A `git worktree` never replays a checkout's cache entry for a task that lists the workspace root
維護者通常 1 天內回覆
評估
- 難度
- 3/5
- 預估耗時
- 1-2 天
- 新手友好度
- 35/100
- Issue 類型
- 缺陷
- 描述清晰度
- 描述清楚
- 活躍度
- 停滯
- 技術堆疊
- git, rust
- 領域
- build-system, tooling
研究方向
Start with crates/vt/src/session/execute/fingerprint.rs, especially should_ignore_entry and the directory listing fingerprint path; compare it with the .git filtering in crates/vt/src/session/execute/tracked_accesses.rs. Add an end to end case that changes .git from a directory to a file and confirms the cached task replays. The issue says the fix is complete when root listings no longer cause this cache miss across checkout and worktree layouts.
由索引模型根據 Issue 內容生成。
描述
Written by an AI agent (Claude) on behalf of @jasonkuhrt.
A task that lists the workspace root records each entry's name and kind in its fingerprint, and .git is one of those entries. In an ordinary checkout .git is a directory; in a git worktree it is a file that points at the main repository. So a cache entry recorded in a checkout never replays in a worktree of the same commit, and the reverse, though every file a task can read is the same. The miss reads:
○ cache miss: '.git' added in workspace root, executing
We share one remote cache between CI, which checks out with a .git directory, and developers and agents, who work in git worktrees. In a fresh worktree at our trunk's head (vite-plus 1.4.0), 56 of the 143 vp lint tasks that missed missed for this reason alone, against entries CI recorded for that commit. Linters, Vitest's project discovery and anything that resolves the workspace list the root, so most of our tasks carry the root listing.
Steps to reproduce
repro.sh <path to vt> runs a task that lists the workspace root in a checkout, adds a worktree of the same commit, copies the checkout's cache into it (standing in for a shared or remote cache), and runs the task there.
repro.sh
#!/usr/bin/env bash
set -euo pipefail
vt=$(cd "$(dirname "$1")" && pwd)/$(basename "$1")
work=$(mktemp -d)
cd "$work"
mkdir checkout && cd checkout
git init -q -b main
cat > package.json <<'JSON'
{ "name": "repro", "private": true }
JSON
cat > vite-task.json <<'JSON'
{
"tasks": {
"list-root": {
"command": "node -e \"console.log(require('fs').readdirSync('.').length)\"",
"cache": true
}
}
}
JSON
echo node_modules > .gitignore
git add -A && git -c user.name=r -c user.email=r@r commit -qm init
echo "== checkout: first run"
"$vt" run list-root
echo "== checkout: second run"
"$vt" run list-root
git worktree add -q ../worktree
mkdir -p ../worktree/node_modules/.vite
cp -R node_modules/.vite/task-cache ../worktree/node_modules/.vite/
cd ../worktree
echo "== worktree: same commit, checkout's cache"
"$vt" run list-root
With vt built from 40cb9cac (macOS 27.2 arm64, Node 26.13.0):
== checkout: first run
$ node -e "console.log(require('fs').readdirSync('.').length)"
5
== checkout: second run
$ node -e "console.log(require('fs').readdirSync('.').length)" ◉ cache hit, replaying
5
---
vt run: cache hit, 412ms saved.
== worktree: same commit, checkout's cache
$ node -e "console.log(require('fs').readdirSync('.').length)" ○ cache miss: '.git' added in workspace root, executing
5
Cause
.git paths are already kept out of the fingerprint when a task reads them: tracked_accesses.rs skips every path under .git. A directory listing doesn't go through that filter. should_ignore_entry drops only ., .., .DS_Store and dist, so the root listing keeps .git with its DirEntryKind, Dir in a checkout and File in a worktree.
vite-plus 1.4.0 ships the same code (fingerprint.rs at a3c087a).
Proposed fix
Skip .git in directory listings the way .DS_Store and dist are skipped, which matches what tracked_accesses.rs already does for reads under .git. A PR that does this follows, with an e2e case where a task that lists the root records its entry with .git as a directory and replays it after .git becomes a file.
If you'd rather keep .git's presence in the listing, the alternative is to record .git without its kind, so a directory and a file compare equal. We went with dropping it because a read of .git itself is already dropped, so its presence doesn't count anywhere else.
The miss message
A changed kind is reported as added: determine_folder_change_kind returns InputChangeKind::Added when an entry's name matches and its kind differs. '.git' added in workspace root sent us looking for a new .git, when it had been there all along as a directory. A message such as '.git' changed from a directory to a file in workspace root would have pointed at the cause. That changes the output, so it isn't in the PR.
Related
#504 is about negative input patterns not reaching listing entries. A fix there would let a task exclude .git by configuration, but .git is never a task input, so it needs no configuration.
- 主要語言
- Rust
- 星號
- 480
- 分支
- 50
- 平均合併
- 1 天 5 小時
- 30 天內合併 PR
- 63
環境準備
在瀏覽器裡用你自己的 GitHub 帳號啟動這個專案的開發容器。
- 沒有 Dockerfile 或 Docker Compose 檔案
- 沒有 Pull Request 範本
- 閱讀貢獻指南
從這裡開始
- 先讀完整個 Issue，再讀專案的貢獻指南。
- 在 Issue 下留言說明你要接手 —— 這能避免兩個人做同樣的事。
- Fork 儲存庫，在一個分支上完成修改。
- 送出 Pull Request，並在描述裡引用這個 Issue 編號。
voidzero-dev/vite-task 的其他 Issue
-
難度 5/5 一週以上 新手友好度 35/100
voidzero-dev/vite-task#921 ·
維護者通常 1 天內回覆
-
難度 4/5 3-5 天 新手友好度 56/100
voidzero-dev/vite-task#910 · 1 則留言 ·
維護者通常 1 天內回覆
-
Automatic output tracking misses files a task renames into place, so a cache hit restores nothing未關閉
難度 4/5 3-5 天 新手友好度 48/100
voidzero-dev/vite-task#908 ·
維護者通常 1 天內回覆
-
Export OpenTelemetry client spans for the native remote cache可能已有人在做 @jasonkuhrt 於 1 天前認領。 未關閉
難度 4/5 3-5 天 新手友好度 25/100
voidzero-dev/vite-task#906 ·
維護者通常 1 天內回覆
-
難度 4/5 3-5 天 新手友好度 56/100
voidzero-dev/vite-task#905 ·
維護者通常 1 天內回覆
查看 voidzero-dev/vite-task 的全部 Issue
相似的 Issue
-
難度 2/5 1-3 小時 新手友好度 62/100
calcit-lang/caps#16 ·
-
難度 2/5 1-3 小時 新手友好度 85/100
penso/herdr-gpui#934 ·
維護者通常 1 天內回覆
-
難度 2/5 1-3 小時 新手友好度 76/100
Y-ASLant/ElegantClipboard#178 ·
-
難度 2/5 1-3 小時 新手友好度 82/100
linxiv-dev/linXiv#384 · 1 則留言 ·
維護者通常 1 天內回覆
-
priority middle
難度 2/5 1-3 小時 新手友好度 70/100
KATO-Hiro/AtCoderClans#12945 ·
維護者通常 1 天內回覆