perf(server): avoid rereading unchanged files in review previews

[shivamhwp]

Review previews can reread thousands of unchanged files from the preceding second when Git's index timestamp falls on a whole second. In a constructed 30,000-file repository, this reduces median preview time from 1,305 ms to 167 ms, about 7.8× faster. It preserves the missing-edit fix from #12613. This is a timestamp-boundary improvement, not a general 7.8× speedup for every repository.

Remove the extra one-millisecond subtraction when setting the temporary index timestamp. The existing Node FileSystem service already reads bigint stats and truncates to whole milliseconds before constructing its Date; flooring that Date preserves the source second. There is no new native import, diagnostic suppression, or error-handling path. Regular- and split-index tests verify that unchanged preceding-second files do not run their clean filter and that the real index stays unchanged.

Validation

Updated with main on September 25 at 9c41485408623e339b799e417be8e7e6f52999e8. Reran all 104 focused tests, server typecheck, formatting, and 32 live API requests on this integration commit; all passed. Targeted lint still has the same preexisting warning.

Detailed matrix and benchmark head: 4a0fd4fcb165dc61eaa088e4e23d30a50b64e57f. Baseline: b2b43bef73447c483ceae486890cb79f01c369cb, main immediately after #12613. No uncommitted code was included.

• All 104 focused core tests pass with installed Git 2.39.5 and with a locally built Git 2.39.5 using USE_NSEC. The latter uses system Git helpers for submodule commands.
• All 62 live API scenarios pass in each of three configurations: Node 24.21 / installed Git, Node 24.21 / USE_NSEC Git, and Node 22.16 / installed Git. These cover index versions 2/4, split indexes, ctime trust on/off, timestamp boundaries, partial staging, sparse indexes, untracked cache, assume-unchanged/skip-worktree, binary content, symlinks, executable mode, unusual filenames, linked worktrees, concurrent previews, and preservation of the real index.
• Another 32 API requests pass for repeated previews, individual file previews, file contents, regular/split indexes, linked worktrees, and whole/fractional timestamps.
• Additional exact-nanosecond/year-2038 and read-only/unreadable/corrupt/locked/symlinked/missing-index and nested-directory checks ran against both Git builds. Each passed 17 of 19 scenarios; the two failures are the same inherited first-Unix-epoch-second limitation described below. Permission/corruption checks expect an error rather than an empty successful preview.
• Server typecheck and formatting pass. Targeted lint reports one preexisting warning. The two new cache tests fail on the baseline and pass here; original same-size-edit regressions remain passing.

Performance

Linux, 30,000 files / 469 MiB, warm filesystem cache, forced timestamps. Times are medians of complete WebSocket requests; correctness and original-index preservation were asserted throughout.

Scenario	Baseline	This PR	Requests per version
One ambiguous edit; 29,999 unchanged preceding-second files	1,305 ms	167 ms	12
All 30,000 files in the ambiguous second	1,298.5 ms	1,308 ms	12
Ordinary timestamps, short benchmark	151.5 ms	163 ms	12
Ordinary timestamps, longer alternating run	137 ms	134 ms	80
No untracked files, changed function bypassed	97 ms	94 ms	80

The short benchmark used baseline/PR/PR/baseline order with one warmup per batch. The longer run used baseline/PR/PR/baseline/PR/baseline/baseline/PR with five warmups and 20 measurements per batch. Small ordinary-case differences change direction between runs, including variation in the bypass control. The earlier 163 → 182 ms slowdown is not a stable result; these measurements do not prove zero overhead. Files that actually need checking still cost about 1.3 seconds in this fixture.

Why the existing timestamp is precise enough

The pinned @effect/platform-node-shared adapter calls fs.stat(path, { bigint: true }) and returns its mtime Date. Node's BigIntStats computes mtimeMs = mtimeNs / 1000000n before constructing that Date. Ordinary non-bigint stats can round across a second; this bigint-backed path does not. We verified the installed adapter, Node 22/24 runtime behavior, and exact final-nanosecond fixtures. See Node 22.16 BigIntStats.

The previous revision added a separate native stat call unnecessarily. This revision keeps the existing service and its error handling. The existing fractional-timestamp regression tests protect this dependency on the Node adapter's conversion behavior.

Raw test and benchmark results

Limits

Timestamps within the first second after the Unix epoch still reproduce a missed edit with both baseline and PR, because a zero-second index disables Git's racy check. This PR fixes the baseline failure at exactly epoch second 1, but does not claim to repair all epoch-zero behavior. Missing-index behavior is also inherited.

Windows, macOS, controlled cold-disk tests, concurrent external index rewrites, and real-world frequency remain unmeasured. Both T3 Browser status and open report no automation host, so the requested before/after UI recording could not be captured. No frontend, provider, or wire-contract code changes are included.

Implemented and tested with GPT-6 in the Codex harness. Claude Opus 5.5 reviewed the earlier native-stat version; the simplification was verified separately in this pass.

Summary by CodeRabbit

• Bug Fixes
  • Improved review previews for repositories using standard or split Git indexes. Previews now correctly include untracked files without unnecessarily running clean filters or changing the existing index.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.5 KiB	13.5 KiB	−36 B (−0.3%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	−2 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.4 KiB	−34 B (−0.5%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.3 KiB	56.2 KiB	−44 B (−0.1%)	66.4 KiB	✅
Codex	Live turn messages	10	9	−1 (−10.0%)	21	✅
Claude	Total thread wire	13.5 KiB	13.5 KiB	+5 B (+0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	+3 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.4 KiB	+2 B (+0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.0 KiB	57.0 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	9	9	0 (0.0%)	21	✅

Baseline: e5a46d6 · PR result: 9c41485 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 114.0 KiB
• Claude decoded thread snapshot: 114.7 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: pingdotgg/t3code/.coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: dc77bc6d-2f7b-4508-a8b1-76f4973645a4

📥 Commits

Reviewing files that changed from the base of the PR and between 4a0fd4f and 9c41485.

📒 Files selected for processing (1)

• apps/server/src/vcs/GitVcsDriverCore.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 8 remain after this review.

---

📝 Walkthrough

Walkthrough

prepareReviewIndex timestamps the copied temporary index using the index modification time floored to whole Unix seconds and clamped to zero. Tests cover review previews with split indexes enabled and disabled.

Changes

Review index timestamp handling

Layer / File(s)	Summary
Read the index timestamp and verify preview behavior apps/server/src/vcs/GitVcsDriverCore.ts, apps/server/src/vcs/GitVcsDriverCore.test.ts	prepareReviewIndex floors the index modification time to whole Unix seconds and clamps it to zero, using zero when the time is absent. Tests verify that previews include only the untracked file and leave the clean filter and original index unchanged for both split-index settings.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 9c414

Preview tests cover both avoiding unnecessary reads of unchanged files and preserving edited tracked files at the timestamp boundary. No merge-blocking preview risk is evident.

Architecture Summary

Architecture risk: 🔵 Low · up to 9c414

The change affects 1 system.

Changed systems: apps/server

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

• observed — apps/server (service) was modified; 2 changed files map to changed impact.

Before / after behavior

• observed — Modified behavior in apps/server/src/vcs/GitVcsDriverCore.test.ts: Adds the same preceding-second review-preview test for both split-index settings. It asserts that only untracked.txt appears in the working-tree preview, while the clean filter remains unrun and the index bytes and modification time remain unchanged.
• observed — Modified behavior in apps/server/src/vcs/GitVcsDriverCore.ts: prepareReviewIndex updates the timestamp comments and removes the one-millisecond subtraction before flooring the index modification time; the resulting time remains clamped to zero, with zero used when the time is absent.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 2…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly identifies the main performance change: avoiding rereads of unchanged files in server review previews.
Description check	✅ Passed	The description clearly explains what changed, why it changed, validation results, performance measurements, scope, and known limitations. It does not use the template headings or include the checklis…

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

_{Comment @coderabbitai help to get the list of available commands.}

[macroscopeapp]

Approvability

Verdict: Approved at 4a0fd4f

Macroscope's review found this PR approvable — This is a small, test-covered adjustment to temporary Git index timestamps that avoids rereading unchanged files during review previews while preserving racy-edit detection and the real index. It introduces no product-default, schema, sensitive-code, or static-analysis configuration changes.

No code changes detected at 9c41485. Prior analysis still applies.

^{You can add or adjust custom eligibility rules. Learn more.}