V8 implementation: does unicode property escapes behavior in character classes range differ from other classes intentionally?

[vsemozhetbyt]

V8 5.8 — 6.1

> 'a-'.match(/[\s-\w]/gu)
SyntaxError: Invalid regular expression: /[\s-\w]/: Invalid character class

> 'a-'.match(/[\p{Script=Latin}-\p{Script=Cyrillic}]/gu)
[ 'a', '-' ]

Is this by design or should it be posted as a bug?

[mathiasbynens]

Good catch. On first glance, it seems like this should throw — so this would be a bug.

[mathiasbynens]

cc @hashseed @schuay

[hashseed]

Probably a bug in the regexp parser. Will take a look soon.

[mathiasbynens]

This V8 bug is now fixed: https://chromium.googlesource.com/v8/v8.git/+/7924985f9fce4ca6ff8b132f55131bd3e28ed849%5E%21

[vsemozhetbyt]

Thank you!

[mathiasbynens]

The spec already handles this case in step 1 of the CharacterRange abstract operation.

If A does not contain exactly one character or B does not contain exactly one character, throw a SyntaxError exception.

However, it seems like this doesn’t cover property escapes that expand to only a single character (although I don’t think those exist). Should the proposal explicitly throw for those too?

[hashseed]

I think the behavior should not differ depending on what exactly a property class contains.

[littledan]

I agree; the current spec is good.

EDIT: Sorry, I was misreading. I think it's important that whether a Unicode property escape contains multiple code points or not should not affect the grammar, as this may change over versions.

[mathiasbynens]

With tc39/ecma262#984 merged, this is now fixed upstream.