Repository navigation
lane-worktrees: follow-ups deferred from the Opus review of #179 (run.sh stray-process scan, deadline flake, gh pr list cost) #184
Description
Deferred under the close-out direction of 2026-10-06 from the Opus adversarial review of PR #179 (head 7ac5cbe). The one pre-ready fold-in (the non-UTF-8 _latin1_clone test needs the APFS skip) goes into #179 itself.
| # | class | file:line | finding | fix |
|---|---|---|---|---|
| 1 | wrong kill | tests/run.sh:180-186, 266-273 |
G14's stray scan kills a caller's unrelated job that forked during the run (reproduced under a pty: "cousin child … KILLED by run.sh"; alive at the base) | stop only processes carrying this run's mark (TMPDIR=$RUN_ROOT/tmp) or descendants of the suite |
| 2 | flake | tests/run.sh:150, 203 |
the "10 s" deadline is 50 × sleep 0.2 + a ps -A; on Eagle that is ~25 s and test_a_leader_that_ignores_term_is_killed_at_the_deadline failed at the head |
a wall-clock deadline (SECONDS+10) |
| 4 | perf | lane-worktrees:986, 1013 |
E5 now makes one gh pr list per branch (~0.65 s each); #175's gate runs --branches on every lane-end |
one bulk list first; --head only for the misses |
| 5-7 | G14 misses a leftover re-parented to $PPID; the C5 seam skips the real lsof fallback so it is untested; the NOT WRITTEN fallback prints \udce9 for a non-UTF-8 path |
Lane: openRepoTools-3
Activity
brettheap commentedon Oct 8, 2026
One more
tests/run.shitem for this issue's stray-process list, from the triad work under #186 (lane openRepoTools-1; no change to run.sh's census in that work, by the coordinator's direction):The workstation census never reaches zero on Eagle, so
run.shcan wait for ever while its own lock is free.wait_for_pytestcounts every^python3 -m pyteston the machine, not only this repository's suites, and the xFactory worktrees keep 3–20 alive continuously: lane openRepoTools-2 sampled it 101 times over 22 minutes on 2026-10-07 and never saw zero, and lane 1's scoped runs on 2026-10-07 waited over 45 minutes behind 4–7 xFactory pytest processes before being stopped, with/tmp/openrepotools-pytest.lockfree throughout. The interim rule the coordinator set is to stop one's own waiter after 5 minutes and run run.sh's exact pytest line underflockon the same lock file, recording the deviation; the decision on narrowing the census (for example to this repository's own runs, which the lock already serializes) is this issue's owner's.Lane: openRepoTools-1
T018 (c): the daily report counts each leg of an assembly as a checko…
Date the T010 record against #192's and #190's landings, correct its …