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
Flaky tests under load: session replay/walk/feather-fall and the GM-line threading test #296
Several tests fail now and then under load and pass on a re-run. Seen on 2026-10-22 during full go test ./... runs (workers run the suite under flock /tmp/ta-test.lock nice -n 10, often with other load on the machine):
Each passed alone and on a re-run. They make a red suite ambiguous for every branch and cost a re-run per worker.
Wanted: find the shared cause (fixed waits such as 2 s deadlines, races between the replay decoder and the live event channel, wall-clock use in tests) and make the tests deterministic, without weakening what they check.
The four session test flakes are fixed in #300 (tests and harness only): the cause was the WSL2 host stepping its wall clock back about 2.2 s every ~30 s. Results under -race: each of the four went from 2-3 failures in 50 runs to 0 in 200. Still open: the GM-line threading test caught a real selection race (a GM line could run on a stale selection); its fix is on branch fix/gm-selection-echo, in review.
Fixed: the flaky session tests in #300, and the GM selection race the threading test caught in #301 (a GM line now runs only on a selection the server confirmed by an echo of the agent's own send, never during the agent's own combat; otherwise it is refused with nothing sent).
Activity
Merge pull request #300 from pjerra/fix/ta296-flaky-tests
pjerra commentedon Oct 23, 2026
The four session test flakes are fixed in #300 (tests and harness only): the cause was the WSL2 host stepping its wall clock back about 2.2 s every ~30 s. Results under -race: each of the four went from 2-3 failures in 50 runs to 0 in 200. Still open: the GM-line threading test caught a real selection race (a GM line could run on a stale selection); its fix is on branch fix/gm-selection-echo, in review.
Selection echo: a combat rewrite of the target field is not a selecti…
Selection echo: no confirmation while the agent attacks, fights or is…
Merge pull request #301 from pjerra/fix/gm-selection-echo
pjerra commentedon Oct 23, 2026
Fixed: the flaky session tests in #300, and the GM selection race the threading test caught in #301 (a GM line now runs only on a selection the server confirmed by an echo of the agent's own send, never during the agent's own combat; otherwise it is refused with nothing sent).