perf: reduce scan startup and workspace contention

[aidenybai]

Summary

• share Oxlint subprocess capacity across workspace projects and preserve nested project capabilities
• load a lightweight rule-metadata entry for core scans while keeping the full plugin entry unchanged
• invalidate browser-guard and Next.js manifest caches when their real dependencies change
• reuse semantic, traversal, filesystem, and render-dependency analysis on hot rule paths
• keep registry generation working on the declared Node 20.19.0 floor without a Vite config subprocess
• accept complete Astro reports in Daytona parity tooling and ignore explicitly disabled inline Remotion CSS timing

This incorporates the applicable ideas from #1463, #1478, #1479, and #1480 in the current architecture.

Performance measurement

Exact base 5dc936e11 and final head b1a966e07 were built separately and benchmarked on the same 16-core Neoverse-V2 host with Node 24.18.0. Every series used excluded warmups followed by 5–9 measured samples; diagnostic counts and hashes matched exactly.

Target	Mode	Base median	Head median	Elapsed reduction	Throughput
deterministic 302-file stress project	lint, no cache	3.434s	2.619s	23.7%	1.31×
React Doctor workspace	lint, no cache	3.134s	1.858s	40.7%	1.69×
deterministic 302-file stress project	full, no cache	3.976s	3.156s	20.6%	1.26×
React Doctor workspace	full, no cache	4.157s	2.841s	31.7%	1.46×
deterministic 302-file stress project	lint, hot cache	1.182s	1.046s	11.6%	1.13×
React Doctor workspace	lint, hot cache	522ms	490ms	6.1%	1.07×

The core scan startup payload also falls from 2,793,346 bytes to 374,924 bytes (86.6% smaller). Existing scan.duration telemetry remains the production success metric; revert the lightweight path if cold-scan p50 does not improve over two releases.

Validation

• full root test suite: 15 of 15 tasks passed
• lint, typecheck, format check, full build, JSON report smoke, and diff hygiene passed
• exact Node 20.19.0 container: registry generation, plugin build, and the complete plugin test suite passed
• parity-tool suites: 33 of 33 tests passed, including complete Astro reports
• complete plugin suite: 25,137 tests passed, 201 skipped
• strict 500-iteration fuzzing passed for remotion-no-css-transition; regression corpus covers disabled inline timing
• deterministic stress fixture preserved 1,200 diagnostics and hash afcd70784219b21dcc2fc570b47f423e37e19a1e7df6bcafd8712bcb8b0e2605
• React Doctor latest diff scan: score 100, zero new diagnostics

Daytona parity

Canonical pinned run at 4d4fb9d07:

• 2,000/2,000 projects compared; 0 failed; 0 skipped
• baseline 1,205,762 diagnostics; candidate 1,205,779
• +17, -0, 1,205,762 unchanged
• all 17 additions were source-audited: 14 valid findings and 3 explicit transition: "none" false positives

Exact final-head targeted run at b1a966e07 covers every repository affected by the Remotion delta:

• 4/4 pinned projects completed; 0 failed; 0 skipped
• +6, -0, 2,946 unchanged
• both real CSS transitions and all four native-media findings remain
• all three explicit-disabled-transition false positives are gone

The final commit changes diagnostic behavior only in that audited Remotion path; the parity-tool Astro change does not affect scanner output. Composing the complete canonical run with the exact final-head affected-project rerun leaves 14 valid additions and no removals or known regressions.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/eslint-plugin-react-doctor@1533

npm i https://pkg.pr.new/oxlint-plugin-react-doctor@1533

npm i https://pkg.pr.new/react-doctor@1533

commit: b1a966e

[github-actions]

React Doctor found no new issues. 🎉

_{Reviewed by React Doctor for commit b1a966e.}

[github-actions]

Interactive terminal E2E

Recorded from the built CLI at b1a966e in a real terminal. The fixture holds Git busy for three seconds, so Scanning... must appear immediately after project selection, then exercises the compact interactive report.

Download the GIF and MP4 artifact

[aidenybai]

Controlled performance benchmark

Exact base 5dc936e11 and final head b1a966e07 were built separately and benchmarked on the same 16-core Neoverse-V2 host with Node 24.18.0. Each series used excluded warmups followed by 5–9 measured samples.

Target	Mode	Base median	Head median	Elapsed reduction	Throughput speedup
Deterministic 302-file stress project	lint, no cache	3.434s	2.619s	23.7%	1.31×
React Doctor workspace	lint, no cache	3.134s	1.858s	40.7%	1.69×
Deterministic 302-file stress project	full, no cache	3.976s	3.156s	20.6%	1.26×
React Doctor workspace	full, no cache	4.157s	2.841s	31.7%	1.46×
Deterministic 302-file stress project	lint, hot cache	1.182s	1.046s	11.6%	1.13×
React Doctor workspace	lint, hot cache	522ms	490ms	6.1%	1.07×

Correctness controls:

• Stress diagnostics remained byte-identical: 1,200 findings, hash afcd70784219b21dcc2fc570b47f423e37e19a1e7df6bcafd8712bcb8b0e2605.
• Workspace diagnostics remained byte-identical in each compared mode.
• The benchmark harness verified matching source fingerprints, diagnostic hashes, host, Node/V8 version, mode, cache cohort, and worker configuration before producing comparisons.

Conservative result: uncached/default-like scans show 1.26×–1.46× throughput, while hot-cache scans show 1.07×–1.13× throughput.