Repository navigation
Conversation
Contributor
There was a problem hiding this comment.
All reported issues were addressed across 11 files
Reply with feedback, questions, or to request a fix.
View guided diff | Turn on auto-fix | Re-trigger cubic
Contributor
🟢 e2e android agent: 20 passed24 agent steps · 15 replayed from cache · 53 model calls · 349.8k tokens (85% cached) · $0.0209 All 20 tests in 1 file
Tested by e2e 0.23.1 · 6m 24s · android-emulator · run artifacts |
Contributor
🟢 e2e ios: 56 passed, 1 skippedAll 57 tests in 7 files
Tested by e2e 0.23.1 · 6m 54s · ios-simulator · run artifacts |
Contributor
🟢 e2e android: 58 passed, 1 skippedAll 59 tests in 7 files
Tested by e2e 0.23.1 · 8m 33s · android-emulator · run artifacts |
Contributor
🟢 e2e ios agent: 21 passed, 1 skipped25 agent steps · 12 replayed from cache · 53 model calls · 325.0k tokens (81% cached) · $0.0226 All 22 tests in 1 file
Tested by e2e 0.23.1 · 8m 14s · ios-simulator · run artifacts |
Collaborator
Author
|
[claude-sonnet-5-5] responding on behalf of kdy1 Left for the human:
|
kdy1
force-pushed
the
kdy1/mobile-gesture-rest-wait
branch
from
November 10, 2026 03:46
1240b61 to
3789edd
Compare
kdy1
force-pushed
the
kdy1/mobile-gesture-rest-wait
branch
from
November 10, 2026 11:54
3789edd to
c255a6c
Compare
kdy1
force-pushed
the
kdy1/mobile-gesture-rest-wait
branch
from
November 13, 2026 08:57
c255a6c to
de86937
Compare
`mobile({ transition })` takes a number, the same wait after every action, or
`{ default, gesture, launch }` (new exported type `TransitionBudgets`). The
engine records the kind of the last action: `gesture` for a root scroll, node
swipe, or coordinate swipeTo, `launch` after a launch, relaunch, close, or
openLink. A scroll can return while its content moves and needs a longer wait.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…moving After a scroll or swipe the `gesture` budget is only the least wait. The engine then reads until two sightings of the control and its look-alikes agree, at most 4 s after the gesture, and that read is also the relocation. `scroll` can return while a fling still moves the list, and a tap aimed at a moving row lands on the row passing under it. A budget of 0 is still no wait, and an action whose timeout cannot fit another read goes ahead on the latest one. The default and launch paths and the agent are unchanged. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A fill or clear after a scroll or swipe now takes the read that found the field at rest as its re-read, so it costs no read of its own and a short timeout does not fail it where a tap goes ahead on the latest read. The checks `relocated` makes for an input move into one method both reads use. The changeset and docs name the cap (4 s, or the `gesture` budget when that is larger) and the timeout exit, and the skill reference lists the link case of `launch`. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The rest wait stopped reading once the action's timeout could not fit another read and its pause, but a long press also needs its hold left, or `InputBudget.send` refuses it unsent. The hold is now part of that exit condition, so a long press after a scroll goes ahead on the latest read. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
kdy1
force-pushed
the
kdy1/mobile-gesture-rest-wait
branch
from
November 14, 2026 05:17
de86937 to
33388e3
Compare
…est-wait Resolves the conflict in packages/mobile/src/surface.ts: this branch's per-kind `transitionWaits` replaces `transitionMs`, beside main's new `interactiveSnapshots` and `hideTouches` fields and their constructor assignments. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
gesturebudget is the least wait, then the engine reads until two sightings of the control and its look-alikes agree, at most 4 s after the gesture. That read is also the relocation read, so a list that already sits still costs no extra read.steadied), which compared the control's own bounds, 100 ms apart, for up to 2 s after the gesture and failed the action withOPERATION_TIMEOUTwhen its budget ran out first. This one compares the control with its look-alikes, 50 ms apart, runs up to 4 s, and goes ahead on the latest read when the budget runs out. perf(mobile): skip the post-gesture stabilization wait for a test's swipe and scroll #1668's opt-out of agent-device's own post-gesture settle (postGestureStabilization: false, feat(client): let callers opt out of post-gesture stabilization callstack/agent-device#4253, released in 0.21.37) stays as it is.scrollcan return while a fling still moves the list. On an iOS simulator with agent-device 0.21.36, 3 of 120 scrolls kept moving 1.66 to 1.84 s after it returned, and the benchmark's 1.5 s sleep sends the next tap about 1.7 s after. In CI the same shape shows assticky chrome: accept below the fold(scrollUntilVisible, then tap): flaky in the final iOS report of 6 of the 101 PRs updated since Oct 15, the most of any test. These numbers are from 0.21.36, where agent-device ignored the opt-out and its own settle ran in full. Main pins 0.21.37 now, which also reworks agent-device's wait for a moving target before a press (feat(interaction): pre-dispatch admission owner for runtime press/click callstack/agent-device#4461). They were not repeated on it.mobile({ transition })also takes{ default, gesture, launch }(newTransitionBudgetstype), so the least wait after a gesture can differ from the wait after a tap. The wait fix(mobile): wait for transitions before screen swipes #1564 puts before a screen swipe follows the same kinds. A number still sets one wait for every action, and 0 still means no wait. The benchmark config keeps 1.5 s.flowchart LR A[scroll or swipe returns] --> B[sleep the gesture budget] B --> C[read the screen] C --> D{two sightings agree?} D -- no --> E{cap or timeout reached?} E -- no --> C D -- yes --> F[act on the control as read] E -- yes --> FVerified
Ran it locally: yes
apps/mobile-benchmarkthrough the built CLI on an iOS simulator and an Android emulator,@e2e-dev/mobilebuilt from main and from this branch, both on the pinned agent-device 0.21.36 (--workers 1 --retries 0, one run at a time). Both builds carried a wrapper around the agent-device client that records every call (not in this PR).scrollUntilVisiblethen tap on the huge virtualized list,--repeat-each 10, two runs per build in the order main, branch, branch, main (main was f6aba97, before fix(mobile): wait for transitions before screen swipes #1564 and perf(mobile): skip the post-gesture stabilization wait for a test's swipe and scroll #1668 landed). That is 840 scroll-then-tap pairs, and all 40 repeats pass. A tap counts as sent on a moving list when the last two snapshots before it disagree about where the rows sit (more than 1 px).Fisher exact, one-sided: p = 0.015. Five of main's six were in one run (the other main run had one), so without that run the contrast alone is 1/210 against 0/420. Slow flings are 16 of 840 scrolls (1.9%). On the branch each waited 2.0 to 2.5 s for two reads to agree, with 3 to 5 reads, and none reached the 4 s cap. The baseline is main before #1668, with agent-device's own settle in place. The table was not repeated against main with #1668's re-read, so it shows what the rest wait adds over that settle, not over
steadied.pnpm checkpass. The previous head (33388e3, the PR's four commits on f64e7db) passed CI with all four device jobs (iOS and Android, deterministic and agentic) green. Locally, the iOS deterministic suite on a simulator with 0.21.37 gave 57 passed, 1 skipped, 1 failed (672 s, 1-minute load 3.5 at start after the 4-minute wait cap). The failure isopenLink opens the scenario a deep link names, new from test(mobile-benchmark): cover device.openLink with a real deep link #1805: the app installed on that simulator is the 11-03 build, whose bundle has noe2e-benchmark://routing, so the link opens and the app ignores it. CI builds or repacks the app from each head's own source. Android was not run locally.Not covered: this branch against main with #1668's re-read on 0.21.37, a swipe that starts a UIKit transition (the tree reports that screen's final frame at once, so only the
gesturebudget protects the tap, as before), and real devices.The scratch test
🤖 Generated with Claude Code