mruby-regexp: return Arrays from Regexp#named_captures - #7026

Regexp#named_captures returns Integer values where CRuby returns Arrays.

/(?<a>x)/.named_captures       # CRuby: {"a" => [1]}, mruby: {"a" => 1}

It also hands out the Hash stored in the @named_captures instance variable, so a
caller that writes into the returned Hash corrupts the Regexp for every later caller:

re = /(?<a>x)/
re.named_captures["a"] = 99
re.named_captures              # CRuby: {"a" => [1]}, mruby: {"a" => 99}

CRuby uses an Array because one name can map to several group numbers, as in
/(?<x>a)|(?<x>b)/. mruby-regexp accepts such a pattern but does not handle it
correctly yet: the later group silently overwrites the earlier one in the Hash, and a
name lookup resolves to the first entry in the C table regardless of which alternative
matched. Duplicate names therefore still resolve wrongly with this change applied:

/(?<a>x)|(?<a>b)/.named_captures  # CRuby: {"a" => [1, 2]}, mruby: {"a" => [2]}
/(?<a>x)|(?<a>b)/.match("b")[:a]  # CRuby: "b",             mruby: nil

That is a separate defect in the compiler and in name resolution, left out of this
change. The return type is aligned so that portable code can be written today.

Changes

• Regexp#named_captures derives a fresh Hash from the @named_captures table on
  every call and wraps each group number in an Array. regexp_init() keeps filling the
  instance variable with the internal name to group-number table, and the public method
  becomes a derivation of it, which is also where the shared-Hash aliasing goes away.
• MatchData#named_captures is unaffected. It rebuilds its result from the compiled C
  table on every call and already answers name to captured string.
• Regexp#named_captures is added to the Ruby API block in
  mrbgems/mruby-regexp/README.md.

Tests

A new assert("Regexp#named_captures") in mrbgems/mruby-regexp/test/regexp.rb covers
the Array value type, the empty Hash for a pattern with no named group, and that
mutating the returned Hash does not affect a later call.

rake test passes.

Summary by CodeRabbit

• New Features

  • Enhanced Regexp#named_captures to return each capture name with its corresponding group index.
  • Returns an empty hash when no named captures are present.

• Bug Fixes

  • Ensured each call returns an independent result, preventing changes from affecting later calls.

• Documentation

  • Added documentation describing the named_captures return format.