Release M275.3#2000
Merged
Merged
Conversation
This commit does four related things:
1. Add an AGENTS.md at the repo root, following the 2025-2026 cross-tool
convention (Linux Foundation Agentic AI Foundation). It is natively read
by GitHub Copilot, Claude Code, OpenAI Codex, Cursor, Gemini CLI,
Windsurf, Aider, Devin, Zed, and others; agents that have their own
native file (e.g. CLAUDE.md) fall back to AGENTS.md when no native file
is present.
The file captures project-specific knowledge that isn't obvious from a
fresh `git clone` and which routinely trips up AI coding assistants:
- The wrapped-output layout (build outputs land one level up from the
working tree -- catches agents reaching for ./out or similar). Notes
the documented `src\` convention and that all commands assume CWD at
the enlistment root.
- Three explicit build paths scoped to scenario, so agents stop reaching
for the full installer build (~5 min, NativeAOT publish + Inno Setup,
no incremental support) when they just want to rerun a unit test:
Path A -- unit-test inner loop: ~10-15 s (dotnet build .csproj)
Path B -- functional-test loop: ~30-60 s (dotnet publish w/
PublishAot=false +
SkipCreateInstaller=true)
Path C -- installer build: ~5 min (Build.bat, only when you
actually need an installer)
Path B explicitly calls out the native C++ vcxproj prerequisite, since
dotnet publish won't build those.
- The NUnit filter footgun: this codebase uses NUnitLite, which supports
only --test (not --where). --where is silently ignored and runs every
test. For unit tests that's just slow; for functional tests it's hours
of wasted runtime against fresh enlistments.
- Fully-qualified-name requirement for --test (short names silently match
nothing).
- The vcpkg caching behavior in Build.bat (don't manually re-run vcpkg
unless overlay ports changed).
Coding standards are NOT duplicated here -- agents are pointed at
CONTRIBUTING.md, which already covers StyleCop, error handling,
TryXxx patterns, logging conventions, etc.
No new precompiled binaries, no helper scripts, no opinionated workflow
recommendations (worktree model etc.) -- just the minimum project-specific
knowledge needed to keep an agent productive in this repo.
2. Remove the "maintenance mode" notice from Readme.md. VFS for Git has
resumed active development; the line about "only required updates as a
reaction to critical security vulnerabilities will prompt a release" is
no longer accurate. The Scalar recommendation paragraph (for new
deployments) is unchanged.
3. Remove the build-badge table from Readme.md. The four pipelines it
referenced are a mix of still-running (gvfs/ci CI - Windows and
CI - Windows - Full Functional Tests, both currently green) and
never-configured for master (mseng/AzureDevOps GVFS/GitHub VFSForGit
Large Repo Perf Tests and Large Repo Build both render the "set up
now" placeholder). Active CI signal lives in the GitHub Actions
checks on each commit/PR; keeping a separately-maintained badge
table just creates a "is this still accurate?" question that no one
answers.
4. Update the "Building VFS for Git" section of Readme.md to reflect the
current build system:
- The .NET 10 SDK has replaced the .NET Core 8 SDK requirement.
- The "Install nuget.exe" prerequisite is gone (modern dotnet SDK
handles restore).
- vcpkg is now required (for native libgit2 et al.); call it out and
point at the official getting-started doc.
- The .NET Core cross-platform development workload and runtime
bullets are obsolete; dropped.
- The build script is `Build.bat`, not the historical
`BuildGVFSForWindows.bat`.
- The installer output path is `out\GVFS.Installers\bin\<Config>\win-x64\`,
not the historical `BuildOutput\GVFS.Installer.Windows\bin\x64\<Config>\`.
- Added a pointer at the end of the section to AGENTS.md for AI
coding assistants.
Assisted-by: Claude Opus 4.7
Signed-off-by: Tyrie Vella <tyrielv@gmail.com>
…ndingUpgradeHandler Add CurrentVersionPath to Configuration that resolves a Current\ junction under the install root, falling back to the flat layout when no junction exists. GVFSLocation is now computed dynamically so junction re-targeting takes effect without restarting the service. Update PendingUpgradeHandler.GetInstalledMountProcesses to recognize mount processes from any Versions\ subdirectory, not just the flat install root. These changes are backward-compatible: on existing flat installs, CurrentVersionPath falls back to AssemblyPath and all behavior is identical. This is groundwork for a versioned install layout (Plan 002) where each version lives in its own directory and a Current junction points to the active version, enabling non-disruptive upgrades. Assisted-by: Claude Opus 4.6 Signed-off-by: Tyrie Vella <tyrielv@gmail.com>
Add AGENTS.md and remove maintenance-mode notice
The GetFileStreamException catch in GetFileStreamHandlerAsyncHandler silently completed the ProjFS command without logging requestMetadata (SHA, virtual path). Other catch blocks in the same handler all log this metadata. Add a RelatedWarning with the HResult value for diagnostic visibility while keeping the severity appropriate for expected/benign HResult.Handle results. Assisted-by: Claude Opus 4.6 Signed-off-by: Tyrie Vella <tyrielv@gmail.com>
At startup, the deferred telemetry attacher only checked the physical console session via WTSGetActiveConsoleSessionId. On Cloud PCs and RDP-only machines, the console session has no logged-in user — the real user is in an RDP session. This meant the startup attach always failed, and the retry timer could never succeed either (SYSTEM has no git global config). A SessionLogon event would fix it, but that only fires for NEW logins — not when the service restarts while a user is already connected. Enumerate all interactive sessions (Active/Connected, session > 0) via WTSEnumerateSessions and try each until the pipe attaches. Assisted-by: Claude Opus 4.6 Signed-off-by: Tyrie Vella <tyrielv@gmail.com>
Service: add versioned install layout support
Fix deferred telemetry pipe attach for RDP sessions
Add diagnostic logging to GetFileStreamException catch block
Worktree mounts shared the primary enlistment's .gvfs metadata (placeholder database, RepoMetadata, diagnostics, corrupt objects) instead of using their own per-worktree .gvfs directory. This caused stale file content after git reset --hard in worktrees, because placeholder tracking was shared with the primary mount. Root cause: code throughout the codebase derived .gvfs paths from EnlistmentRoot, which always resolves to the primary enlistment root even for worktree mounts. The correct property is DotGVFSRoot, which resolves to the per-worktree .gvfs for worktree mounts and the primary .gvfs for primary mounts. Product fixes: - GVFSDatabase: accept dotGVFSRoot directly instead of deriving from EnlistmentRoot; update InProcessMount, SparseVerb, DiskLayoutUpgrade, EnlistmentPathData, ProfilingEnvironment call sites - GitRepo: resolve DotGVFSRoot at construction for corrupt object paths - GVFSVerb.InitializeLocalCacheAndObjectsPaths: use DotGVFSRoot for RepoMetadata (affects dehydrate, prefetch) - DiagnoseVerb: use DotGVFSRoot for .gvfs file collection and RepoMetadata initialization - CacheVerb: use DotGVFSRoot for RepoMetadata - RequiredGitConfig: narrow parameter to GVFSEnlistment, use DotGVFSRoot - GVFSLogsRoot: derive from DotGVFSRoot instead of re-deriving - GitMaintenanceStep: check WorkingDirectoryRoot for existence - HooksInstaller: use WorkingDirectoryRoot in error messages Rename to prevent recurrence: - Rename Enlistment.EnlistmentRoot to PrimaryEnlistmentRoot, forcing every caller to explicitly choose the right path property. Audit of all ~280 references across ~75 files confirmed each usage is correct. Test: - Add WorktreeUsesPerWorktreePlaceholderDatabase functional test Assisted-by: Claude Opus 4.6 Signed-off-by: Tyrie Vella <tyrielv@gmail.com>
fix worktree mounts using primary enlistment's .gvfs paths
KeithIsSleeping
approved these changes
Jun 3, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Changes: