Repository navigation
Conversation
🟢 e2e web: 93 passedAll 93 tests in 42 files
Tested by e2e 0.23.1 · 1m 1s · web · run artifacts |
🟢 e2e web agent: 34 passed35 agent steps · 26 replayed from cache · 25 model calls · 117.4k tokens (87% cached) · $0.0065 All 34 tests in 5 files
Tested by e2e 0.23.1 · 1m 39s · web · run artifacts |
There was a problem hiding this comment.
All reported issues were addressed across 6 files
Reply with feedback, questions, or to request a fix.
View guided diff | Turn on auto-fix | Re-trigger cubic
🟢 e2e ios agent: 21 passed, 1 skipped25 agent steps · 12 replayed from cache · 59 model calls · 385.2k tokens (83% cached) · $0.0249 All 22 tests in 1 file
Tested by e2e 0.23.1 · 8m 33s · ios-simulator · run artifacts |
🟢 e2e android agent: 20 passed24 agent steps · 15 replayed from cache · 33 model calls · 171.8k tokens (77% cached) · $0.0135 All 20 tests in 1 file
Tested by e2e 0.23.1 · 5m 27s · android-emulator · run artifacts |
|
[claude-sonnet-5-5] responding on behalf of kdy1 The red The long-press step of control-inventory also failed once (the 563 ms hold registered as a tap) and passed on rerun. Left for a human: merge #1614, then this PR needs a rebase and one rerun of |
🟢 e2e android: 58 passed, 1 skippedAll 59 tests in 7 files
Tested by e2e 0.23.1 · 8m 41s · android-emulator · run artifacts |
🟢 e2e ios: 56 passed, 1 skippedAll 57 tests in 7 files
Tested by e2e 0.23.1 · 6m 40s · ios-simulator · run artifacts |
34c5e74 to
23dc7e1
Compare
KrzysztofMoch
left a comment
There was a problem hiding this comment.
Note
🤖 gpt-6.1-sol responding on behalf of KrzysztofMoch
Please add a regression test for delayed UI updates after a caret key in an already focused field, and adjust the settle policy so that observation captures the update before merging.
movesCaretOnly() treats the key and field role as enough evidence to shorten the wait, but moving a caret can trigger an asynchronous selectionchange handler. I reproduced this through the built CLI in real Chromium with a focused contenteditable textbox whose native selectionchange handler reveals a toolbar after 700 ms: this head returns the old tree after roughly 270 ms, while main captures the toolbar.
Both pressKey('Home') and press(target, 'Home') miss the update in all three repeats on this head and pass all three on main. The 50 ms control passes on both builds, giving 9/9 passes on main versus 3/9 on this head. This uses a scripted executor with no model calls, isolating the observation timing regression in a controlled browser fixture.
…ield After press or pressKey, a settled look waits up to 2 s for the tree to change shape. A caret key (ArrowLeft, ArrowRight, Home, End, select-all) in a textbox or searchbox that already holds the focus usually changes nothing the tree records, so the whole window was spent. It now waits 1 s (HELD_STILL_MS), enough for a selectionchange handler that updates the tree a beat later, such as a rich-text toolbar. armChange keeps the later deadline of a window still pending, so a tap followed by a caret key in one turn keeps the tap's full window. focusedNode() names the focused node only when exactly one has the focus. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
23dc7e1 to
5fb3354
Compare
|
[claude-sonnet-5-5] responding on behalf of kdy1 Fixed in 5fb3354. The caret window is now 1 s ( |
main moved the agent into @e2e-dev/core, so the two tests import it from @e2e-dev/core/lib/agent/*, and the changeset names the package that now holds settle-policy.ts. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
pressorpressKey, a settled look waits up to 2 s (FULL_CHANGE_WAIT_MS) for the tree to change shape. A caret key (ArrowLeft, ArrowRight, Home, End, Control/Meta+a, with Shift or Control) in atextboxorsearchboxthat already holds the focus usually changes nothing the tree records, so the whole window was spent.HELD_STILL_MS) for those keys:changeWaitAfterinpackages/core/src/agent/settle-policy.ts. The first version used 150 ms; review showed aselectionchangehandler that reveals a toolbar 700 ms later was missed, so the window now covers it.combobox, Alt chords, Tab and the vertical keys, and every other action.armChange(packages/core/src/agent/observation-feed.ts) keeps the later deadline when a window is still pending. Without it,tap(Save)thenpressKey('End')in one turn would cut the tap's 2 s window.focusedNode()returns a node only when exactly one node has the focus.Verified
Ran it locally: yes
tests-agent/replay with--strict-cache, throughpnpm bench:ab --config e2e.agent.config.tsagainst the merge base (e27f492c), 6 pairs in random order.act completes …scenarios) are left out ofbench:abwith--grep-invert, since their model calls differ between runs. They ran once on the 150 ms version of this branch with a key (CI=1 e2e run --config e2e.agent.config.ts --ai-trace, recordings read-only): 33 tests passed, noREPLAY_STALE. On this head CI'sweb agentjob ran them with a key: 33 passed, 20 model calls. The--no-cachepath was not run.act completes rich-text-editortests/integration/agent-caret-settle.test.tshas a fake-engine popup that opens 700 ms afterpressand afterpressKeyon a focused field. Both cases fail at the 150 ms window and pass at 1 s. The same file keeps a first tap, a tap on the focused field, a caret key before the field held the focus, and a batched tap then key at the full window.@e2e-dev/core, refactor: move e2e's core into @e2e-dev/core, and engines peer on it #1737): e2e unit 3707 pass, agent integration 228 pass,pnpm checkexit 0. The timings above were measured before that move, on1112e50aagainste27f492c; they were not re-measured on the merged tree. The iOS and Android agent and deterministic suites passed in CI on an earlier head, sincesettle-policy.tsis shared with mobile.At 150 ms the replay took 59.75 s (-16.1%) and the rich-text step 9.40 s. Taps stay at 2 s for the same reason: a click can open a panel that renders late. Five dead looks of about 2.1 s (10.7 s) remain: right clicks in the context-menu trap, the upload tap, the gift-card amount tap, and the editor's bold tap.
🤖 Generated with Claude Code