perf(snapshot): add mutex lock, incremental add, and batched revert

[johnnyeric]

Port snapshot optimizations from upstream OpenCode to reduce git process spawning and prevent concurrent corruption.

Relates to #8379

Port snapshot optimizations from upstream OpenCode to reduce git process spawning and prevent concurrent corruption:

• Per-gitdir mutex lock (upstream #17878, v1.3.0): serializes concurrent snapshot operations to prevent race conditions
• Incremental git add (upstream #17878, v1.3.0): replaces naive git add . with diff-files + ls-files + size filtering, skipping files >2MB
• Batched revert (upstream #20564, v1.3.14): groups up to 100 files per git checkout call with path-conflict detection, falling back to single-file revert on failure

Benchmarks (200 files):

• revert: 2,851ms → 191ms (15x faster, 8x fewer git processes)
• concurrent track: 2/5 succeed → 5/5 succeed (corruption fixed)

Context

Users reported high RAM usage and excessive git processes, especially on Windows with Agent Manager. A potential root cause is the snapshot module spawning many concurrent/redundant git processes during track, revert, and diff operations. OpenCode upstream has addressed these with Effect-TS based optimizations, this PR ports the key improvements as plain async/Promise code without adopting the Effect runtime until we are able merge newest upstream releases.

Implementation

Three optimizations ported from upstream, all contained in packages/opencode/src/snapshot/index.ts:

• Per-gitdir mutex lock (upstream #17878, v1.3.0): A simple async lock (withLock) serializes concurrent snapshot operations per gitdir. Prevents race conditions where parallel track() calls corrupt the snapshot index. Without the lock, only 2/5 concurrent calls succeed; with it, all 5 return consistent results.
• Incremental git add (upstream #17878, v1.3.0): Replaces naive git add . (which walks the entire worktree) with git diff-files + git ls-files --others to find only changed/new files, filters out files >2MB via concurrent stat, then runs git add --sparse. Reduces work on large repos.
• Batched revert (upstream #20564, v1.3.14): Groups up to 100 files per git checkout call with path-conflict detection. Falls back to single-file revert on failure. Reduces 60+ sequential git process spawns to ~6.

Local Adhoc Benchmarks (200 files)

Operation	Before	After	Improvement
revert (60 files)	2,851ms / 65 git processes	191ms / 8 git processes	15x faster, 8x fewer processes
5x concurrent track	2/5 succeed	5/5 succeed	Corruption fixed

Screenshots

N/A -- backend optimization, no UI changes.

How to Test

1. Run the snapshot test suite: cd packages/opencode && bun test
   test/snapshot/snapshot.test.ts test/kilocode/snapshot-cache.test.ts
2. All 57 tests should pass (11 new tests covering concurrency, batching, and large file exclusion)
3. For manual testing: use Agent Manager to edit multiple files in a session and monitor git process count / RAM usage --should see significantly fewer git processes and more stable memory
4. PENDING: Windows testing

Get in Touch

[kilo-code-bot]

Code Review Summary

Status: 1 Issue Found | Recommendation: Address before merge

Overview

Severity	Count
CRITICAL	0
WARNING	1
SUGGESTION	0

Fix these issues in Kilo Cloud

Issue Details (click to expand)

WARNING

File	Line	Issue
packages/opencode/src/snapshot/index.ts	427	Oversized tracked files are still staged because git add --sparse . updates tracked entries even when they are listed in info/exclude.

Other Observations (not in diff)

Previously reported non-ASCII batched revert issue is resolved via core.quotepath=false. Previously reported unborn-HEAD git reset -- <paths> issue is resolved by removing that fallback.

Files Reviewed (2 files)

• packages/opencode/src/snapshot/index.ts - 1 issue
• packages/opencode/test/snapshot/snapshot.test.ts

---

_{Reviewed by gpt-5.4-20260305 · 221,966 tokens}