Skip to content

Fix main's CI: canonical eval destination, atomic fake records - #1316

Merged
JacobStephens2 merged 1 commit into
mainfrom
fix-main-ci-races
Oct 25, 2026
Merged

JacobStephens2 merged 1 commit into
mainfrom
fix-main-ci-races

Conversation

@JacobStephens2

Copy link
Copy Markdown
Owner

Fixes the last two failures in main's push CI. #1301 fixed the third.

  • macOS: evals::repository::tests::complete_trial_repository_preserves_reviewed_patch_inputs_and_custom_branches compared the canonical prepared repository with the /var/... temp path. macOS resolves /var to /private/var. The fixture's destination is now canonical.
  • Linux (flaky): experiments::stop_and_resume_preserve_finished_trials_and_refuse_a_changed_bar read claude-calls.json while it was empty. The fakes' append_record truncated the file before writing it. It now writes a sibling file and renames it over the record, so a reader, or a fake killed by stop, never sees a partial list. Every fake (gh, claude, codex, muse, agy, grok, opencode) uses this writer.

Why main keeps getting these failures: #1315.

scripts/affected-tests main: 3004 tests passed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LSAzXiA6BD9neiPgVQRMKD

The eval repository test compared the canonical prepared path with the
/var temp path, which differs on macOS. Fakes truncated their record file
before writing it, so an unlocked reader, or a fake stopped mid-write,
could see it empty; they now write beside it and rename.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LSAzXiA6BD9neiPgVQRMKD
@JacobStephens2 JacobStephens2 added the test:macos On a PR into main: run the full 4-shard macOS suite instead of the one-job macOS gate (#893) label Oct 25, 2026
@JacobStephens2
JacobStephens2 merged commit 6ecf071 into main Oct 25, 2026
21 of 28 checks passed
@JacobStephens2
JacobStephens2 deleted the fix-main-ci-races branch October 25, 2026 12:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

test:macos On a PR into main: run the full 4-shard macOS suite instead of the one-job macOS gate (#893)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant