Repository navigation
Follow-ups from PR #2048 (bot-thread retry): event-path scan, thread-unreadable notice, owner vs dev-lead replies #2077
Description
Deferred review findings from #2048 (fixes #2046). These were judged valid but out of scope for that PR, so it could merge.
- Run the thread-retry scan on a path that doesn't depend on the comment gate. The scan in
review-one-pr.shis nested in themc_gate_rc == 1branch. A PR blocked only by an unreplied bot review thread is therefore recovered only by the 2-hour cron, not on the review event. Also call the thread scan fromdev-lead-resume.sh. (Codex P1, cubic P2 on feat: implement issue #2046 — dev-lead: unprocessed bot review threads are never retried — review-thread counterpart of #2017 #2048.) - Make
thread-unreadablevisible. A thread with more than 100 comments classifies asthread-unreadable. It is skipped forever: it is never markedexhausted, so it never gets a notice. Treat it like exhaustion and post a one-time human notice. (cubic P2.) - Tell owner replies apart from dev-lead replies. The replied-thread check matches on login only. The maintainer and dev-lead share
don-petry, so an owner's clarifying reply marks the threadrepliedand blocks recovery. Require the dev-lead automation marker on dev-lead replies. (Codex P1; same root cause as dev-lead silently ignores @dev-lead mentions from the maintainer, because the maintainer's account is dev-lead's own identity (self-comment skip) #2027.) - (Low) Concurrent scans can post a duplicate exhaustion notice. It is harmless, since the hidden marker dedups later scans; a claim-before-post would remove it.
Activity
don-petry commentedon Oct 9, 2026
Dev-Lead Implementation Plan — blocked on #2048
Issue: #2077 — Follow-ups from PR #2048 (bot-thread retry)
Scope
None yet. Every item in this issue changes code that only exists on the unmerged PR #2048.
Why this is blocked
CONFLICTING/DIRTYagainstmainand labelleddev-lead:hands-off. The last dev-lead rebase attempt was aborted for a human decision:main(dev-lead fix-reviews reads reviewThreads(first:50) with no pagination — open threads past the first 50 are invisible, so the pass does nothing #2056) has its ownort_fetch_open_threads, which overlapsscripts/lib/open-review-threads.shandscripts/dev-lead-fix-reviews.sh.main(42dc38d), none of the targets exist:scripts/lib/bot-thread-retry.shis missing, and so istests/dev-lead/unit/test_bot_thread_retry.bats.review-one-pr.shhas only the bot-comment scan (scan_pr_for_undispositioned_bot_comments) insidemc_gate_rc == 1. There is no thread scan to move.thread-unreadable/exhausted/repliedclassifications and the exhaustion notice are all feat: implement issue #2046 — dev-lead: unprocessed bot review threads are never retried — review-thread counterpart of #2017 #2048 code.main, so the PR would carry ~880 unrelated commits.Proposed path
main(dev-lead fix-reviews reads reviewThreads(first:50) with no pagination — open threads past the first 50 are invisible, so the pass does nothing #2056) and merges it.mc_gate_rc == 1branch inreview-one-pr.sh, and also call it fromdev-lead-resume.sh.thread-unreadablelike exhaustion and post a one-time human notice, deduped by the hidden marker.don-petrylogin no longer marks the threadreplied.tests/dev-lead/unit/test_bot_thread_retry.batsfor each item, failing first.No files were changed in this run.