grep: support POSIX equivalence classes

[wondr-wclabs]

Fixes #35.

This adds a small BRE/ERE pre-compile normalization step for POSIX equivalence-class entries inside bracket expressions. Because uu_grep does not currently implement locale collation, the implementation only normalizes single-character [=c=] entries to the equivalent literal character, which matches the C-locale behavior described in the issue. Multi-character collating elements are left unchanged rather than guessing at locale-specific behavior.

The parser handles POSIX character-class and collating-symbol tokens inside bracket expressions so patterns such as [[:alpha:]][[=1=]] keep their existing meaning, and fixed-string mode is intentionally left literal.

[codspeed-hq]

Merging this PR will not alter performance

✅ 10 untouched benchmarks
⏩ 17 skipped benchmarks¹

---

_{Comparing wondr-wclabs:codex/posix-equivalence-classes (c1d4187) with main (d28bf76)}

Footnotes

1. 17 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. ↩

[lhecker]

I'm personally willing to accept this as a broken feature until we have an alternative to Oniguruma. The equivalence classes are somewhat niche. Not sure how others view it...

[wondr-wclabs]

That is a fair concern. This patch is deliberately narrow: it only normalizes single-character [=c=] entries to the C-locale literal form, and it does not attempt locale collation or multi-character collating elements. That means it fixes the concrete C-locale behavior from #35, but it still leaves the larger POSIX equivalence-class semantics unimplemented.

Given the feedback on #55, I agree the main tradeoff is not whether this small parser can pass the local case, but whether grep should grow local pre-compile normalizers around Oniguruma at all. If the project direction is to avoid those until there is a broader regex-engine strategy, then this PR is probably not worth merging even though the current checks were green before main moved.

My view is: this is acceptable only if maintainers want the narrow C-locale compatibility improvement now. If the preferred direction is “leave equivalence classes broken until the engine story changes,” I can close this PR rather than rebase and keep adding review noise.

[lhecker]

the engine story

I'll close this PR for now then. If other maintainers consider this a priority to fix, let's reopen the PR!