[Bug]: Desktop backend dies with SIGTRAP in fff's background scan thread (realloc trap)

未关闭
#38,975 1 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
45/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
electron, typescript
领域
backend, desktop

调研方向

Start in the apps/server area named in the report and trace how workspace file search uses @ff-labs/fff-node and its native @ff-labs/fff-bin-darwin-arm64 binding during directory scans. The crash reports point to realloc on an fff background scan thread, but there is no deterministic repro or specific source file or test named. Investigate the native allocation or binding behavior and define a reproducible test that confirms indexing no longer crashes the backend.

由索引模型根据 Issue 内容生成。

描述

Before submitting
  • I searched existing issues and did not find a duplicate.
  • I included enough detail to reproduce or investigate the problem.
Area

apps/server

Steps to reproduce

No deterministic repro yet. What happened:

  1. Run the desktop app on macOS (Apple Silicon) with several projects open and agents running.
  2. About an hour after launch, the bundled backend child was killed by SIGTRAP three times in six minutes (23:53, 23:54 and 23:59 local). Each time the app restarted it within a second.

All three crash reports have the same shape: the faulting thread is one of @ff-labs/fff's own threads (fff-bg-2, fff-bg-6, and an unnamed fff scan thread), inside realloc, while other fff threads are still listing directories (getattrlistbulk). The JavaScript main thread is in ordinary V8 code each time; nothing is calling finder.destroy() at that moment.

Expected behavior

File search indexing never takes the backend down.

Actual behavior

The backend process dies with EXC_BREAKPOINT (SIGTRAP). macOS shows " quit unexpectedly" while the window stays up, and any agent turn running at that moment is cut off; providers resume in a fresh native session.

The trap is raised by Electron's allocator from a realloc that libfff_c.dylib makes on its own background thread, so it looks like heap damage inside fff (or its Node binding) that Electron's allocator catches. Because the backend runs inside the Electron binary (ELECTRON_RUN_AS_NODE), malloc/realloc go through Electron's allocator, which may be why this shows up on desktop.

Impact

Major degradation or frequent failure

Version or commit

0.1.10 (main as of 2026-11-18). This is an internal build of T3 Code with the desktop app renamed. apps/server/src/workspace has only upstream commits (it lacks #38337 and #38585, which do not touch native memory), and @ff-labs/fff-node / @ff-labs/fff-bin-darwin-arm64 are 0.11.0, the same as main.

Environment

macOS 26.6.2, arm64, desktop app, Electron 44.10.1, @ff-labs/fff-node 0.11.0

Logs or stack traces
# desktop log (server-child.log)
backend child process failure output end  details="pid=2198 killed by SIGTRAP"   05:53:33Z
backend child process failure output end  details="pid=21068 killed by SIGTRAP"  05:54:22Z
backend child process failure output end  details="pid=27689 killed by SIGTRAP"  05:59:12Z

# crash report (.ips), process = the backend child, parent = the app
exception: EXC_BREAKPOINT (SIGTRAP), codes 0x1, 0x...3c4
termination: Trace/BPT trap: 5

# faulting thread fff-bg-6 (same in the other two reports)
0-4  Electron Framework   (allocator trap; symbols resolve to unrelated names)
5    Electron Framework
6    Electron Framework
7    libsystem_malloc.dylib  _malloc_zone_realloc
8    libsystem_malloc.dylib  _realloc
9-23 libfff_c.dylib          (no symbols)
24   libsystem_pthread.dylib _pthread_start

# other threads at the same moment
fff-bg-0..8   libfff_c.dylib, waiting on a condition variable (except the faulting one)
4-7 threads   libsystem_kernel getattrlistbulk < libfff_c.dylib   (directory scan in progress)
main thread   Electron Framework / V8 only, no fff or ffi-rs frames
Workaround

None found. The backend restarts by itself, but running agent turns are interrupted each time.

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。