fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (residual scroll jump-back root fix) by allenliang2022 · Pull Request #5392 · nesquena/hermes-webui

allenliang2022 · 2026-07-02T02:22:02Z

Follow-up to #5338 (shipped in v0.51.797). That PR routed the sync anchor-realign write and the async postProcess reflow through overflow-anchor suppression, but a residual mobile scroll jump-back remained on a real device, captured mid-stream.

Root cause (confirmed with real on-device data)

I instrumented the running instance with a scrollTop flight-recorder and captured the jumps on the actual phone. The decisive signal: every captured jump (dTop -101 / +350 / +748 / -400) had a call stack of the rAF sampler ONLY — no JS render function on the stack. The jumps are not our JS writing scrollTop; they are the browser's native scroll-anchoring engine re-compensating scrollTop in the layout phase whenever above-viewport content changes height:

virtual-scroll topPad spacer recompute
worklog live to settled collapse
the STREAM_DONE multi-render sequence (renderMessages fires several times back-to-back)
CSS max-height collapse/expand animations on worklog rows — .activity-body (.34s), .tool-group-body (.3s), .tool-card-detail (.26s) — the dominant driver during streaming

Because that compensation runs in the browser's layout step it is independent of which frame our JS wrote scrollTop in. A single-rAF _fixMobileScrollJank released before the collapse/reflow landed, and even a fixed deferred window (~150ms) lifts mid-animation for the 260–340ms CSS transitions — so the remaining frames of each animation still jump.

Fix

_fixMobileScrollJank now does two things:

Defers its restore — each call re-arms suppression and cancels any pending release (clearTimeout + cancelAnimationFrame), so a burst of renders shares one window with a 400ms settle floor for non-animated churn (virtual topPad recompute, image-decode, katex measure).
Tracks CSS animations — binds transitionrun/transitionstart + transitionend/transitioncancel for max-height on #messages (bubbled from descendant worklog rows). An animation start holds suppression (cancels the pending release); an animation end schedules a short 90ms settle after the last one. Hard-capped at 1200ms so a looping transition can't pin overflow-anchor:none forever.

Desktop rests at overflow-anchor:none, so _browserOverflowAnchorActive() returns false and the whole guard is a no-op there.

Verification (end-to-end, isolated debug instance, real 500-message session, Playwright)

step	scenario	result
reproduce	bare `auto`, above-viewport -400px height change	scrollTop jumped -400
fixed (instant)	`auto` + armed guard, same -400px change	0
fixed (animated)	real 340ms CSS max-height animation	overflow-anchor stays `none` through 160/250/340ms (old fixed window released at ~153ms), restores after
no-regression	desktop computed `none`	guard no-op (inline untouched, normal scroll works)

Tests

Updates the bug: WebUI is unusable on Android - transcript jumps to top on every interaction (regression in v0.51.576) #4856 source-string test that keyed on the old single-rAF cleanup body to match the deferred-release structure (same "clears only none" invariant).
Adds test_fix_mobile_scroll_jank_defers_release_across_height_churn (deferred re-arm/cancel + settle floor) and test_fix_mobile_scroll_jank_tracks_css_max_height_animations (transitionrun/transitionend max-height extender + hard cap) — both base-fail on the old single-rAF body / head-pass on the new one.

static/ui.js + the one test file only. @nesquena

greptile-apps · 2026-07-02T02:26:46Z

Greptile Summary

This PR extends the overflow-anchor suppression guard introduced in #5338 to cover the full height-churn window on mobile. The previous single-rAF restore lifted suppression before CSS max-height collapse/expand animations (up to 340ms on worklog rows) and the STREAM_DONE multi-render burst had finished, allowing the browser's native scroll-anchoring engine to re-compensate scrollTop in the layout phase.

Deferred, re-armable release: each _fixMobileScrollJank() call cancels any pending rAF/timer and re-arms the window, so consecutive renders in a burst share one suppression window (two rAF hops + a 400ms settle floor).
CSS animation tracking: _bindMobileAnchorTransitionExtender binds transitionrun/transitionend for max-height on #messages, holding suppression for the full animation duration and scheduling a short 90ms post-transition settle.
Independent hard-cap timer: _mobileAnchorMaxHoldTimer is separately re-armed on every fix() call and is never cancelled by onRun, so a missed transitionend cannot pin overflow-anchor:none indefinitely.

Confidence Score: 5/5

Safe to merge; the change is mobile-only (desktop is an unconditional no-op) and every release path converges through the single _liftMobileAnchorSuppression helper.

The state machine is internally consistent: all three release paths route through _liftMobileAnchorSuppression which resets all shared state atomically. The independent hard-cap timer is correctly re-armed on every call and never suppressed by onRun. Behavioral node-harness tests are mutation-verified against both gate-cert defects.

No files require special attention.

Important Files Changed

Filename	Overview
static/ui.js	Replaces the single-rAF overflow-anchor restore with a deferred, re-armable suppression window that tracks CSS max-height animations; logic is sound, state machine is consistent, and all release paths converge through _liftMobileAnchorSuppression.
tests/test_issue4856_android_scroll_regression.py	Updates existing cleanup-invariant test to target the new shared _liftMobileAnchorSuppression helper; adds two source-string tests and two mutation-verified behavioral harness tests covering re-arm extension and hard-cap release on missed transitionend.

_{Reviews (3): Last reviewed commit: "fix(chat): make re-arm reachable + hard ..." | Re-trigger Greptile}

greptile-apps · 2026-07-02T02:26:50Z

+  if(_mobileAnchorSuppressReleaseTimer){ clearTimeout(_mobileAnchorSuppressReleaseTimer); _mobileAnchorSuppressReleaseTimer=null; }
+  if(_mobileAnchorSuppressRafId&&typeof cancelAnimationFrame==='function'){ cancelAnimationFrame(_mobileAnchorSuppressRafId); }
+  _mobileAnchorSuppressRafId=0;
+  const rafHop=(cb)=>{ if(typeof requestAnimationFrame==='function') return requestAnimationFrame(cb); return setTimeout(cb,16); };


The rafHop fallback returns a setTimeout ID (a number) and stores it in _mobileAnchorSuppressRafId, but the re-arm code unconditionally calls cancelAnimationFrame(_mobileAnchorSuppressRafId) to cancel it. cancelAnimationFrame ignores setTimeout IDs, so the pending 16ms timer is never actually cancelled in the fallback path — consecutive calls won't properly extend the suppression window there. Real mobile browsers always have requestAnimationFrame, so this path is never taken in production, but a parallel clearTimeout call would close the gap and avoid a future maintainability trap.

Suggested change

const rafHop=(cb)=>{ if(typeof requestAnimationFrame==='function') return requestAnimationFrame(cb); return setTimeout(cb,16); };

let _rafHopIsTimeout=false;

const rafHop=(cb)=>{ if(typeof requestAnimationFrame==='function') return requestAnimationFrame(cb); _rafHopIsTimeout=true; return setTimeout(cb,16); };

… to kill the residual scroll jump-back Follow-up to nesquena#5338 (shipped in v0.51.797). That PR routed the sync anchor-realign write and the async postProcess reflow through overflow-anchor suppression, but a residual mobile scroll jump-back remained on a real device, captured mid-stream. ## Root cause (real on-device data) A scrollTop flight-recorder on the running instance captured the jumps on the actual phone. Decisive signal: every captured jump (dTop -101/+350/+748/-400) had a call stack of the rAF sampler ONLY -- no JS render function on the stack. The jumps are not our JS writing scrollTop; they are the browser's native scroll-anchoring engine re-compensating scrollTop in the LAYOUT phase whenever above-viewport content changes height: virtual-scroll topPad recompute, worklog live->settled collapse, the STREAM_DONE multi-render sequence, and -- dominant during streaming -- CSS max-height collapse/expand animations on worklog rows (.activity-body max-height .34s, .tool-group-body .3s, .tool-card-detail .26s). Because that compensation runs in the browser's layout step it is independent of which frame our JS wrote scrollTop, so the previous single-rAF _fixMobileScrollJank released before the collapse/reflow landed and every per-write / single-frame suppression missed it. ## Fix _fixMobileScrollJank now (1) DEFERS its restore -- each call re-arms and cancels any pending release so a burst of renders (STREAM_DONE fires renderMessages several times back-to-back plus a deferred postProcess reflow) shares one suppression window with a 400ms settle floor for non-animated churn; and (2) TRACKS CSS animations -- it binds transitionrun/transitionend for max-height on #messages, so an animation start holds suppression (cancels the pending release) and an animation end schedules a short 90ms settle after the last one. Hard-capped at 1200ms so a looping transition can't pin overflow-anchor:none forever. Desktop rests at overflow-anchor:none so _browserOverflowAnchorActive() returns false and the whole guard is a no-op. ## Verification (isolated debug instance, real 500-message session, Playwright) - Reproduce: bare auto, above-viewport -400px height change -> scrollTop jumped -400. - Fixed: same change with the guard armed -> 0. - Real 340ms CSS max-height animation: overflow-anchor stays none through 160/250/340ms sample points (the old fixed window released at ~153ms) and restores after the animation ends. - No-regression: desktop computed none -> guard no-op, inline untouched, normal scrolling works. ## Tests Updates the nesquena#4856 source-string test that keyed on the old single-rAF cleanup body to match the deferred-release structure, and adds two behavioral source-invariant tests (deferred re-arm/cancel + settle floor; transitionrun/transitionend max-height extender + hard cap), both base-fails on the old body / head-passes on the new one. static/ui.js + the one test file only.

nesquena-hermes · 2026-07-02T03:04:20Z

🔬 Gate certification — RED ⛔ (2 SILENT defects that undercut the fix's own goal — dead re-arm + missed-transitionend leak)

Certified head: sha:737e4967 (rebased onto current master, git apply clean) · PR: #5392 · allenliang2022, fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (residual scroll jump-back root fix)
Verdict: The approach is right (defer the overflow-anchor release across the full render/collapse/reflow window, extend on consecutive renders, hard 1.2s cap) and consistent with the #5338 review. But the full gate found 2 real SILENT defects: the re-arm is dead code (the guard bails before it on repeated calls, so consecutive renders don't extend the window — defeating the PR's own goal), and a missed transitionend leaves overflow-anchor:none pinned (no independent max-hold release timer). Fix both, then it's a visible fix → Nathan screenshot.

What I ran (rebased worktree `/tmp/wt-rebase-5392`)

Gate	Result
Rebase onto current master	✅ `git apply` clean; node -c OK
Codex (reproduce)	SHIP-WITH-FIXES — 2 SILENT; I confirmed the first by inspection
Full pytest suite	1 failed / 11747 passed — the 1 failure (`test_issue1567_nous_picker…falls_back_to_static_4`) is a pre-existing env-dependent flake (fails isolated too; Nous Portal model-fetch env, unrelated to this ui.js-only change)
PR's own test	✅ 20/20 (but they don't exercise the repeated-call re-arm or missed-transitionend paths — which is why they're green while the defects exist)

Findings

⛔ SILENT (I CONFIRMED) — the re-arm/extend is dead code on repeated calls (static/ui.js:12324): _fixMobileScrollJank early-returns if(!_browserOverflowAnchorActive(el)) return;, and _browserOverflowAnchorActive reads the computed overflow-anchor (getComputedStyle(el).overflowAnchor==='auto'). After the FIRST call sets inline overflow-anchor:none, the computed value is none → the predicate returns false → every subsequent back-to-back render (the STREAM_DONE settle fires renderMessages several times + a deferred postProcess reflow) bails out before the clear/reschedule re-arm logic. So the "consecutive renders EXTEND the suppression window" behavior — the PR's core "full height-churn window" fix — never actually happens; suppression only lasts the first call's base window.

Fix (Codex-exact): allow already-owned inline suppression through the guard — const alreadySuppressed = el.style.overflowAnchor==='none'; if(!alreadySuppressed && !_browserOverflowAnchorActive(el)) return; — then the existing clear/reschedule runs. Add a runtime test that calls the helper twice and asserts a cancel/reschedule occurs.

⛔ SILENT — missed transitionend pins overflow-anchor:none (regresses #5338's mobile resting-auto contract) (static/ui.js:12290): onRun cancels the pending base release when a max-height transition starts, relying on transitionend/transitioncancel to reschedule. But the _MOBILE_ANCHOR_MAX_HOLD_MS=1200 is only a guard check inside onRun (whether to cancel), NOT an independent release timer. If transitionend/transitioncancel is missed (interrupted animation, element removed mid-transition), nothing restores → overflow-anchor:none stays inline indefinitely, regressing #5338's mobile-rests-at-auto contract.

Fix: add a separate max-hold release timer, armed on every suppression window, NOT canceled by onRun, restoring only if el.style.overflowAnchor==='none'; clear it from every normal release path.

Recommendation to the next agent

RED — gate-fail/changes-requested (2 SILENT fixes): (1) let already-owned inline suppression through the guard so the re-arm/extend actually works; (2) add an independent max-hold release timer so a missed transitionend can't pin overflow-anchor:none. The approach + consistency with #5338 (_browserOverflowAnchorActive predicate, guarded restore, single bound listener) are good — these fixes make the intended behavior real and leak-safe. Once green, it's a VISIBLE mobile-scroll fix → Nathan screenshot/live-drive (follow-on to #5338, also parked for his sign-off). (The 1 suite failure is a pre-existing Nous-picker env flake, not this PR.) concept 4/5 (#4856 residual scroll-jump-back root fix). Author @allenliang2022 (T1). crit=3.

Gate-certifier layer (warm-up → gate → release). I do not merge/tag/deploy. Rebased onto current master; re-arm-dead-code confirmed by inspecting _browserOverflowAnchorActive (reads computed → false after inline none), missed-transitionend leak is real (1200ms is a guard not a release timer). Cert valid for sha:737e4967.

…ate-cert defects) Addresses two SILENT defects the gate-cert caught: 1. Re-arm was dead code. _fixMobileScrollJank opened with if(!_browserOverflowAnchorActive(el)) return; and that predicate reads the COMPUTED overflow-anchor. The first call sets inline 'none', flipping computed to 'none', so the 2nd..Nth call of a STREAM_DONE burst returned before the re-arm logic — the 'consecutive renders extend the window' behavior never happened, collapsing to a single first-call window. Now: treat an inline 'none' WE set as still-armed (alreadySuppressed) so re-arm runs; desktop has computed 'none' with EMPTY inline so it stays a no-op. 2. The hard cap was only a guard clause inside onRun, not a release. A missed transitionend (interrupted animation / detached element) pinned overflow-anchor:none forever, breaking the nesquena#5338 resting-'auto' contract. Now an independent _mobileAnchorMaxHoldTimer, armed every call and cancelled ONLY by an actual lift (never by re-arm or onRun), force-lifts at the cap. All release paths route through a shared _liftMobileAnchorSuppression(). Tests: replaces the source-string assertions with a behavioral node harness that executes the REAL function against a mock #messages + fake timers and asserts the time-delta / failure-path behavior: - test_behavioral_rearm_extends_suppression_window: a 2nd call at t=200 must keep anchor 'none' at t=500 (dead-code re-arm releases ~432ms after call nesquena#1). - test_behavioral_hard_cap_releases_when_transitionend_missed: transitionrun with no transitionend must still restore 'auto' by the cap. Both mutation-verified: reverting either fix flips the matching test to FAIL. Also re-points test_raf_cleanup_checks_none to the extracted _liftMobileAnchorSuppression helper (avoids orphaning it). static/ui.js + the test file only.

allenliang2022 · 2026-07-02T03:26:43Z

Both gate-cert defects fixed in 22b4b8da — thank you, both were real and both were verification failures on my side, not just code.

1. Re-arm was dead code. You are exactly right: the guard if(!_browserOverflowAnchorActive(el)) return; reads the computed value, and the first call's inline overflow-anchor:none flips computed to none, so the 2nd..Nth call of a STREAM_DONE burst returned before the re-arm logic ever ran — the whole "consecutive renders extend the window" behavior collapsed to a single first-call window. Fixed by treating an inline none we set as still-armed:

const alreadySuppressed = el.style.overflowAnchor===none;
if(!alreadySuppressed && !_browserOverflowAnchorActive(el)) return;

Desktop rests at computed none with EMPTY inline, so alreadySuppressed is false there and it stays a verified no-op.

2. The hard cap was a guard clause, not a release. _MOBILE_ANCHOR_MAX_HOLD_MS was only read inside onRun; a missed transitionend (interrupted animation / detached element) pinned overflow-anchor:none indefinitely, violating the #5338 resting-auto contract. Fixed with an independent _mobileAnchorMaxHoldTimer, armed on every call and cancelled ONLY by an actual lift (never by re-arm or onRun), so it force-lifts at the cap regardless. All release paths now route through a shared _liftMobileAnchorSuppression().

On the tests — the deeper problem was that my assertions checked absolute state on the happy path, which could not distinguish "re-arm works" from "re-arm is dead but the first call already set none." Replaced them with a behavioral node harness that executes the REAL function against a mock #messages + fake timers and asserts the time-delta / failure path:

test_behavioral_rearm_extends_suppression_window: a 2nd call at t=200 must keep the anchor none at t=500 (with the dead-code bug it releases ~432ms after call Portability #1).
test_behavioral_hard_cap_releases_when_transitionend_missed: transitionrun with no transitionend must still restore auto by the cap.

Both are mutation-verified: reverting either fix flips the matching test to FAIL (I checked by reverting each hunk and re-running). Also re-pointed test_raf_cleanup_checks_none to the extracted _liftMobileAnchorSuppression helper so it isn't orphaned. static/ui.js + the test file only; CI matrix green locally.

nesquena-hermes · 2026-07-02T04:11:12Z

🔬 Gate certification — GREEN (engineering) ✅ · CONVERGED (bounce closed) · ⏸️ visible → Nathan screenshot

Certified head: sha:9fd34989 (clean rebase, branch gate-rebase/5392-mobile-anchor-defer) · PR: #5392 · allenliang2022, fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (#4856)
Verdict: Bounce closed — both my prior SILENT defects are fixed (verified): the re-arm is now reachable on repeated calls, and there's an independent max-hold release timer so a missed transitionend can't pin overflow-anchor:none. Codex SAFE, suite green (one unrelated pre-existing flake). Engineering GREEN; as a visible mobile-scroll fix (follow-on to #5338) it now needs Nathan's screenshot/live-drive sign-off.

What I ran (rebased worktree `/tmp/wt-rebase-5392b`)

Gate	Result
Rebase onto current master	✅ `git apply` clean; node -c OK
Codex (reproduce)	SAFE TO SHIP — gated the rebased worktree, 0 findings
Full pytest suite	1 failed / 11749 passed — the 1 failure (`test_issue1567_nous_picker…falls_back_to_static_4`) is the SAME pre-existing env flake as round 1 (fails isolated; Nous Portal model-fetch env, unrelated to this ui.js-only change)
PR's own test	✅ 22/22 (grew 20→22: added re-arm + max-hold coverage)

Findings — both prior SILENT closed (verified)

Finding 1 (re-arm was dead code) → FIXED: the guard is now const alreadySuppressed=el.style.overflowAnchor==='none'; if(!alreadySuppressed && !_browserOverflowAnchorActive(el)) return; (ui.js:12350-51). Reading the inline value (own-set) instead of only the computed predicate means a repeated call while already-suppressed passes the guard and reaches the clear/reschedule re-arm — so consecutive renders now genuinely EXTEND the window (the PR's core goal). Comment credits the gate feedback.
Finding 2 (missed-transitionend leak) → FIXED: a new independent _mobileAnchorMaxHoldTimer (ui.js:12364) armed on every suppression window — setTimeout(() => _liftMobileAnchorSuppression(el), _MOBILE_ANCHOR_MAX_HOLD_MS=1200) that "NOTHING cancels except an actual lift", so a missed transitionend/transitioncancel can't pin overflow-anchor:none past 1.2s. Release logic was consolidated into _liftMobileAnchorSuppression which clears ALL timers (base + max-hold + rAF) — no double-fire, no orphaned timers. Guarded restore (only clears if still 'none').
Routes through the same _browserOverflowAnchorActive predicate as fix(chat): suppress browser overflow-anchor during JS scroll-anchor realign (mobile scroll jump-back) #5338 (desktop none = no-op); single bound transition listener. Codex SAFE, suite green (bar the env flake).

Recommendation to the next agent

Engineering-GREEN — merge from branch gate-rebase/5392-mobile-anchor-defer (sha:9fd34989), NOT the PR's stale head 22b4b8da — AFTER Nathan's visible sign-off. Both SILENT defects closed via a clean centralized-release refactor (_liftMobileAnchorSuppression + independent hard-cap timer); Codex SAFE + 22/22 tests + suite green (the 1 failure is a pre-existing nous-picker env flake, not this PR). Because it's a visible mobile-scroll motion fix (residual jump-back root fix, follow-on to #5338), it's not autonomous: live-drive the jump-back scenario on a mobile viewport + screenshot/GIF for Nathan (pair with #5338's sign-off). concept 4/5 (#4856 root fix). Credit @allenliang2022 (T1, co-authored). crit=3.

Gate-certifier layer (warm-up → gate → release). I do not merge/tag/deploy. Rebased onto current master; both SILENT closed (re-arm reachable via inline-value guard + independent max-hold release timer, verified by reading the centralized _liftMobileAnchorSuppression), Codex SAFE + 22/22 + suite green (bar the pre-existing nous env flake). Engineering green; visible → Nathan sign-off. Cert valid for sha:9fd34989.

Release — mobile transcript jump-back root fix (#5392, residual #4856)

nesquena-hermes · 2026-07-02T19:50:47Z

Shipped in v0.51.829 📱 — thanks @allenliang2022!

Merged via release #5440, with Nathan's sign-off (mobile motion fix — proof is the Codex repro + the 22-test regression file, since a still can't show absence-of-jump).

Full gate (rebased onto current master v0.51.828):

Codex regression gate: SAFE TO SHIP — mobile-scoped, desktop no-op, hard-cap timer only cleared by an actual lift; verified the interaction with fix(#5367): preserve transparent stream live rows on rerender #5400 (transparent-stream reconcile, also ui.js, shipped v0.51.827): unchanged from master, interaction only through the shared #messages scroll path — both coexist intact.
Full pytest suite green (11800 passed). PR-own tests 22/22 (grew from 20 with re-arm + max-hold coverage). streaming-scroll-hardening 7/7. Browser-smoke + ESLint runtime + node -c CLEAN.
Both prior gate defects closed + code-verified: the re-arm is reachable on repeated calls (reads the inline own-set overflow-anchor), and the independent hard-cap timer (1.2s) can't be pinned by a missed transitionend.

This is a root-layer fix — it targets the browser's layout-phase scroll-anchoring compensation the mobile flight-recorder pinned, not our JS scroll writes.

Credit preserved via Co-authored-by. Addresses #4856.

… + independent hard cap) nesquena#4856 Clean rebase of allenliang2022's nesquena#5392 (rebase-first). Co-authored-by: allenliang2022 <allenliang2022@users.noreply.github.com>

…height-churn window (nesquena#4856) [allenliang2022]

…esquena#4856 jump-back)

allenliang2022 changed the title ~~fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (往回大跳 root fix)~~ fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (residual scroll jump-back root fix) Jul 2, 2026

greptile-apps Bot reviewed Jul 2, 2026

View reviewed changes

allenliang2022 force-pushed the fix/mobile-overflow-anchor-defer-release branch from cb1e8cf to 737e496 Compare July 2, 2026 02:39

nesquena-hermes added gate-fail Gate found blocking issue(s); fix-spec in comment; awaiting fix/re-push changes-requested Maintainer left detailed feedback requesting changes; PR is waiting on author to address labels Jul 2, 2026

nesquena-hermes added size:L Large PR (>10 files or >250 LOC) and removed changes-requested Maintainer left detailed feedback requesting changes; PR is waiting on author to address labels Jul 2, 2026

nesquena-hermes mentioned this pull request Jul 2, 2026

Release — mobile transcript jump-back root fix (#5392, residual #4856) #5440

Merged

nesquena-hermes added a commit that referenced this pull request Jul 2, 2026

Merge pull request #5440 from nesquena/release/stage-5392

7cddabb

Release — mobile transcript jump-back root fix (#5392, residual #4856)

nesquena-hermes closed this Jul 2, 2026

rzyns pushed a commit to hermegeddon/hermes-webui that referenced this pull request Jul 2, 2026

Merge nesquena#5392: defer mobile overflow-anchor suppression across …

2839c16

…height-churn window (nesquena#4856) [allenliang2022]

rzyns pushed a commit to hermegeddon/hermes-webui that referenced this pull request Jul 2, 2026

docs(changelog): nesquena#5392 mobile overflow-anchor defer (residual n…

75891ea

…esquena#4856 jump-back)

allenliang2022 mentioned this pull request Jul 3, 2026

fix(chat): recover viewport anchor via sessionIdx when the content-derived key goes stale (desktop scroll jump-back) #5469

Closed

nesquena-hermes mentioned this pull request Jul 3, 2026

Release v0.51.841 — desktop scroll jump-back fix (#5469) #5485

Merged

allenliang2022 mentioned this pull request Jul 5, 2026

Scroll jump-back: renderMessages() wipe-and-rebuild collapses scrollHeight → browser clamps scrollTop (amplified by transcript virtualization #500) #5636

Closed

Navigation Menu

Uh oh!

fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (residual scroll jump-back root fix) - #5392

fix(chat): defer mobile overflow-anchor suppression across the full height-churn window (residual scroll jump-back root fix)#5392
allenliang2022 wants to merge 2 commits into
nesquena:masterfrom
allenliang2022:fix/mobile-overflow-anchor-defer-release

allenliang2022 commented Jul 2, 2026 •

edited

Loading

Uh oh!

greptile-apps Bot commented Jul 2, 2026 •

edited

Loading

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

greptile-apps Bot Jul 2, 2026

Uh oh!

Uh oh!

nesquena-hermes commented Jul 2, 2026

Uh oh!

allenliang2022 commented Jul 2, 2026

Uh oh!

nesquena-hermes commented Jul 2, 2026

Uh oh!

nesquena-hermes commented Jul 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	const rafHop=(cb)=>{ if(typeof requestAnimationFrame==='function') return requestAnimationFrame(cb); return setTimeout(cb,16); };
	let _rafHopIsTimeout=false;
	const rafHop=(cb)=>{ if(typeof requestAnimationFrame==='function') return requestAnimationFrame(cb); _rafHopIsTimeout=true; return setTimeout(cb,16); };

Uh oh!

Conversation

allenliang2022 commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Root cause (confirmed with real on-device data)

Fix

Verification (end-to-end, isolated debug instance, real 500-message session, Playwright)

Tests

Uh oh!

greptile-apps Bot commented Jul 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Greptile Summary

Confidence Score: 5/5

Important Files Changed

Uh oh!

greptile-apps Bot Jul 2, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

nesquena-hermes commented Jul 2, 2026

🔬 Gate certification — RED ⛔ (2 SILENT defects that undercut the fix's own goal — dead re-arm + missed-transitionend leak)

What I ran (rebased worktree /tmp/wt-rebase-5392)

Findings

Recommendation to the next agent

Uh oh!

allenliang2022 commented Jul 2, 2026

Uh oh!

nesquena-hermes commented Jul 2, 2026

🔬 Gate certification — GREEN (engineering) ✅ · CONVERGED (bounce closed) · ⏸️ visible → Nathan screenshot

What I ran (rebased worktree /tmp/wt-rebase-5392b)

Findings — both prior SILENT closed (verified)

Recommendation to the next agent

Uh oh!

nesquena-hermes commented Jul 2, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

allenliang2022 commented Jul 2, 2026 •

edited

Loading

greptile-apps Bot commented Jul 2, 2026 •

edited

Loading

What I ran (rebased worktree `/tmp/wt-rebase-5392`)

What I ran (rebased worktree `/tmp/wt-rebase-5392b`)