Fix the Android emulator gate and update every workflow - #79
Merged
Merged
Conversation
The Android smoke test has been failing on every run since master last built. It is not a flake and it is not the code: re-running master's own 2026-08-04 build today fails identically, on the same commit that passed then. The runner image moved from ubuntu24/20260720.247 to 20260823.283 in between. The emulator process started, printed "Found systemPath" and its core-count warning, then died without ever registering with adb. It was being given 2 cores, 2GB of RAM and a 2GB disk with a further -partition-size 2048 on top — a starved configuration that the older, smaller system images tolerated and the current ones do not. It now gets what the runner actually has, and the reclaim step that used to run for the emulator's benefit reports free space either side, so a future failure to boot can be told apart from a failure to run. Switching from google_atd to aosp_atd drops the Google APIs the app never used, for a lighter image and a faster boot. The emulator action was pinned to a SHA that no longer resolves to any commit in its repository, so what was actually running could not be audited. It is now pinned to v2.38.0 by SHA, with the version in a comment. Every other action across all four workflows — and the one the template ships to a user's game — was one to four major versions behind. All are current. Concurrency groups cancel superseded PR runs but never a release, and each workflow declares its own least-privilege permissions. Claude-Session: https://claude.ai/code/session_01NkneJGya3dNr2huxsgfuja
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The Android smoke test has been red on every run since master last built. It is not a
flake and it is not the code.
Proof: I re-ran the Android job from master commit
30951244265— untouched code thatpassed on 2026-08-04 — and it failed today with the identical
Timeout waiting for emulator to boot. The runner image moved fromubuntu24/20260720.247toubuntu24/20260823.283in between.What was wrong
The emulator started, printed
Found systemPathand its core-count warning, then diedwithout ever registering with adb. It was being given 2 cores, 2GB RAM and a 2GB disk,
with a further
-partition-size 2048on top. The older, smaller system images toleratedthat; the current ones do not.
It now gets what the runner actually has (4 cores, 4GB RAM, 8GB disk, no redundant
partition override), and
aosp_atdreplacesgoogle_atd— the app never used Google APIs,so this is a lighter image and a faster boot.
The reclaim step that existed for the emulator's benefit now reports free space either
side, so the next failure can be told apart at a glance: no disk, or a real crash.
The pin nobody could read
ReactiveCircus/android-emulator-runnerwas pinned to4c44018e59b437e86cdfc41da381398f93ed8808, which does not resolve to any commit in thatrepository —
GET /commits/{sha}returns 422. Actions could still fetch it, but nobodycould audit what was running. Now pinned to
v2.38.0by SHA with the version in a comment.Everything else was stale too
Every action in the repo was one to four major versions behind:
actions/checkoutactions/setup-dotnetactions/setup-javaactions/upload-artifactactions/download-artifactandroid-actions/setup-androiddocker/setup-buildx-actiondocker/login-actiondocker/build-push-actionsoftprops/action-gh-releaseI checked the two risky bumps against their
action.ymlrather than assuming:download-artifact@v8still hasmerge-multiple, andsetup-android@v4's inputs are alloptional with workable defaults.
Also updated the workflow the template ships to a user's game — a generated project
should not start life four majors behind.
Added
concurrencygroups (cancel superseded PR runs, never cancel a release) andper-workflow least-privilege
permissions.Verification
All five workflow files parse. The real check is this PR's own CI run: if the Android job
goes green, the gate is fixed, since that is the same job on the same runners that has
failed three times in a row today.
https://claude.ai/code/session_01NkneJGya3dNr2huxsgfuja