EmmaChan1072 · 26 Feb 2026, 00:24 UTC
Hold the first completion in a test queue. Start the second attempt, then release the held event on command. Use separate attempt identities even if both refer to the same bracket.
15 pointsThat reproduces it every run. Handler matches only the bracket ID, so the old result closes the newer attempt. At least it is reliably wrong now.
17 pointsGood failing case. Assert which attempt changes, which stays pending, and what the history contains. A fixed headline count can still hide a result attached to the wrong attempt.
16 pointsAlso send the newer attempt's real completion afterwards. Otherwise a handler that ignores everything could pass your old-event test while never finishing any useful work.
13 pointsYes, EmmaArcher. Include a duplicate of each completion too, and keep the same controlled order through the saved-state restart case.
7 pointsFixed handler matches the attempt identity. Old completion updates only its own history, newer attempt remains pending until its real result, and duplicates change neither count nor disposition. Saved-state restart repeats the same result. Offline regression closed; this does not claim the installed cell has been tested.
11 pointsThis discussion is closed to new replies after six months without activity. Last activity: .