Wrong handling for metacharacters in sets, e.g. [\w0-9]

[Zac-HD]

[\w0-9] gets translated to [[a-zA-Z]0-9], which is invalid. See python-jsonschema/jsonschema#612.

I think fixing this may require us to actually parse the pattern rather than just doing local replacements, and it certainly indicates that we need to test against a much wider range of regular expressions! (i.e. write a better strategy, and contribute a larger set of patterns upstream)

[Zac-HD]

See tests: https://github.com/Zac-HD/js-regex/compare/fix-sets

[jayvdb]

One possible partial workaround, for at least \w and similar, is to use the Python re engine in bytes mode which doesnt support Unicode. That would not help if the regex also needs Unicode support, but often that wont be needed, as JS regexp historically hasnt been used for unicode because it was poorly supported. The user could turn this on explicitly to ensure they are informed about the side effects.

[jayvdb]

Or simpler, use re.ASCII.