tools: late-push — #224's instrument, detection by CONTENT not SHA

[jobordu]

#224 specified tools/late-push.py and it did not exist.

A squash merge captures every commit in the PR at merge time. It cannot capture one the author has not pushed yet. Measured there: 37 of 100 merged inside 120s of creation, 25 inside 60s, one at 42 seconds. ★ At 42 seconds the author is not late; the merger is early — and MERGED erases the difference.

⛔ Detection by CONTENT, not SHA — #224 is explicit

A squash does not preserve shas, so git merge-base --is-ancestor answers a different question: whether this commit is on main, never whether its content is. Every squashed commit is a non-ancestor, so ancestry alone reports the whole board lost. This compares git patch-id --stable.

The known-positive — #224's own, re-verified before building

PR #339          merged   2026-08-20T19:56:18Z
commit 7447b1d   authored 19:57:18Z          ← 60 seconds later
                 docs/DEFECT-CLASSES.md +10
                 not an ancestor · patch-id absent from 400 commits of main

--pr 339 reports exactly that, exit 1. Ten lines genuinely lost. (⚠ #224 says 61s — author-vs-commit time.)

⛔ Two deviations from #224's stated commands, both honest

1. --since 24h no longer reaches its own specimen. #339 is 16 days and ~250 merges down; the command was written when it was fresh. --pr N keeps the known-positive runnable without a git-log per PR across the board — and --limit is documented as a recency window, not a filter, on every run.

2. --since 1s returns exit 2 VOID, not the 0 #224 expected. With no commits on main in the window the patch-id comparison set is empty, so every candidate would read as LOST. ⇒ Refusing is correct; a clean 0 there would be a lie. The condition asked for a detector that "cannot return both is not one" — it returns three, and the third is the one that matters.

⚠ A finding is a CANDIDATE, not a proven loss

Content that reached main via a conflict resolution or an altered rebase has a different patch-id and reads as lost here. #445 holds: ORPHANED ≠ CONTENT-LOST, and only the first is mechanically decidable. An unreadable head ref is counted and named UNCHECKED, never folded into clean.

Verification

--self-test 5 legs both directions, driven with synthetic state (#402) · suite 7/7 including the empty-patch-id VOID and test_control_can_fail · --help 0 · bogus flag 2 · check-tools-index 0.

⚠ One test failure was my stub, not the tool: patch_id() calls git show with allow_fail=True, so keying the stub on that returned a stat block where a diff was expected and the finding silently vanished. Keyed on --stat instead.

🤖 Generated with Claude Code

[coderabbitai]

Warning

Review limit reached

Next included review available in 24 minutes.

Check out review usage here.

View limit details

Limit details: You’ve used the included review currently available.

You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository.

Learn how review limits work.

Review configuration:

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Team

Run ID: e1b7c78c-6c42-48e2-b500-1e823a3ca28a

📥 Commits

Reviewing files that changed from the base of the PR and between a268c99 and 1ce8bdd.

📒 Files selected for processing (3)

• tools/README.md
• tools/late-push.py
• tools/test_late_push.py

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}