Source-based recognition

Every terminal, combinator, Ruleset and RFC import uses the same source-based runtime. Public calls are matchPrefix(Source, int $byteOffset = 0) and parseAll(Source, int $byteOffset = 0), returning ParseResult. Old string matching, ASCII bridges and source-compatibility dispatch have been removed.

Text matches full literals one recognition unit at a time. Char, Set, Range and Any consume one code point in UTF-8 mode or one octet in byte mode. Text literals must be valid UTF-8 in text mode; octet bounds/characters must be one byte in byte mode. Range performs code-point ordinal comparisons for text. Case mapping uses explicit UTF-8 in text mode and ASCII byte mapping in byte mode. Range rejects case mappings that expand into more than one code point.

Regex matches must begin at the supplied byte position and consume a nonempty match. Text mode enables PCRE UTF-8, rejects \C byte escapes and checks complete code-point endpoints. Byte mode rejects regex UTF options. Regex execution errors throw rather than masquerading as ordinary failures. General regexes retain full-subject semantics and may access the whole source. For quoted literals, Regex::literalWindow(string) and the facade's literalWindow() factory provide bounded matching. See file input for source-memory costs.

Start/End use source byte positions. Nothing succeeds with a zero-width span; repeating a nullable parser throws. Combinators retain byte consumption and selected children. A failure consumes zero bytes publicly even when its internal record has observed progress. Diagnostics preserve that distinction.

See source APIs, consumption, custom extensions and the runnable execution examples.