mruby-regexp: refuse a byte search offset inside a character - #7152

Follow-up to #7148, out of a review comment on it.

ecd891b gave String#byteindex and String#byterindex a check: a byte offset that lands inside a character names no position the string has, so a byte search refuses it rather than start from the middle of one. This gem takes both methods over for a Regexp argument, and that path asks nothing about the offset it is handed, so the same search answers one way for a String and another for a Regexp:

"あいうあいう".byteindex("い", 1)    # IndexError
"あいうあいう".byteindex(/い/, 1)    #=> 3, searched from inside the first character
"あいうあいう".byterindex("い", 1)   # IndexError
"あいうあいう".byterindex(/い/, 1)   #=> nil

CRuby raises for all four.

Where the question is asked

At the point the C methods ask it: after the offset has been read against the byte length and the ends have been answered, and of the offset the search will start from rather than the one the caller wrote. So an offset outside the subject stays a miss rather than becoming an error, at either end, and byterindex refuses over the position its clamp left, which for a position past the end is the end of the subject and a boundary.

str_check_byte_pos() becomes mrb_str_check_byte_pos() and moves into mruby/internal.h, so both paths refuse over one rule rather than two spellings of it. A binary receiver has a position per byte and is refused nothing, which is the core function's own first answer and comes along with it. The gem reaches it through Regexp.__check_byte_pos, beside Regexp.__check_encoding and Regexp.__check_pattern.

The walk is not asked. __regexp_rsearch resumes one byte past a match start, and String#scan one byte past a zero-width match; both are offsets inside a character on purpose, and the engine steps over one on its own.

Tests

The first commit records where the two forms stand before anything moves, and where they already agree: an offset on a boundary, every offset of a string that indexes by byte, and an offset outside the string at either end. The second flips the four answers that were the gem's own.

Beyond the suite, a differential over 6 subjects (ASCII, multibyte, mixed, 4 byte characters, binary, empty) by 6 patterns (literal, any, ASCII literal, groups, zero-width, no match) at every offset from past one end to past the other, comparing the answer or the exception of byteindex and byterindex in both forms: 756 cases, of which 741 answer as CRuby does. The 15 that do not are a binary receiver searched for a pattern spelled in UTF-8, where CRuby raises Encoding::CompatibilityError and mruby reads the bytes; the two forms agree with each other there, and did before this.

Verification

Full suite green at every commit on build_config/ci/gcc-clang.rb: full-core with MRB_GC_STRESS (2287), bintest (2287), the C++ ABI (2286), and the default gembox, where the new test skips (2071). 0 failures, 0 crashes, no new compiler warnings.

Summary by CodeRabbit

• Bug Fixes
  • Improved String#byteindex and String#byterindex handling for UTF-8 strings.
  • Positions inside a multibyte character now raise IndexError.
  • Valid character-boundary offsets continue to return correct byte positions.
  • Binary strings continue to accept byte-based offsets.
  • Out-of-range offsets consistently return nil for string and regular expression searches.
• Tests
  • Added coverage for UTF-8 boundaries, invalid offsets, binary strings, and out-of-range searches.