[dinwwwh]

Requested by dinwwwh · project thread

What

expand tracks how many of the previous route's parts the next route can reuse (literals). It used to lower that count in two places: once where a route waits, and once where it adds a param. Both happen only for a part that starts with { once decoded, so the count is now lowered in one place, right after that check.

That check also catches a part such as {a.b}, which starts with { but is literal text. The count now drops for that part too, so the next route shares fewer parts with this one and parses them again. Matching is unaffected. The comment on shared and literals now describes the rule this way.

Measurements

Instruction counts from callgrind, node --single-threaded, averaged over hash seeds 1 and 2:

	before	after
min / min+gzip	5012 / 2092 B	5001 / 2091 B
REST cold+3 match	3.498M	3.485M (−0.35%)
REST cold+3, one request per method	5.075M	5.064M (−0.2%)
REST cold+10	6.010M	5.998M (−0.2%)

expand only runs while the tree is being built, so steady-state matching does not change.

Checks

• Tests pass with 100% coverage.
• The model fuzzers, including the late-add ones, find 0 diffs.
• Stryker finds 66 survivors, down from 67, and none of them are new. The if (true) mutant of the merged check is killed by the late-add test added in perf(router): make fewer calls per part when building the tree #195.

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed]

Merging this PR will not alter performance

1 warning

⚠️ 1 benchmark is dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

✅ 29 untouched benchmarks
⏩ 2 skipped benchmarks¹

---

_{Comparing claude/project-thread-m32ra9 (fa5ae66) with main (f18cd06)}

Footnotes

1. 2 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. ↩

[pullfrog]

✅ No new issues found.

Reviewed changes

• Single-site literals lowering — if (index < literals) { literals = index } moved from the wait branch and the param branch to one spot immediately after if (segment[0] === '{'), so the shared-parts count drops once per braced part.
• Comment update — the shared/literals comment now describes the rule as "the parts before the first part starting with { once decoded", which matches the code.
• Behavioral delta — a literal part that starts with { after decoding (e.g. {a.b}) now lowers literals too. This is the one semantic change the PR calls out, and I verified it is safe: new literals is always ≤ old (lowering happens at a superset of indices), the count resets per route, and a smaller count only stops the next route's shared-prefix walk earlier, re-walking idempotent literalChild/child.param ??= steps. The extra expand(child.pending, …) those re-walks can trigger is a no-op, since the route that set literals already flushed that region. A differential fuzz over {a.b} / %7Ba.b%7D permutations with late-add warmups (forcing whole === true and partial builds) found 0 mismatches, and late-adds.test.ts:49 already covers the duplicate {a.b} case.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}