Add basic support for named capture group

[hhugo]

Summary

Based on top of #174
Add support for as bindings in sedlex patterns, allowing users to capture sub-matches by name:

match%sedlex buf with
| ("hello", (Star any as rest)) -> print_string (Utf8.of_submatch rest)
| _ -> ()

This uses a tagged DFA approach (Laurikari-style) that records sub-match positions in a single forward pass, with no backtracking penalty.

What's supported

• Basic captures: (p as x) binds x to the text matched by p
• Multiple captures: (p1 as x, p2 as y) in a sequence
• Or-patterns: (p1 as x) | (p2 as x) — both branches must bind the same names; a discriminator tag determines which branch matched
• Nested match%sedlex: inner and outer lexers maintain independent tag state
• Zero overhead when unused: rules without as generate identical code to before

What's rejected (with clear error messages)

• as inside repetition operators (Star, Plus, Rep, Opt)
• as inside set operators (Compl, Sub, Intersect)
• as in named regexp definitions (let%sedlex)
• Or-patterns where branches bind different names

Implementation

• NFA: nodes carry optional tags; Sedlex.bind wraps a regexp with start/end tag epsilon nodes
• DFA: epsilon closure collects tags; transitions carry tag action lists
• Runtime: new __private__mem arrays on the lexbuf store tag positions, saved/restored on mark/backtrack, adjusted on refill
• PPX: regexp_of_pattern returns tag info alongside the regexp; gen_binding_code emits let bindings that extract submatch from tag positions
• New submatch type in the public API for structured access to sub-match positions

Testing

• Runtime tests in test/basic.ml covering captures, or-patterns, nested match, and edge cases
• Codegen baseline tests in test/codegen/ tracking generated code and automata structure
• DOT graph output in test PPX for visual inspection of tagged automata

This is a minimal implementation — no tag optimizations are applied yet. The codegen tests document current (unoptimized) baselines to track future improvements (tag coalescing, offset propagation, etc.).

Next

If accepted, 4 PRs will follow to improve performances:
See https://github.com/hhugo/sedlex/pulls

[hhugo]

cc @toots

[pmetzger]

Let me know what you would like me to do with these things.

[hhugo]

Let me know what you would like me to do with these things.

This PR is ready to review. It would be nice to have another human review on it.

[pmetzger]

@toots @Drup Thoughts?

[hhugo]

@toots, do you think you could review this or should we look for some other reviewer ?

[toots]

@toots, do you think you could review this or should we look for some other reviewer ?

The feature looks very cool and I want to help but I'm very unfamiliar with that part of the codebase.

Therefore, my feedback will be about long term maintenance:

• Is it possible to document the implementation better?
• Is it possible to organize the implementation so that different parts are isolated and easy to reason about etc.

Maybe at first you could see if you can document the implementation? I gather that you're using a coding assistant. Those can be useful to do the base work and let your review using your own understanding of it to make sure that it is correct?

[toots]

Thanks for the documentation. This seems pretty detailed and if you're confident it is accurate, that's a great addition!

My next and hopefully last feedback: I see the tests checking a lot of the happy part.

Are we testing the error paths? Are errors and limitations of the implementations clear, expected, tested and documented for the user to understanding?

Thanks!

[hhugo]

Are we testing the error paths? Are errors and limitations of the implementations clear, expected, tested and documented for the user to understanding?

I've added documentation in the readme.

[hhugo]

@pmetzger, this can be merged after #190

[toots]

@pmetzger I still would like to have admin rights to this repo to enable the merge queue and auto merge.

[pmetzger]

@toots Not sure what you mean by merge queue and auto merge?

[toots]

@pmetzger the repository can be set to protect the main branch from direct merge and auto-merge PRs when they have been accepted and pass the CI. This removes friction when reviewing. For instance, I can review some changes, approve the approach and mark for auto merge when the author has finished fixing minor CI issues.