Scrolling a long thread stutters after loading earlier turns (timeline measure pass and per-scroll message re-derivation)
メンテナーはふだん 1 日以内に返信
まだ誰も着手していません。
評価
- 難易度
- 4/5
- 見積もり時間
- 3〜5日
- 初心者へのやさしさ
- 45/100
- issue の種類
- バグ
- 明瞭さ
- 明確に書かれている
- 活発さ
- 活発
- 技術スタック
- react, typescript
- 領域
- frontend, performance
調査の方向性
Start by locating useTimelineFollow and the legend-list measure pass identified in the CPU trace (W$ in the generated bundle); the issue gives no source file paths, so search the repository for those entry points. Profile scrolling after repeatedly loading earlier turns and inspect the layout effects and row derivation they trigger. Done means the long-thread scroll no longer repeatedly measures every mounted container or rebuilds unchanged message rows, with the initial-page behavior preserved.
索引モデルが issue の本文から書いたものです。
説明
What happened
Scrolling the chat timeline of a long thread (about 3,250 items, 318 turns) stutters on the Windows desktop app. The jank appears after "Load earlier turns" has been used. At the initial page the same thread scrolls smoothly.
Diagnosis
Repro in the web client served by the thread's own server, headless Chromium at 1600x1000, with a scripted 8 s rAF scroll sweep on .messages-timeline-scroll:
| State | Scroll height | DOM nodes | p50 | p99 | Max frame | Frames >50 ms |
|---|---|---|---|---|---|---|
| Initial page | 3,812 px | 2,053 | 16.7 ms | 17.2 ms | 67 ms | 1 |
| After 8x "Load earlier turns" | ~197,000 px | ~3,000 | 16.7 ms | 83 ms | 200 ms | 9 |
The DOM stays near 3,000 nodes because the list is virtualized, so the cost scales with the amount of loaded history rather than with the DOM.
The CPU trace for the loaded state (8 s of scrolling) shows:
- 14,980 forced style and layout flushes totalling 1.13 s, plus 640 ms of self time in
getBoundingClientRect. All of it comes from the legend-list measure pass (r0->W$in main-*.js), which runs from layout effects during commits and measures every mounted container. InW$,getBoundingClientRectis followed byoffsetWidth/offsetHeightand sometimesgetComputedStyle. - Scroll-time work in useTimelineFollow-*.js that rebuilds message rows:
formatMessagetakes 218 ms,encodeabout 116 ms, andlt/Q/ytin session-logic run underLp/gh/Dh. This re-derivation runs while scrolling, even though the thread is idle and nothing new is arriving. - Long Animation Frame attribution names
DOMWebSocket.onmessageas the entry point for most long frames (2.45 s), because the runtime picks up work queued behind socket messages. The handler itself takes under 20 ms; the cost is the rendering work above.
Rendering itself is small: about 300 ms in the render phase and 190 ms in paint.
Caveat: headless Chromium rasterizes in software, so absolute frame times are higher than in the desktop app. The layout and JS costs above are CPU-side and carry over.
Steps to reproduce
- Open a thread with a few thousand timeline items.
- Click "Load earlier turns" until a few hundred thousand px of history is loaded.
- Scroll up and down with the wheel or trackpad.
Version
Desktop app 0.1.11-nightly.20261119.4954 (Windows). The thread lives on a linked Linux environment that runs 0.1.11-nightly.20261119.4941, and the headless repro used that server's web client.
Environment
Windows 11 Pro 26300, Ryzen 7 5800X, 128 GB RAM, RTX 3080 Ti, 60 Hz. The machine was idle during the repro (no process above about one core of 16, GPU 3D under 4%).
Evidence
Local scroll traces, LoAF summaries and the CPU trace are available on request.
Workaround
Switch away from the thread and back; it reopens at the first page, which scrolls smoothly.
Related
- #13925: the same symptom on the Android app, which has a separate native feed. That report says web and desktop are smooth, which no longer holds once older turns are loaded.
- #12689: Windows desktop renderer stalls from synchronous topology IPC. This repro ran in a browser with no
desktopBridge, so that path is excluded here and may add to it on desktop. This desktop'sdesktop.trace.ndjsonshows 4,111sendSynccalls in 8.3 h, with handlers at a 1.6 ms maximum. - #37639: a forced layout from equations when a thread opens; a different trigger.
- #22350: iOS LegendList
scrollBycorrections; this one is about scroll position, not CPU.
Agent: Claude Code (Opus 5.5), following the t3 triage playbook.
- 主要言語
- TypeScript
- スター
- 24.8k
- フォーク
- 6.4k
- 平均マージ
- 4時間 29分
- マージ済み PR(30日)
- 488
環境構築
このプロジェクトの開発コンテナを、あなたの GitHub アカウントでブラウザ上に起動します。
- Dockerfile・Docker Compose ファイルなし
- プルリクエストのテンプレートあり
- コントリビューションガイドを読む
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
pingdotgg/t3code のほかの issue
-
難易度 2/5 1〜3時間 初心者へのやさしさ 75/100
メンテナーはふだん 1 日以内に返信
-
needs more info
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
pingdotgg/t3code#39400 · コメント 1 件 ·
メンテナーはふだん 1 日以内に返信
-
[Bug]: Wrapped PR titles lose bottom padding in the thread details panel対応中かも @iamdatvt94 が 1 日前に担当しました。 オープンbug needs-triage
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
メンテナーはふだん 1 日以内に返信
-
bug needs-triage
難易度 2/5 1〜3時間 初心者へのやさしさ 66/100
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 73/100
メンテナーはふだん 1 日以内に返信
pingdotgg/t3code の issue をすべて見る
似ている issue
-
area:plugins kind:bug priority:p2 ready
難易度 2/5 1〜3時間 初心者へのやさしさ 90/100
メンテナーはふだん 1 日以内に返信
-
Escaped `*` followed by `*` (`\**`) never matches対応中かも @bardrop が今日担当しました。 オープン
難易度 2/5 1〜3時間 初心者へのやさしさ 71/100
editorconfig/editorconfig-core-js#172 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 82/100
vega/vega-webgpu#40 ·
-
type: bug 🐛
難易度 2/5 1〜3時間 初心者へのやさしさ 63/100
puckeditor/puck#1929 ·
メンテナーはふだん 1 日以内に返信
-
compression() compresses every response at Brotli quality 11対応中かも @staylor が今日担当しました。 オープン
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
メンテナーはふだん 1 日以内に返信