You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@maka.apache.org by "orangeCatDeveloper (via GitHub)" <gi...@apache.org> on 2026/09/06 00:40:56 UTC
[I] bug(desktop): "Return to latest" is pulled back by the reading anchor; in-flight history loads drop later requests [maka]
orangeCatDeveloper opened a new issue, #4882:
URL: https://github.com/apache/maka/issues/4882
### What happened
Two defects in Desktop transcript history navigation. Both are user-visible, and together they are the `transcript-scroll-cost.spec.ts:239` flake that has been failing `Desktop e2e` on `main` (run 33962670601) and on unrelated PR branches since the spec landed in #4796.
**1. "Return to latest" is pulled back to the Turn the reader just left.**
The tail range does arrive, and is immediately replaced by the head range again. `ChatView` pins the scroller to the tail only after `await props.returnToLatest.onClick()` resolves, but the tail batch reaches the store before that promise settles. In that window the scroller is still unpinned, so its reading-anchor report writes the top Turn back, and the `restoreRange` effect in `app-shell.tsx` (which re-runs on every `messages` change) finds that Turn missing from the new range and calls `loadAround(anchor)`. In 60/60 local runs the tail was resident for 1–243 ms before being swapped back. The test only passes when Playwright's poll happens to land inside that window, which a loaded CI runner misses more often.
For a user: scroll far back into history, click **返回最新消息** — the transcript flashes the tail and jumps back to where they were, with the notice still showing.
**2. A history load in flight silently drops the requests behind it.**
`loadTranscriptHistory` in `app-shell.tsx` guards with `historyLoadPendingRef` and `return`s when a load is pending, without queueing or retrying. The scroller asks for history on every reader movement, so one wheel gesture fires several `earlier` requests; when the in-flight one settles, nothing re-asks, and paging stops until the reader moves again. Same shape for `latest`: clicking the notice while an `earlier` load is settling is a dead click.
For a user: fast wheel scrolling through old history occasionally stalls one page short.
### How to reproduce
Defect 1 reproduces on `main` 411512bd9 under load:
```
cd apps/desktop
npx playwright test e2e/transcript-scroll-cost.spec.ts -g "paging back" \
--config e2e/playwright.config.ts --repeat-each=20 --workers=4
# 3 failed / 17 passed locally (macOS); every run swaps tail -> head, the
# passing ones only polled inside the window.
```
Instrumented timeline of one failing run (renderer console + main-process replica log, ms):
```
666603 test clicks 返回最新消息
666659 loadTranscriptHistory('latest') starts range hasNewer=true durableThrough=239 oldest=0 newest=19
666676 reading anchor set: turn-prompt-rail-1/0 <- scroller report, unpinned
666735 batch hasNewer=false frags=20 ids 220..239 <- tail resident
666749 restoreRange -> controller.loadAround(0) <- anchor not in range
666761 loadLatest resolves -> ChatView pinToTail -> anchor cleared (too late)
666811 batch hasNewer=true frags=20 ids 0..19 <- back at the head
...30 s later: expect(turn-prompt-rail-120).toHaveCount(1) times out
```
Defect 2 showed up once the first was fixed (1/60): ten consecutive `earlier` requests dropped by the guard during one wheel gesture, the in-flight load settled, no further load, `expect.poll(turns.first()).not.toBe(...)` timed out at `transcript-scroll-cost.spec.ts:263`.
### Environment
- Maka commit: `411512bd9` (`main`, 2026-09-05)
- OS: macOS (Darwin 25.5) for the local repro; CI failures on ubuntu runners
- Surface: Desktop
- Node.js: v26.8.1
### Logs, screenshots, or additional context
- CI failures: https://github.com/apache/maka/actions/runs/33962670601 (`main`), https://github.com/apache/maka/actions/runs/33969566059
- #4560 rebuilds this area and clears the anchor in `app-shell.tsx` before the load; locally that ordering still loses the race (the scroller re-reports the anchor 17 ms after the load starts, see timeline). Pinning before the load is what closes it.
- The guard in defect 2 is a useful throttle for `earlier` (2979 drops in 20 runs); it only needs to remember the last request, not go away.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [I] bug(desktop): "Return to latest" is pulled back by the reading anchor; in-flight history loads drop later requests [maka]
Posted by "orangeCatDeveloper (via GitHub)" <gi...@apache.org>.
orangeCatDeveloper commented on issue #4882:
URL: https://github.com/apache/maka/issues/4882#issuecomment-5556026380
Cross-Session repro for the second half of the title (in-flight history loads dropping later requests), fixed by 76692daa8 on #4883:
1. Session A: scroll up to trigger "load earlier" (`loadBefore` in flight).
2. While it is in flight, switch to Session B. The transcript controller is rebuilt for B, but the pending-load gate was one shell-wide object, so B stayed `pending`.
3. Click "返回最新消息" on B. The request is queued behind A's load and the click appears dead (users click again).
4. A's load settles. The settle path reads A's captured `isCurrent()` — false on B — so the queue is cleared without replaying. B's navigation is gone.
A stale A result could also land on B: the replay used A's captured `controller`/`isCurrent`/`setPending`/`onError`, not B's.
#4883's second commit keys the gate per controller (`WeakMap<object, TranscriptHistoryGate>`): each Session's in-flight load and queued request stay to themselves, same-Session queue semantics unchanged. Covered in `apps/desktop/src/main/__tests__/app-shell-session-ui-state.test.ts` ("starts another Session history load without waiting behind an in-flight load elsewhere" and the six sibling cases).
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org
Re: [I] bug(desktop): "Return to latest" is pulled back by the reading anchor; in-flight history loads drop later requests [maka]
Posted by "Astro-Han (via GitHub)" <gi...@apache.org>.
Astro-Han closed issue #4882: bug(desktop): "Return to latest" is pulled back by the reading anchor; in-flight history loads drop later requests
URL: https://github.com/apache/maka/issues/4882
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@maka.apache.org
For additional commands, e-mail: commits-help@maka.apache.org