Improve resiliency of git worktree commands by tyrielv · Pull Request #1943 · microsoft/VFSForGit

tyrielv · 2026-04-13T15:25:16Z

Summary

MountNewWorktree() in the post-command hook silently discarded the exit code of gvfs mount, leaving worktrees unprojected when concurrent adds caused mount contention.

Review Guide

3 files changed — read in this order:

1. `GVFS.Hooks/Program.cs` (smallest change)

Added GetHookExitCode() — parses --exit_code=N from hook args. Generic infrastructure any post-hook handler can use. Placed alongside GetHookType and GetGitCommand.

2. `GVFS.Hooks/Program.Worktree.cs` (product fix)

RunWorktreePostCommand — now checks exit code. Skips add/move post-processing if git itself failed; always runs remove cleanup.
TryMountWithRetry (new) — retries gvfs mount 2× with 100ms/250ms delays. First attempt shows console output; retries are quiet. Returns bool.
MountNewWorktree — captures git checkout exit code (was discarded). On checkout failure, emits warning with recovery command and skips mount. On mount failure, emits warning.
MountMovedWorktree — same TryMountWithRetry treatment.

3. `WorktreeTests.cs` (test improvements)

Adaptive scaling — starts at Environment.ProcessorCount worktrees (min 4). If concurrent adds overwhelm the primary mount ("does not appear to be mounted"), halves count and retries.
Pipe-based mount assertion — probes worktree-specific named pipe directly instead of File.Exists. Definitive signal that ProjFS is active.
Barrier synchronization — CountdownEvent ensures all threads launch simultaneously.
Diagnostics on failure — captures directory listing, .git contents on assertion failure.

Verification

Tested locally with 16 concurrent worktrees (Environment.ProcessorCount on dev machine):

Scenario	Runs	Failures	Rate
Without product fix	5	5	100%
With product fix	5	0	0%

Note: On CI runners with fewer cores (4), the lower concurrency may not trigger the race. The test still validates correctness of concurrent worktree operations and the adaptive scaling logic — the product fix is independently justified by code inspection (silent exit code discard).

CI: all 45 checks passed.

MountNewWorktree() in the post-command hook silently discarded the exit code of 'gvfs mount', leaving worktrees unprojected when concurrent adds caused mount contention. Also discarded the git checkout exit code. Product fix: - Check git checkout exit code; skip mount on failure - Check gvfs mount exit code with retry (2 retries, 100ms/250ms) - Extract and check --exit_code from hook args to skip post-processing when the git command itself failed - Return int? from GetHookExitCode so callers decide null semantics - Apply same TryMountWithRetry to MountMovedWorktree - Remount old worktree when 'git worktree move' fails (pre-hook unmounts) - Remove maxRetries parameter; retry count driven by delay array length - Emit actionable warning to stderr on mount failure with recovery command Test fix: - Increase concurrent worktrees from 2 to ProcessorCount for reliable mount contention - Use CountdownEvent barrier for tight synchronization of launches - Add pipe-based mount verification as primary assertion (probes the worktree-specific named pipe directly instead of relying on File.Exists) - Add diagnostic capture on failure (directory listing, .git contents) - Use dynamic branch names with GUID suffixes to avoid collisions - Use numeric indices instead of char arithmetic for labels/paths to support >26 concurrent worktrees - Only retry at lower concurrency when ALL failures are overload-related; mixed failure sets (overload + real regression) are reported immediately Bug: https://dev.azure.com/microsoft/OS/_workitems/edit/61784115 Assisted-by: Claude Opus 4.6 Signed-off-by: Tyrie Vella <tyrielv@gmail.com>

tyrielv force-pushed the tyrielv/fix-concurrent-worktree-mount branch 3 times, most recently from 3cb3366 to 2d9b375 Compare April 13, 2026 19:08

tyrielv changed the title ~~GVFS: fix silent mount failure on concurrent worktree add~~ Improve resiliency of git worktree commands Apr 13, 2026

tyrielv marked this pull request as ready for review April 13, 2026 21:51

tyrielv requested a review from KeithIsSleeping April 13, 2026 21:51