Huge, but mostly mechanical change. Also brings in a phf
update unfortunately, albeit a fairly low risk one.
This patch makes us stop carrying tokens and source locations the error
path, which adds a lot of bloat to the generated code unnecessarily, for
a feature that is almost always off by default. The worst offender is
the Tokenizer::next() error, which starts getting passed by registers
after this change.
Instead of carrying the location in the error, rule and declaration
parsers hand out the source location of the start of the offending
declaration, and that's what ParseErrorReporter gets.
This means the reported line / column numbers now change, which needs
tests changes. The unprintables test also gets fixed and cleaned up now
that we don't have a ton of different codepaths to escape.
Differential Revision: https://phabricator.services.mozilla.com/D320989
Emilio Cobos Álvarez
94c7ff59f3