perf(linter): compute comment fix span only for directive comments

[connorshea]

This is a micro-optimization but in a repo with 10k+ files with 10k+ comments, this would have some (very minor) impact that is worth doing, given that the change is so small.

What this does

DisableDirectivesBuilder::build_impl pre-computed the fix span for every comment in the file — before match_directive filters out non-directive comments, which is the overwhelming majority. The fix span is only consumed after that filter (all uses sit below the early continue), so for ordinary comments the computation was pure waste. This PR moves the computation below the directive check. 5 insertions, 4 deletions, no logic change.

Benchmark

Interleaved paired samples (N=40 pairs, warmup, time.monotonic_ns()), single rule enabled (eslint/no-debugger, zero diagnostics on the fixtures so the common non-reporting path is measured), baseline binary built from pristine main:

Fixture	Baseline	Optimized	Speedup
Single-line file, 10k inline /* */ comments (~318 KB)	91.96 ms	8.90 ms	10.3× (t=297.6, 40/0 wins)
30k short // comment lines	12.15 ms	11.74 ms	1.04× (t=2.7, 29/11)
Control: comment-free 30k-line file, all rules off	13.11 ms	13.32 ms	0.98× (t=-1.0, noise)

The adversarial single-line case is where the quadratic bites (minified bundles, generated code). Typical multi-line code sees a small win proportional to comment density. The comment-free control shows the deltas aren't binary-layout noise. (A rules-off "parse-only" run on the comment fixture shows the same ~10× — the directives builder runs regardless of rule config, which is exactly why this path is worth trimming.)

Correctness

• All 1169 oxc_linter lib tests pass, including the fix_span_* unit tests covering compute_comment_fix_span.
• Diagnostics parity (--format=json, diagnostics array compared) is exact between baseline and optimized binaries on both fixtures and on a canary exercising eslint-disable-line, eslint-disable-next-line, block eslint-disable/eslint-enable, an unused directive, and comments sharing lines with code.
• --fix --fix-suggestions --fix-dangerously output is byte-identical between the two binaries on that canary.
• The reorder computes the same value from the same inputs for every comment that reaches a use; only the timing moves.

AI disclosure

This change was implemented with Claude Code. Reviewed by me manually to confirm the fix is safe.

🤖 Generated with Claude Code

[codspeed-hq]

Merging this PR will not alter performance

✅ 5 untouched benchmarks
⏩ 66 skipped benchmarks¹

---

_{Comparing perf/linter-disable-directives-lazy-fix-span (688a1b6) with main (8de6fca)}

Footnotes

1. 66 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[camc314]

thanks!

[camc314]

Merge activity

• Jul 13, 9:27 AM UTC: The merge label '0-merge' was detected. This PR will be added to the Graphite merge queue once it meets the requirements.
• Jul 13, 9:27 AM UTC: camc314 added this pull request to the Graphite merge queue.
• Jul 13, 9:36 AM UTC: Merged by the Graphite merge queue.