perf: micro-optimize resolver normal path

[hardfist]

Goal

Micro-optimize the resolver normal path by at least 5% using exact Callgrind instruction references as the primary metric.

Benchmark

Target benchmark:

RUSTFLAGS="-C debuginfo=1 -C strip=none -g" cargo codspeed build
CODSPEED_CARGO_WORKSPACE_ROOT=$PWD valgrind --tool=callgrind --callgrind-out-file="$out_dir/callgrind.out.%p" "$bench_bin" "benches/resolver.rs::resolver::bench_resolver::resolver[single-thread]" --exact

Metric: Callgrind Ir from CodSpeed Valgrind.

Baseline: 116,917,041 Ir.
5% target threshold: 111,071,189 Ir.
Final: 109,733,110 Ir (-7,183,931, -6.15%).

Changes and Measurements

Commit	Ir	Delta vs baseline	Check	Notes
05be16b	116,554,214	-362,827 (-0.31%)	cargo test --lib -- --skip tests::pnp	Avoid UTF-8 conversion while appending extensions on Unix.
16c1795	114,956,513	-1,960,528 (-1.68%)	cargo test --lib -- --skip tests::pnp	Skip impossible bare alias checks for absolute requests.
278d640	112,025,825	-4,891,216 (-4.18%)	cargo test --lib -- --skip tests::pnp	Partition aliases and fallback aliases by absolute vs relative request kind.
e6517c6	109,733,110	-7,183,931 (-6.15%)	cargo test --lib -- --skip tests::pnp	Precheck path aliases before path-to-string conversion.

Artifacts

• Baseline: optimization-artifacts/valgrind/callgrind/20260517T043326Z-single-thread-baseline/
• Unix extension buffer: optimization-artifacts/valgrind/callgrind/20260517T043726Z-single-thread-unix-extension-buffer/
• Absolute alias filter: optimization-artifacts/valgrind/callgrind/20260517T044703Z-single-thread-absolute-alias-filter/
• Partition aliases: optimization-artifacts/valgrind/callgrind/20260517T045216Z-single-thread-partition-aliases/
• Path alias precheck: optimization-artifacts/valgrind/callgrind/20260517T045641Z-single-thread-path-alias-precheck/
• Final clean measurement: optimization-artifacts/valgrind/callgrind/20260517T045822Z-single-thread-final/

Test Status

cargo test --lib -- --skip tests::pnp passes (123 passed).

A full cargo test --lib still fails the existing PnP fixture cases (125 passed; 6 failed) with NotFound fixture paths. Those failures were observed before these perf changes and are tracked here as a residual environment/fixture gap, not as a new regression from this branch.

Status

Target achieved for exact resolver[single-thread]: 109,733,110 Ir is 6.15% below the clean baseline.

[codspeed-hq]

Merging this PR will improve performance by 9.94%

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 5 improved benchmarks
✅ 7 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	resolver[[single-threaded]resolve with many extensions]	147.2 ms	141.4 ms	+4.08%
⚡	Simulation	resolver[multi-thread]	65.1 ms	61.5 ms	+5.91%
⚡	Simulation	resolver[resolve from symlinks]	195.6 ms	172.2 ms	+13.58%
⚡	Simulation	resolver[single-thread]	56.9 ms	54 ms	+5.43%
⚡	Simulation	resolver[resolve from symlinks multi thread]	123.7 ms	101.7 ms	+21.67%

Tip

Curious why this is faster? Comment @codspeedbot explain why this is faster on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing perf/micro-optimize-resolver (91e539f) with main (8975e38)}