deps(regex): take perex 0.1.10; ready for the perex#3 matcher release

[proggeramlug]

Draft. Blocked on a perex release that contains PerryTS/perex#3, and on that release clearing the 7-day min-publish-age soak. This PR is ready to receive that version bump.

Part of #10166

What it does today:

• Moves Perry from perex 0.1.9 to 0.1.10. 0.1.10 was published 2026-09-20 and has cleared the soak.
• Adapts to 0.1.10's one API change. Search::run now returns RunError { error, remaining_work, buffers }, so both call sites in perex_runtime.rs record the work a failed run left instead of keeping the entry budget. That retires the "under-counts what the failed call spent" caveat those sites carried. Nothing observes it today, because WORK is usize::MAX on these paths.

The step still missing is perex = "0.1.11" (or whichever release carries perex#3), which brings the matcher changes measured below. Nothing else in Perry changes for it.

What the perex#3 matcher changes do in Perry

Measured on qb2 (AMD EPYC 9254, idle), --release, PERRY_NO_AUTO_OPTIMIZE=1. The perex#3 branch is built in through a local [patch.crates-io] path. Both arms sit on top of #11543, and the only difference between them is the perex source plus this adaptation. Instructions come from perf stat -e instructions:u as a two-N differential.

Package workloads (scripts/package_bench.py run --arms node,perry --modes instr). Every Perry run's output was byte-identical to Node 26.5.1:

workload	node	before	with perex#3	Δ
jsonwebtoken/decode	34,850	233,972	112,469	−51.9%
uuid/v4	6,556	63,660	33,225	−47.8%
uuid/v7	18,331	101,423	71,090	−29.9%
uuid/v5_parse	48,301	251,964	191,072	−24.2%
nanoid/generate	2,884	68,670	55,080	−19.8%
dotenv/parse	203,214	2,948,837	2,877,221	−2.4%
decimal.js/parse_sum	1,722,290	43,623,719	42,867,092	−1.7%
node-cron/validate_parse	427,893	4,357,781	4,284,555	−1.7%
validator/batch	957,163	28,593,841	28,303,073	−1.0%
moment/parse_format	115,070	3,899,511	3,879,889	−0.5%
date-fns/format_add	85,895	1,166,059	1,163,517	−0.2%
commander/parse_argv	193,121	4,024,680	4,038,212	+0.3%
dayjs, date-fns/diff_interval, validator/sanitize				within ±0.1%

Microbenchmarks, instructions per call:

probe	before	with perex#3	Δ
uuid REGEX.test (/^(?:[0-9a-f]{8}-…)$/i)	23,500	8,227	−65.0%
jws JWS_REGEX.test, 155-char token	96,711	15,395	−84.1%
/^\d+$/.test	4,453	4,164	−6.5%
dotenv LINE.exec loop, 36-line document	1,609,579	1,559,671	−3.1%
split(/%..|./)	166,370	166,526	+0.1%

Peak RSS (median of 5, harness n2) moves by less than 1% on every workload measured except moment/parse_format, which is −6.1%. The matcher allocates nothing new; its scratch is unchanged.

Tests (qb2, with the perex#3 patch)

• cargo test --release -p perry-runtime --lib -- perex split replace regex (RUST_TEST_THREADS=1): 274 passed.
• run_parity_tests.sh (PERRY_SKIP_BUILD=1 PERRY_NO_AUTO_OPTIMIZE=1) with filters regex, regexp, split, replace, match, against Node 26.5.1: 65 PASS, plus the same pre-existing COMPILE_FAIL main has (test_gap_regex_replace_dyn_regex_with_http).
• test-files/test_gap_regex_engine_package_shapes.ts (added in perf(regex): cut split and replace host overhead on ASCII subjects #11543) is byte-identical to Node with the patch.
• perex#3 has its own full check list: cargo test in debug and release, every differential harness including q1/q17 pausing and relocation, and test262.

Run on this branch as pushed, with perex 0.1.10 from the registry: cargo test --release -p perry-runtime --lib -- perex split replace regex (RUST_TEST_THREADS=1), 274 passed. Not run with 0.1.10: the gap filters. The adaptation is the same code that passed them against the patched perex. Also not run: the full gap sweep, the auto-optimize arm, macOS, and cargo test --workspace.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[proggeramlug]

perex 0.1.11 is published (crates.io, tag v0.1.11, commit 80845f531 = perex#3 merged + version bump). It clears Perry's 7-day publish-age soak around 2026-10-04 21:40Z. Then this PR only needs the version bump from 0.1.10 to 0.1.11 and a re-measure. Measured with perex#3 patched in locally on top of #11543: jsonwebtoken decode −52%, uuid v4 −48%, v7 −30%, v5 parse −24%, nanoid −20%, RSS within 1%.