Invalid unicode escapes in regexp are not a syntax error

[Earlopain]

For example:

/\u{21!3}/

$ ruby -c test.rb
Syntax OK

$ ruby --parser=parse.y -c test.rb
ruby: test.rb:1: invalid Unicode list: /\u{21!3}/ (SyntaxError)

Prism knows this is invalid but intentionally ignores the error:

prism/src/prism.c

Lines 9810 to 9814 in 3340d87

	if (flags & PM_ESCAPE_FLAG_REGEXP) {
	// If this is a regular expression, we are going to
	// let the regular expression engine handle this
	// error instead of us.
	pm_buffer_append_bytes(regular_expression_buffer, start, (size_t) (parser->current.end - start));

This is because this regex is valid (invalid escape after comment in extended mode):

/# \u{21!3}/x

At runtime everything is right but it would be nice to also report this at parse-time.

[Earlopain]

I don't think this and #2104 are able to be implemented realistically. There errors come from onigmo and not the parser itself.

parse.y calls into ruby regex code here https://github.com/ruby/ruby/blob/c32eb727bdde8b3e840198aaffb049a955d94783/parse.y#L15426-L15432. onigmo is a library but as far as I know ruby regularly applies its own patches (if it is even developed upstream anymore, not sure), so basically it is its own thing. Resolving this seems pretty complicated.

Basically prism provides no checking of regex content (and if it did it would need to be properly versioned which doesn't make it any easier). Maybe you could add this somewhere into ruby so that -c is aware of it similar to how parse.y does it but for prism itself I don't see a super obvious solution. Reimplementing regexp parsing in in its entirety seems like a huge waste of time.

regexp.c includes some errors, not sure by what metric these were added there. Probably the bare minimum to make cruby tests pass?

[kddnewton]

Yup, just the bare minimum to get everything passes.