lsm / neokai / 28667520162

82%

Build Type

push

github

Committed by web-flow

Commit Message

fix(web): settle-scroll to bottom on refresh/cold-mount in chat + task threads (#2224)

* fix(web): settle-scroll to bottom on refresh/cold-mount in chat + task threads

On refresh/cold-mount, the single-shot mount-scroll in useAutoScroll lost the
race against the browser's scroll-position restoration and late content/layout
growth (markdown/code highlighting, image loads, banners, composer inset),
leaving the user stranded above the latest messages. Add a bounded settle-scroll
that re-pins to the bottom over a short window on first non-empty content,
cancels when the user intentionally scrolls away, and self-terminates so normal
scroll behavior resumes. Also opt out of browser scroll restoration
(history.scrollRestoration = 'manual') as defense-in-depth.

* fix(web): plug settle/observer leaks and cancel settle on load-older

Address review feedback on the settle-scroll change:
- P1: when containerRef.current is null on first run, the retry-timeout path
  installed listeners/ResizeObserver via setupScrollDetection but discarded its
  cleanup, so they leaked (SpaceTaskUnifiedThread hits this during its loading
  placeholder). Capture the cleanup in a local and invoke it from the outer
  effect cleanup for both the timeout and direct paths.
- P2: cancel the in-flight settle when loadingOlder flips true. A fast Load More
  resolving before the last tail re-pin let that timer see loadingOlderRef=false
  and snap to the bottom, clobbering ChatContainer's restored read position.

Add regression tests for both. Note: messageCount stays in the scroll-detection
deps on purpose — it re-triggers setup when the container mounts late
(SpaceTaskUnifiedThread placeholder); the churn is intentional, not a leak.

Coverage Stats

11537 of 16627 branches covered (69.39%)

Branch coverage included in aggregate %.

50 of 52 new or added lines in 1 file covered. (96.15%)

1 existing line in 1 file now uncovered.

98254 of 116560 relevant lines covered (84.29%)

301.66 hits per line