Non-deterministic matching starting in v1.21.2

[caroline-matsec]

We've recently found that there are cases where the matches returned from the same exec call are non-deterministic. Sometimes a part of the result is truncated, and sometimes it returns special characters that were not in the original input string.

Here's an example script that runs the same value through 1 million times and prints when the output is unexpected. I've run this several times and have also provided some sample output below. It fails on v1.21.2 and 1.21.3, but succeeds on 1.21.1, 1.21.0 and 1.20.9.

const RE2 = require('re2');

// Regex from RFC https://www.rfc-editor.org/rfc/rfc2396#appendix-B
const URL_RFC_PATTERN = new RE2('^(([^:/?#]+):)?(//([^/?#]*))?([^?#]*)(\\?([^#]*))?(#(.*))?');

function parseUrl(str) {
    const matches = URL_RFC_PATTERN.exec(str);
    const rv = {
        part1: matches[2],
        part2: matches[4],
    };
    return rv;
}

for (let i = 0; i < 1000000; i++) {
    const result = parseUrl('https://my-random-example.site/');
    if (result.part1 !== 'https') {
        console.log('Unexpected result for part1 on attempt #' + i + ': ' + result.part1);
    }
    if (result.part2 !== 'my-random-example.site') {
        console.log('Unexpected result for part2 on attempt #' + i + ': ' + result.part2);
    }
}

Sample output:

$ node repro.js
Unexpected result for part1 on attempt #1396: `��)�
Unexpected result for part2 on attempt #1396: ~m-example.site
Unexpected result for part2 on attempt #2528: om-example.site
Unexpected result for part1 on attempt #844837: ���)�
Unexpected result for part2 on attempt #844837: m-example.site

[uhop]

Thank you for identifying the affected versions — it’ll help me to analyze the changes.

[uhop]

I think I fixed the problem. Please verify the fix and let me know if it works for you.

[caroline-matsec]

I think I fixed the problem. Please verify the fix and let me know if it works for you.

Yep, I'm no longer seeing any failures. Thanks!