Two sims fail on main: the apply's context alarm never fires #445
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
What happens
Two simulated runs fail on
main(d4ea96df), and nothing says so, because the sims are a tool and not part of the suite:apply-auto-plentifulapply-group-compactsBoth script an apply turn past its context alarm (380,000 of 400,000) and expect
osf.budget.alarmwithfresh_session, then a respawn from Braid's note. No alarm is recorded. The step asks the model for a third turn, which the play does not have, and the run never ends.Either the alarm no longer fires for a continued apply session, or the scenarios describe a path the engine stopped taking (braid#383 changed when a step forks).
Seven more sims failed on
mainfor another reason, a staleplayfield in the default play's report. That one is fixed with braid#443.Fix
Find which it is. If the alarm is broken, fix it with a test. If the scenarios are stale, rewrite them to the path production takes.
Done when
./tools/sim.shpasses every scenario onmain.Fixed in PR 460. The engine was right and the scenarios were stale.
Since braid#407 an apply step starts a session of its own (
openspec.applyhas nofork_session_of), so at its alarm it is compacted in place.apply-group-compactsandapply-auto-plentifulstill expected the older move to a fresh session, with their next turns underrespawns; the engine compacted, sent its continuation, and the play had no third turn. Both plays now continue in the same session and expectosf.budget.alarmwithcompact,osf.budget.compactedand no respawn. The spec's scenario "A group's apply step" said the same stale thing and is corrected.Evidence. Both pass, and
python -m osf.sim --allpasses all 85 scenarios.