Working JS library for flag x based on details in the proposal

[slevithan]

I recently created the regex package, which is a template tag for regexes as raw strings. Among other features, it always implicitly enables flag x based on details in this proposal (which are fairly limited right now, so I also based it on flag xx from Perl and PCRE).

It always uses flag v (if available) or u implicitly, so it can't be used to test x in Unicode-unaware mode. And since it uses template strings, it doesn't need to worry about ()[]/ in comments (since comments are stripped before passing to the RegExp constructor). But with those caveats, you can use it to test x behavior even for edge cases.

Just for example:

• regex`[a- -b]` is an error (range of a to unescaped/invalid -).
• regex`[a& &b]` is equivalent to /[a&b]/.
• regex`[a && b]` is equivalent to /[a&&b]/v.
• regex`\0 1` is equivalent to /\0(?:)1/.
• \c A and (? :) are errors, and ( ?:) is an error because you can't quantify (.
• Quantifiers following whitespace and/or comments apply to the preceding token, so x + is equivalent to x+.
• Whitespace and/or comments are allowed to separate a quantifier and the ? that makes it lazy.
• Only space and tab are insignificant within character classes and [\q{…}], not # or other whitespace.
• Outside of character classes, the insignificant whitespace characters are those matched natively by \s.
• Excluding [\q{…}], whitespace is significant in enclosed tokens.
  • Outside of character classes: \u{…}, \p{…}, \P{…}, (?<…>), \k{…}, and {…}.
  • Within character classes: \u{…}, \p{…}, and \P{…}.

If additional details are clarified in this proposal and they don't match regex's handling, I will update it to stay in line.