...
Actually, unless it's ambiguous, an ABNF *does* define how to parse.
Actually, no, the purpose of an ABNF is to define the grammar for
generating valid strings and testing strings for validity. It might
be used as a guide by something like lex to create a parser that
enforces validity while parsing, but that generally is not done in
Internet-facing software because of Postel's Law.
...
For example, RFC 3986 has a very specific grammar for generation
and validity of URIs, but also describes one parsing algorithm
(not the only one, but certainly one in common use) in an
Appendix that will accept any string and parse it into the major
components.
And I'll reiterate, again, that the algorithm for reference parsing
in HTML5 is not definitive of URLs -- it is just a variation on the
appendix in RFC3986 that includes a non-ASCII character encoding
step. The entire notion that this has anything to do with IRI or URI
definition, or that we need to fix any of the IETF specs to
incorporate browser-specific reference error-handling, is simply
absurd. They are not the same thing.