String#matchAll polyfill does not appear to correctly set or update the lastIndex property

[Andrew-Cottrell]

closure-compiler/src/com/google/javascript/jscomp/js/es6/string/matchall.js

Line 48 in 3ae91f3

var /** !RegExp */ regexCopy =

The String#matchAll polyfill fails this test case

(function () {
    "use strict";
    var regexp = /\S+/g;
    var string = "one two three";
    regexp.lastIndex = string.indexOf(" ") + 1; // match all after the first space character
    var results = Array.from(string.matchAll(regexp));
    console.log(results.length); // Expected 2
    console.log(results[0]); // Expected ["two", index: 4, ...]
    console.log(results[1]); // Expected ["three", index: 8, ...]
}());

The polyfill does not appear to implement steps 7-8 of RegExp.prototype [ @@matchAll ].

7. Let lastIndex be ? ToLength(? Get(R, "lastIndex")).
8. Perform ? Set(matcher, "lastIndex", lastIndex, true).

The polyfill also doesn't appear to fully implement step 11.a.ii.2 of %RegExpStringIteratorPrototype%.next or AdvanceStringIndex, which require incrementing lastIndex by 2 when the Unicode flag is set and the next 2 code units form a surrogate pair.

Note, the comparison at

closure-compiler/src/com/google/javascript/jscomp/js/es6/string/matchall.js

Line 64 in 3ae91f3

if (regexCopy.lastIndex === previousIndex) {

does not reliably detect empty-string matches in at least Internet Explorer. The following approach, using match.index, seems to work reliably

var fullUnicode = Boolean(regexCopy.unicode); // 8.         Let fullUnicode be O.[[Unicode]].
var matchStr = match[0];                      // 11.a.i.    Let matchStr be ? ToString(? Get(match, "0")).
if (!matchStr) {                              // 11.a.ii.   If matchStr is the empty String, then
    regexCopy.lastIndex = match.index + 1 + ( // 11.a.ii.1. Let thisIndex be ? ToLength(? Get(R, "lastIndex")).
        fullUnicode &&                        // 11.a.ii.2. Let nextIndex be ! AdvanceStringIndex(S, thisIndex, fullUnicode).
        isSurrogatePair(string, match.index)  // 11.a.ii.3. Perform ? Set(R, "lastIndex", nextIndex, true).
    );
}

Of course, without any additional polyfill, fullUnicode would be false in Internet Explorer.

[concavelenz]

Thank you for the detailed report.