pcre2: publish the Unicode regex dependency root

[brandonpayton]

Purpose

Replace the source-only registry entry with a first-class Kandelo Homebrew PCRE2 10.44 formula.

The formula builds through the normal Kandelo SDK and publishes the actual development and tool surface: 8-, 16-, and 32-bit static libraries, the POSIX wrapper, public headers, pkg-config metadata, pcre2-config, pcre2grep, and pcre2test.

Registry reconciliation

packages/registry/pcre2-source/package.toml is authoritative for:

• Version: 10.44
• Source: https://github.com/PCRE2Project/pcre2/releases/download/pcre2-10.44/pcre2-10.44.tar.gz
• SHA-256: 86b9cb0aa3bcb7994faa88018292bc704cdbb708e785f7c74352ff6ea7d3175b
• License: BSD-3-Clause

The downloaded source archive matches that digest. The Homebrew package is named pcre2 because it now publishes built, consumer-visible artifacts rather than an unpacked source cache.

Formula contract

• wasm32 scope is explicit.
• Builds static libpcre2-8, libpcre2-16, libpcre2-32, and libpcre2-posix.
• Installs pcre2.h, pcre2posix.h, all four upstream pkg-config files, and pcre2-config.
• Rewrites installed consumer metadata from the versioned Cellar staging path to the stable opt prefix.
• Builds pcre2grep and pcre2test with Unicode/UTF/UCP and full pcre2grep fork/exec callouts.
• Disables JIT because native executable-memory code generation is not a WebAssembly capability.
• Keeps optional zlib/bzip2 compressed-input support out of this first dependency-free target contract. Tap main does not yet publish the libbz2 header/static-library surface; compression: publish linkable bzip2 and xz libraries #25 can support a separate follow-up.
• Declares pkgconf for both build and test, with WABT and Binaryen as build tools.
• Fork-instruments pcre2grep and rejects ABI mismatch, Asyncify residue, missing fork exports, or an unexpected uninstrumented fork import before installation.
• Treats /bin/sh as Kandelo base-system state supplied by dash, not a formula dependency.

Platform and bottle gates

The source formula may merge independently after formula review. Bottle construction and publication must wait until the applicable platform gates are green:

• Automattic/kandelo#842
• Automattic/kandelo#873
• Automattic/kandelo#874
• Automattic/kandelo#881
• Automattic/kandelo#882
• Automattic/kandelo#885
• Automattic/kandelo#888 makes fork instrumentation byte-reproducible.
• Automattic/kandelo-homebrew#22 is required only for branch-aware pre-merge bottle dry runs; it is not a source-merge gate.

The pthread fork-continuation fix and ABI 18 transition from #867 are merged.

The deterministic-instrumentation gate is fully isolated. Immutable uninstrumented pcre2grep input SHA-256 6d3a835bf6cca612c4b722c8b60291e30f07b51bf51eaf6288c355efbf6592bb and exact instrumenter SHA-256 4f25fdc47f546ded0b042ceff337ac976bd456fbaa42df973eeb8bd258f064d4 produced two same-size outputs:

• 3f15de738b73ee63ab5e898804ae02ae5ad3b78e3fdb9cdad47c67ad20e10667
• effdff934a21c79107332ebb6b572793af2ad84e4d5884d4fe1567b703c2dea8

The specific cause is crates/fork-instrument/src/instrument.rs in instrument_one_function_nested_switch: iterating the regions HashMap keys while building pending carryover plans randomizes subsequent module.locals.add allocation order. That changes only synthetic local indices in the Code section. Section sizes, function/export/data/name ordering, and normalized instructions are identical. Sorting that allocation loop made two outputs byte-identical in an isolated proof; sorting only the later body-parameter loop did not.

No formula workaround or bottle hash is proposed. #888 fixes the tool at its allocation-order root cause; CI must rebuild this formula after that platform change lands and the remaining publisher gates are green.

Validation

Functional validation used scripts/dev-shell.sh, ABI 18 platform artifacts, the #882 ABI parser, and exact formula head 40ad3b52e47a7e44125e31a9a070c49beb573404. The branch was then rebased semantic-preservingly onto current tap main at adb2364. Commit 11d291f subsequently changed only shared-support loading and the pkgconf test dependency; the four artifact-producing commits retain identical stable patch IDs and artifact logic:

• Exact-head source reinstall passed: 229 installed files, 6.1 MB.
• Exact-head brew test passed.
• Node: pcre2grep/pcre2test feature reporting, UTF-8 Unicode property matching, fork/exec callout through /bin/sh, stable pkg-config output, and linked 8/POSIX, 16, and 32-bit consumers passed.
• Chromium: the linked PCRE2/POSIX consumer, Unicode pcre2grep behavior, and fork/exec callout passed with exact expected output.
• Final executable audit: ABI 18; imports only from env and kernel; no Asyncify. pcre2grep has all five fork exports; pcre2test has no fork instrumentation requirement.
• One exact-head local artifact instance: pcre2grep SHA-256 cd92b403c5cb1289b0caf9810ec61f02b964405d55639e01384289baa937cf0e; pcre2test SHA-256 6c4968b2393298439628c7c11378f74a37daa9e843d3fc76f67ee9ebb6c3c48c. These are validation evidence, not publishable bottle hashes while determinism is gated.
• Current tap formula-support tests: 21 runs, 186 assertions.
• homebrew-validate: passed for the unchanged checked sidecar set.
• brew audit --strict: passed.
• brew style: no offenses.
• Ruby syntax and git diff --check: passed.
• Post-rebase brew style, Homebrew Ruby syntax, formula-support tests, and git diff --check: passed.
• At 11d291f, Homebrew Ruby syntax, brew style, a literal formula load through Tap.fetch, formula-support tests (21 runs, 186 assertions), and git diff --check passed through Kandelo's dev shell.

No bottle block or PCRE2 sidecar is included.