to be clean for it to be used as input for `lerna version`.
# How
Replaced the yarn call with a direct invocation of the node Typescript source, bypassing yarn.
# Test Plan
Since the `trigger-release.yml` workflow supports a dry run mode, it's safe to modify it to check out a test branch and run the workflow many times to debug the issue.
Once this change is merged, a dry run invocation of the workflow should succeed even if no version string is passed in.
# Why
Problems can occur if an update intended for a development build is published with the environment defaulting to or set to production, and vice versa.
To prevent such problems in the future, this PR makes the `--environment` flag a required flag for projects using SDK 55 and greater.
# How
Add a new utility function to throw an error if sdkVersion does not meet the requirement, and use that function in `eas update:configure` and `eas update`.
Skip the check if EAS_BUILD env is set.
# Test Plan
Unit tests added.
Testing locally with different projects.