fix(mobile): stop long iOS threads from jumping while scrolling up

[t3dotgg]

Scrolling up through a long thread on iOS was jumpy to the point of feeling broken: content juddered as you dragged through history, and the list could yank you back to the bottom seconds after opening a thread.

Both problems live in @legendapp/list 3.2.0's handling of estimated vs. measured row sizes. Our rows are estimated at 180px but really measure anywhere from 49px (turn folds) to 4000px+ (long markdown messages), so:

1. As each history row mounted during scroll-up, 3.2.0 issued an individual scroll compensation per row — traced bursts of ~10 adjusts in 20ms with deltas like +3974/-131, each racing the native momentum scroll. That's the judder.
2. The initial scroll-to-end never registered as finished (content size kept changing as rows measured), so its retry loop kept firing seconds later — traced it teleporting the viewport 7,600px back to the end mid-drag.

Fix: upgrade mobile to @legendapp/list 3.3.3, which batches row measurements into a single layout pass and fixes the stale end-scroll retries, and rebase our transparent-header inset patch onto it. 19 of 20 hunks applied cleanly; the one reject (onScrollBeginDrag forwarding) exists natively in 3.3.3, so only the didUserDrag flag was folded into its handler. Web keeps its own unpatched 3.2.0.

Verified in the iOS simulator against a clone of real thread data (119-message thread): scroll offset traces went from ±7,600px mid-drag reversals to zero discontinuities, and the judder is gone. Mobile thread tests pass; mobile and web typecheck clean.

---

Change authored by Claude Fable 5 via Claude Code.

🤖 Generated with Claude Code

---

Note

Medium Risk
Touches core list scrolling, initial positioning, and MVCP compensation in a patched dependency used for thread UI; regressions could affect scroll position, keyboard insets, and header blur on iOS, though scope is mobile-only and web keeps 3.2.0.

Overview
Upgrades mobile @legendapp/list from 3.2.0 to 3.3.3 and retargets the pnpm patch to patches/@legendapp__list@3.3.3.patch, addressing iOS thread scroll judder and late snap-to-bottom on long, variably-sized message rows.

The rebased patch layers app-specific behavior on 3.3.3: contentInsetStartAdjustment for transparent-header / automatic UIKit insets, keyboard KeyboardAwareLegendList inset compensation props, contentInsetEndAdjustment on native (not web-only), an inset-end settle watchdog that holds first paint until scroll-to-end stabilizes, safer maintain-at-end when tail sizes are unknown, and Reanimated layout transitions gated off during MVCP adjustments and active scrolling. onScrollBeginDrag now sets didUserDrag for the watchdog. Lockfile/workspace still wire web to unpatched 3.2.0 while mobile uses patched 3.3.3.

^{Reviewed by Cursor Bugbot for commit 7a92d6b. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Fix iOS thread list jumping while scrolling up by upgrading @legendapp/list

Upgrades @legendapp/list from 3.2.0 to 3.3.3 and updates the workspace patch to match. The new version fixes scroll position jumps in long threads by suppressing layout transitions in ReanimatedPositionView during MVCP adjustments and native scroll activity.

• Adds onScrollBeginDrag handler to track user drag state and prepare edge handling
• Defers setInitialRenderState readiness when insetEndRevealHold is active
• Defers maintain-scroll-at-end logic when anchor item sizes are unknown
• Records lastMVCPAdjustTime to coordinate transition suppression in animated rows
• Behavioral Change: initial render is now gated on insetEndRevealHold state, which may delay first paint in some scroll configurations

^{Macroscope summarized 7a92d6b.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 383288d0-18a0-4b43-9f8e-b3661ac3c839

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR updates a third-party library patch to fix iOS scrolling behavior, but an unresolved review comment identifies that lastNativeScrollTime is checked but never assigned, meaning part of the animation-suppression logic may not function as intended.

^{You can customize Macroscope's approvability policy. Learn more.}

Dismissing prior approval to re-evaluate 7a92d6b

[t3dotgg]

[claude-fable-5] RESPONDING ON BEHALF OF THEO

Theo reported jank persisted on this branch (his recording). Root-caused and pushed a second commit (7a92d6b).

The remaining jank was not the library scroll math — it was our own itemLayoutAnimation (LinearTransition.duration(180)) animating repositions it should never touch. Two cases, confirmed by correlating per-frame scroll traces with instrumented requestAdjust calls in the simulator:

1. MVCP compensation shifts. When a history row measures different from its estimate, the list shifts row positions and compensates with an equal contentOffset change — net visual movement should be zero. The layout transition animated the reposition half, so every estimated-to-measured correction became a visible lurch-and-settle (traced deltas up to ±1100px mid-drag).
2. Container reuse. Containers get reassigned to far-away rows during fast scrolls even with recycleItems off. Animating the container tween from the old row's position sends content sliding across the viewport.

Fix extends the patch: the transition is skipped during MVCP adjustments, active scrolling (via the core's existing lastNativeScrollTime), and any container reassignment. At-rest shifts (streaming text, fold toggles) still animate.

Verified in the sim on the same 119-message thread with the same flick pattern: frame-level analysis went from 6 lurch events per pass to 0 after the fix. 📹 after recording

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 7a92d6b. Configure here.}