You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
#403 minors: config file mode not checked (symlink/submodule stalls queue), race wording, cap tests #413
(both lenses) A change to only the config's file type is not caught (fork-ci/actions/head-check/head_check.py:470-474).
The not-a-file check asks cat-file -t, which says blob for a symlink and for a submodule entry pointing at an existing blob. The "unchanged" skip compares only the object id.
A symlink config, or a submodule entry with the config's id, passes. The gate keeps working, but the queue's config.load (pr-queue/config.py:146) then fails, and the queue stops until an owner fixes it by hand. This cannot merge a head with fewer lenses.
Fix: read the entry with git ls-tree, allow only modes 100644/100755, compare mode + id in the skip, and test all three shapes. RELEASE-NOTES.md:24 overclaims until then.
The two-PR race is only narrowed, not closed. The edited trigger does not fire when the base moves. "Require branches to be up to date" closes it. Fix the release notes and README wording.
RELEASE-NOTES.md:39: in checkout mode, re-running a pull_request job reuses the old merge commit. A base fixed by hand needs a new event (push, reopen, title edit). Say so.
Test gaps (each mutation stays green):
the cap values at their boundaries (size 32 KiB, glob length 128, glob total 2048);
a merge-tree failure with exit code 2 or higher that still prints output (refused one step later today);
merge.directoryRenames=true.
Optional: a timeout on the git calls (merge-tree now runs on every gate run; bounded today by the 15-minute job timeout).
Activity
head-check: config file mode + #403 follow-ups (#413) (#419)
head-check: config file mode + #403 follow-ups (#413) (#419)
head-check: config file mode + #403 follow-ups (#413) (#419)