Android: a thread with many subagent rows takes 8–13 s to show its messages, and its forks inherit the delay

未关闭
#39,687 4 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
52/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
android, react-native, typescript

调研方向

Start with apps/mobile/src/features/threads/thread-subagent-group.tsx at the reported subscription point, then trace how thread feed rows are rendered and whether full history or many subagent rows affect that path. Reproduce with the described Android thread and profile JavaScript rendering; compare against a thread with few subagent rows. Done means the messages appear promptly on repeated opens and forks, with profiling evidence identifying the delay.

由索引模型根据 Issue 内容生成。

描述

via-triage
What happened

On the Android app, one thread takes 8–13 seconds to show its messages each time I open it. The screen stays blank under the header until then. A fork of that thread has the same delay. Other threads open at once, including a much larger one (1,752 items, about 23 MB) and a fork that inherits 2.2 MB of history.

Diagnosis
  • The server is not the cause. Each open sends orchestration.subscribeThread with a resume sequence. The server replays in about 1 ms, and no getThreadSnapshotWindow runs. All other RPCs in the window finished in under 15 ms.
  • The delay is in the app, between receiving the data and drawing the rows. On one open of the fork, the app subscribed to the parent thread 8 s after the fork. That subscribe comes from ThreadSubagentGroup (apps/mobile/src/features/threads/thread-subagent-group.tsx:120), which mounts only when the feed draws a subagent group row. So no rows were mounted for 8 s.
  • The phone's vcs.listRefs call came within 1 s of the content appearing on every open. It seems to follow the render; it does not cause the wait.
  • Settled state is not the cause. Unsettling the parent made no difference.
  • Size alone is not the cause:
    • Slow: 118 items, 1.1 MB, 24 subagent, 14 dynamic_tool, 23 command_execution, 2 compaction.
    • Fast: 412 items, 2.2 MB, 2 subagent, 43 dynamic_tool, 117 command_execution.
    • Fast: 1,752 items, 23 MB, 5 subagent. This one probably loads as a bounded window.
  • Unproven guess: the cost is in drawing many subagent rows, or in drawing the full history of a thread that is small enough to skip the bounded window. A JS profile from the device would confirm it.
Steps to reproduce
  1. On desktop, use a thread with about 24 subagent items and some large dynamic tool results, small enough to load in full (about 120 items).
  2. Open it in the Android app 2.3.0 over T3 Connect. The messages appear after 8–13 s.
  3. Fork it at a later run and open the fork. It has the same delay.
  4. Open a thread with few subagent rows but more data. It opens at once.

I have no minimal repro yet. The steps above describe the thread I have.

Version

Android app 2.3.0 (Play Store); server 0.1.11-nightly.20261119.4941

Environment

Samsung SM-S928U1 (Android), connected through T3 Connect; server on macOS arm64

Evidence
# open fork (parent settled)
10:32:00.433 subscribeThread fork        (resume, server work ~1 ms)
10:32:00.73-01.05 getThreadLinkMetadata x50, each <15 ms
               (no RPCs from the app for 7 s)
10:32:08.519 subscribeThread parent      <- subagent group row mounts; content visible

# open parent alone, unsettle, open fork, open parent again
11:46:19.891 subscribeThread parent
11:46:27.071 vcs.listRefs                <- content visible (~8 s)
11:46:30.582 dispatchCommand thread.unsettle parent
11:46:37.301 subscribeThread fork
11:46:37.601 subscribeThread parent
11:46:48.021 vcs.listRefs                <- content visible (~11 s)
11:46:51.901 subscribeThread parent
11:47:04.161 vcs.listRefs                <- content visible (~12 s)
Related issues

#34240 "Mobile sync is slow" (closed, no details). That issue is about sync in general. This one is one thread that is slow to draw while the server answers at once.

Fix applied or workaround

None found. Nothing was changed on the machine. The parent was unsettled for the test and then settled again.

Filed by

Claude Code (Claude Opus 5.5, claude-opus-5-5) inside T3 Code, following the t3 triage playbook

主要语言
TypeScript
星标
24.8k
派生
6.4k
平均合并
4 小时 29 分钟
30 天内合并 PR
488

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

pingdotgg/t3code 的其他 Issue

查看 pingdotgg/t3code 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。