Session page: a late poll answer overwrites a newer one (J5 "withdraw" flake) #932

New issue

Closed

opened 2026-09-29 11:32:00 +00:00 by claude-bot · 1 comment

claude-bot commented

2026-09-29 11:32:00 +00:00

Collaborator

J5 failed on #926's CI (run 11404, desktop, j5-after-the-session.spec.js:402): after Withdraw, The recording still read "Approved"; the phone and keyboard failures that followed were a cascade (shared database, the desktop evening's suggestions left un-added).

Cause. useSessionData.load (SessionPage.jsx) calls setData with whatever useAsync.run returns — and run returns an overtaken answer even though it keeps it out of its own state. The review page polls the session every 15 s while the wiki run is going (it starts at approval). A poll that left before the withdrawal and landed after the page's own refresh put the approved session back until the next poll, past J5's 10 s wait. A real bug, not only a test flake: any GM action during a poll can be visually undone for up to 15 s.

Fix. useSessionData keeps its own newest-read counter; only the newest read may draw the page. Unit test reproduces it (fails without the fix).

J5 failed on #926's CI (run 11404, desktop, `j5-after-the-session.spec.js:402`): after **Withdraw**, The recording still read "Approved"; the phone and keyboard failures that followed were a cascade (shared database, the desktop evening's suggestions left un-added). **Cause.** `useSessionData.load` (SessionPage.jsx) calls `setData` with whatever `useAsync.run` returns — and `run` returns an overtaken answer even though it keeps it out of its own state. The review page polls the session every 15 s while the wiki run is going (it starts at approval). A poll that left before the withdrawal and landed after the page's own refresh put the approved session back until the next poll, past J5's 10 s wait. A real bug, not only a test flake: any GM action during a poll can be visually undone for up to 15 s. **Fix.** `useSessionData` keeps its own newest-read counter; only the newest read may draw the page. Unit test reproduces it (fails without the fix).

rbrooks referenced this issue from a commit

2026-09-29 11:32:13 +00:00

fix(webapp): a late poll answer no longer overwrites a newer one on a session page (#932)

claude-bot referenced this issue from a pull request that will close it,

2026-09-29 12:08:03 +00:00

A late poll answer no longer overwrites a newer one on a session page (#932) #933

claude-bot closed this issue

2026-09-29 12:53:39 +00:00

claude-bot referenced this issue from a commit

2026-09-29 12:53:39 +00:00

Merge PR #933: a late poll answer no longer overwrites a newer one (#932)

claude-bot commented

2026-09-29 12:54:03 +00:00

Author

Collaborator

Fixed in PR #933 (merged as 67b107bf): useSessionData counts its reads, and only the newest draws the page. The new useSessionData.test.jsx reproduces the race (it fails without the fix). CI was green, including the journeys.

Possible follow-up, not done here: useAsync.run still returns overtaken answers to its callers. Any other caller doing run(...).then(setX) on a polled read has the same shape. It wasn't audited here, and a repo-wide change to useAsync semantics wants its own look.

Fixed in PR #933 (merged as 67b107bf): `useSessionData` counts its reads, and only the newest draws the page. The new `useSessionData.test.jsx` reproduces the race (it fails without the fix). CI was green, including the journeys. Possible follow-up, not done here: `useAsync.run` still returns overtaken answers to its callers. Any other caller doing `run(...).then(setX)` on a polled read has the same shape. It wasn't audited here, and a repo-wide change to `useAsync` semantics wants its own look.

claude-bot referenced this issue

2026-09-29 19:11:13 +00:00

useAsync.run hands an overtaken answer back to its caller #941

rbrooks referenced this issue from a commit

2026-09-29 19:19:44 +00:00

fix(ui): useAsync.run hands an overtaken answer back to nobody (#941)

claude-bot referenced this issue

2026-09-29 22:07:07 +00:00

useAsync.run hands an overtaken answer back to nobody (#941) #947