Add tests for RegExp u flag

[jugglinmike]

I'd like to draw attention to two tests in particular since they both concern a very specific condition (for example, good ol' step 8.g.iv.5.d):

• test/built-ins/RegExp/prototype/Symbol.match/u-lastindex.js
• test/built-ins/RegExp/prototype/Symbol.replace/u-lastindex.js

Despite a careful reading, I'm still not convinced those test cases properly express the detail documented in their info/description. I decided I may have looked at it for too long and that some fresh perspective would help

@mathiasbynens Would you mind taking a look?

[anba]

lastIndex is reset to 0 in 21.2.5.2.2 RegExpBuiltinExec, step 15.a.i when the loop in 21.2.5.6, step 8.g is executed the third (and last) time.

[jugglinmike]

Ah, okay. So I think this behavior can't be externally verified through inspection of the lastIndex property. What about this instead:

test/built-ins/RegExp/prototype/Symbol.match/u-lastindex.js, renamed to test/built-ins/RegExp/prototype/Symbol.match/u-advance-after-empty.js

var match = /^|\udf06/ug[Symbol.match]('\ud834\udf06');

assert(match !== null);
assert.sameValue(match.length, 1);
assert.sameValue(match[0], '');

test/built-ins/RegExp/prototype/Symbol.replace/u-lastindex.js, renamed to test/built-ins/RegExp/prototype/Symbol.replace/u-advance-after-empty.js

var str = /^|\udf06/ug[Symbol.replace]('\ud834\udf06', 'XXX');
assert.sameValue(str, 'XXX\ud834\udf06');

[anba]

Yes, that should work. Alternatively it's also possible to use /\b/ and /\B/ assertions to execute those steps. Like in: https://github.com/anba/es6draft/blob/master/src/test/scripts/suite/regress/bug4159.js

[jugglinmike]

I've fixed the bugs that @anba identified in a fixup commit

[goyakin]

These are good tests @jugglinmike. It might be better to add the following tests to improve coverage:

• Coercion of the unicode property to Boolean
• length of the unicode get accessor function
• Symbol.search

[jugglinmike]

Sounds good to me. I've added tests for those three behaviors

[mathiasbynens]

You could also add a test to see if running RegExp.prototype.unicode; (i.e. this is RegExp.prototype) throws.

[jugglinmike]

Hey Mathias! That test actually describes to distinct expectations:

1. RegExp.prototype is not a RegExp instance
2. get RegExp.prototype.unicode requires the "this" value to have an [[OriginalFlags]] internal slot

I think it will be more clear (and maintainable) to test RegExp.prototype instanceof RegExp in one place and then only test for the second behavior generically (i.e. with any object missing the expected internal slot) in all other cases. The former is outside the scope of this pull request (though there is an open issue for it: #278), but this patch includes a test for the latter (https://github.com/tc39/test262/pull/352/files#diff-09b2f28b81fb4f5d4d84056cdced1ba6).

[bterlson]

So I think this is good for 🍆?

[jugglinmike]

Done and done