github / github/copilot-cli

ANSI escape stripping garbles output at PTY read boundaries

未关闭
#2,103 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看
area:terminal-rendering area:tools
主要语言
Shell
星标
11.2k
派生
1.9k
平均合并
14 小时 16 分钟
30 天内合并 PR
6

描述

## Describe the bug

The bash tool's ANSI-stripping logic garbles output when escape sequences are split across PTY read boundaries. Fragments like `1m`, `]8;;`, or URL text from OSC 8 hyperlinks leak into captured output. The bug is non-deterministic — the same command can produce different garbled lines or sometimes render cleanly.

## Affected version

1.0.6

## Steps to reproduce

Run this in a Copilot CLI session (no dependencies needed — pure bash):

```bash
for d in $(seq -w 4 17); do
printf '\e]8;;https://example.com/2026-03-%s\aDate-%s\e]8;;\a \e[1mA\e[22m \e[36mB\e[39m \e[32mC\e[39m \e[31mD\e[39m\n' "$d" "$d"
done
```

Run multiple times. Some runs will show garbled output on one or more lines, such as:
```
Date-13 A B C D
ttps://example.com/2026-03-14Date-14 A B C D ← URL leaked
Date-15 A B C D
```
or:
```
Date-14]8;; A B C D ← OSC fragment leaked
```

Other runs of the exact same command will be completely clean.

## Expected behavior

All escape sequences should be fully stripped in every run, producing 14 identical lines of `Date-XX A B C D`.

## Additional context

The stripping appears to process bytes as they arrive from each PTY `read()` call. Since reads don't align with writes, an escape sequence can be split across two reads. Neither contains the complete sequence, so the regex fails to match and leaves fragments in the output. The non-determinism (different lines garbled across runs, sometimes clean) confirms this is a read-timing issue, not a regex pattern bug.

贡献指南

打开贡献指南

评估

这个 Issue 还没有评估数据。

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。