Conversation
At the end of a list iOS rubber-bands past the edge, so the first post-gesture capture lands mid-bounce with every row shifted and read as moved, then the content sprang back to the pre-gesture tree. A change against a baseline that already ended in the scrolled direction is now credited only once the surface holds still and still differs; a bounce that settles back answers at-edge. Closes callstack#2884.
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
All reported issues were addressed across 2 files
Reply with feedback, questions, or to request a fix.
Fix all with cubic | Re-trigger cubic
The rest requirement resolved the edge state without the swipe point, so a nested screen could answer for a scroller the gesture never ran in. The baseline check now passes the swipe midpoint through readScrollEdgeState, the same selection the edge verdict uses, and the poll loop doc names the edge exception.
There was a problem hiding this comment.
All reported issues were addressed across 3 files (changes from recent commits).
Reply with feedback, questions, or to request a fix.
Fix all with cubic | Re-trigger cubic
|
[claude-fable-5-1] responding on behalf of Oskar CI note: Smoke Tests on this head failed at |
|
Reviewed at 510eb60. The code looks right. Not blocking: the inline comment above the readScrollEdgeState(baseline.nodes, edge, { point: swipe.midpoint }) call at https://github.com/callstack/agent-device/blob/510eb60/src/daemon/scroll-movement.ts#L333 still describes the outer/inner list relationship backwards even though the code and the sibling test are correct, and claimMoved's own edge check a few lines down at https://github.com/callstack/agent-device/blob/510eb60/src/daemon/scroll-movement.ts#L446 resolves its container without the swipe midpoint while baselineEndsInDirection resolves with it, so the two calls could in principle pick different containers on a nested-list screen — passing swipe.midpoint through claimMoved's call would make one selection rule govern every edge decision in this module, but both points can be taken or left as they don't change behavior on the cases this PR covers. Smoke Tests was still running at review time with nothing to attribute against this diff; scroll-movement.ts and scroll-edge-state.ts sit on the scroll-command route that suite exercises, so worth a look if it fails, but there's no failure evidence yet. I could not reproduce the author's iPhone 17 Pro / iOS 26.2 live scroll run myself and I'm relying on the PR body's description of it rather than a raw log, and I did not confirm pnpm check:affected independently since CI hadn't finished. The one open item is Smoke Tests finishing green. |
thymikee
added
the
ready-for-human
|
[claude-fable-5-1] responding on behalf of Oskar CI note for |
Summary
Closes #2884.
observeScrollMovementcreditsmovedon the first capture that differs from the pre-gesture baseline. At the end of a list iOS rubber-bands past the edge, so that first capture lands mid-bounce with every row shifted by a few points, reads aschanged, and the content then springs back to exactly the pre-gesture tree. The next command's stabilization sees the surface identical to its baseline (post_gesture_snapshot_stale_accept,matchedPreGestureBaseline: true) while the scroll had already answeredmoved, and an agent scrolling untilat-edgenever stops.The baseline now carries its nodes, and a
changedreading against a baseline that already ended in the scrolled direction (readScrollEdgeState(baseline.nodes, edge): a container with no hidden content) is credited only once the surface holds still and still differs. A bounce that settles back to the baseline takes the existing settled path and answersat-edge; a bounce that never settles expires assurface-unsettled. A scroll whose baseline still hid content keeps paying one read, so the one-capture cost claim holds everywhere except at the edge. Keyed on typed edge state and the shared at-rest comparator, not on diagnostics text. The decision is logged asscroll_movement_edge_rest_required.Touched: 2 source files, 1 test file.
Validation
Tested commit
4549d4006(review follow-up: edge state resolved at the swipe point through readScrollEdgeState's target; nested-list cases pinned).pnpm check:affected --run: passed (an earlier run flagged the poll loop's complexity; the per-reading verdict is now its own function).at-edgein 3 reads; content that changes at the edge and holds still ismovedin 2 reads; a bounce that never settles isunobserved/surface-unsettled. The existing one-readmovedcase (baseline with hidden content) is unchanged.--debug:open com.apple.Preferences,scroll bottom, then twicesnapshot+scroll down 0.75. Both scrolls answeredmovement: at-edgewithscroll_movement_edge_rest_requiredthenscroll_movement_observed attempts: 3in the request log, i.e. the first read differed from the baseline and the surface settled back. On main that first read is themovedanswer the issue reports.