Repository navigation
chore(ci): run the full suite on sync and release PRs - #490
Merged
Merged
Conversation
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
Owner
Author
|
The description edit restarted the suite as documented: run 43511819370 (event |
This was referenced Nov 13, 2026
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Why
Until now the full suite in
test.yml(unitande2e) ran only frompublish.yml, pushedv*tags andworkflow_dispatch. Pull requests got onlytypecheck.yml. As a result, the upstream v2.3.0 sync (#489) merged without unit tests, and its first unit failure (the PowerShell ShellTool timeout) surfaced in the 2.3.0-shuv.1 publish run.What
test.ymlnow also runs onpull_request(opened,synchronize,reopened,edited,labeled). Bothunitande2eshare one job-levelifthrough a YAML anchor. On pull requests they run only when one of these is true:chore(sync)orchore(release)(the convention across chore(sync): merge upstream dev past v2.2.5 #473 to chore(sync): merge upstream v2.3.0 and prepare 2.3.0-shuv.1 #489, whose branch names vary), orupstream-synclabel, orfull-testslabel ("Run the full test suite (unit and e2e) on this PR").Both jobs run because both gate publish. The aim is that publish never finds a new failure.
Concurrency. This mirrors
typecheck.yml. Pull requests sharetest-pr-<number>withcancel-in-progress: true. Every other run getstest-run-<run_id>, so release runs never share a group and are never cancelled. Thetest-prefix keeps the group apart from any caller's group:publish.yml's ispublish-<ref>-<version>, andtypecheck.yml's istypecheck-….Description edits restart a gated PR's suite (accepted and documented). My first commit skipped body-only edits and unrelated labels, and gave those runs a group of their own. On this PR that showed a real problem. A PR's checks list (
statusCheckRollup), and branch protection, keep only the newestunitande2echeck runs. The skipped run from a description edit replaced the real, still-running suite, andgh pr checksreportedunit skipping. After a real failure, that would hide the failure. GitHub cannot filterpull_requesttriggers by edited field or label name, so there is no cheap way to avoid creating that run. Every event on a gated PR now restarts the suite instead. The comment intest.ymland theAGENTS.mdnote explain why.Cancelled
unitruns stop promptly.unit's three build smoke steps (compiled service lifecycle, Node build runtime, Node build) wereif: always(), andalways()also runs on cancellation. On this PR, a superseded run kept building after cancellation and heldtest-pr-490, so its replacement stayed pending until I force-cancelled the old run about 10 minutes later. Those steps now useif: ${{ !cancelled() }}, the idiome2ealready uses. They still run after a unit-test failure, but stop on cancellation. Release runs are never cancelled by concurrency, so nothing changes for them; a manual cancel of a publish run now stops promptly.e2e's artifact upload keepsalways()because it is quick.Docs. The header comments in
test.ymlandtypecheck.ymlnow describe this. The sync procedure inAGENTS.mdnow requires sync and release PRs to passunitande2ebefore merge, and mentionsfull-testsfor other risky PRs. The CI line inPLAN-fresh-v2-rewrite.mdis updated too. The dateddocs/sync-*.mdaudits are historical, so they are left unchanged.workflow_call still runs both jobs
In a reusable workflow, the
githubcontext is the caller's.publish.ymlis triggered byworkflow_dispatch, sogithub.event_nameintest.ymlisworkflow_dispatchwhenpublish.ymlcalls it. The gate's first clause,github.event_name != 'pull_request', is therefore true, andunitande2erun unconditionally, just as before. Tag pushes (push) and direct dispatches pass the same clause. The concurrency expression'spull_requestbranch is false there too, so those runs get the unique grouptest-run-<run_id>and nothing cancels them. Only a caller that is itself triggered bypull_requestwould hit the PR gate, and none exists.Verification
actionlint1.7.12 is clean ontest.yml. Its only finding is a pre-existing shellcheck SC2016 info note intypecheck.yml's inlinebun -escript, which this PR does not change. Prettier andgit diff --checkare clean on the changed files.openedrun 43504043883 skipped both jobs.full-testsstarted 43504123544, withunitande2ein progress.Notes
t3/shell-timeout-regression. This PR does not touchpackages/core/test/tool-shell.test.ts.Check runs
Results on head
910bb86(thesynchronizerun withfull-tests), run 43504720265:e2e: pass (20m).unit, attempt 1: failed on one test,ShellTool stopped > stops a command the user stopped, keeps its file and fails the call with the tail, file and shell ID (spawned)(packages/core/test/tool-shell.test.ts:2285). It is a timing race: the killed command's file heldfirst\nbecausesecondhad not been written yet. This is not the known PowerShell case, which did not fail here. It is pre-existing, in the file the in-flight shell fix owns, and unrelated to this PR, so I left it alone.unit, attempt 2 (rerun of the failed job): pass (12m46s).typecheck,hygiene,plugin-browserand Socket all pass.This body edit itself starts a fresh
unit+e2erun, as documented above; its result is in a comment below.