Zero width matches produce an error

[gskinner]

Zero width matches produce an error

[gskinner]

This is correct behaviour.

A zero width match will match at the same position infinitely. For example, the pattern "z?" matching against the string "aaa", will match at position 0 infinitely. Because it matches 0 characters, it can satisfy the match without moving the search index, causing it to match at the same position again.

[bkiers]

A zero width match will match at the same position infinitely.

Well, that depends on how you match against the target string. For example, the following:

"aaa".match(/(?=a)/g);

will return an array containing 3 empty strings.

But version 1 of Regexr did not behave this way. Is this also a change because of the use of the JavaScript regex engine?

[gskinner]

This is true, however if you use the same example with .exec() you'll see that lastIndex never advances past 0.

var regex = /(?=a)/g;
regex.exec("aaa");
regex.lastIndex; // 0

The "infinite" error is driven from this condition. Specifically, if the lastIndex is identical for two executions, then we surface the infinite error.

I have to assume that .match automatically advances the index. Warrants some testing.

Version 1 behaves the same. If you plug that pattern in, you'll see it changes to red to indicate the same error - it's just less explicit about it. The main difference is that v1 will show any matches up to the infinite loop, whereas v2 won't. I thought it made more sense to show no matches in an error condition, to draw attention to the error (vs users winding up in infinite loops).

[bkiers]

The main difference is that v1 will show any matches up to the infinite loop, whereas v2 won't.

Ah, that's what the red color means in v1 :)

Odd how JavaScript does not advance when the pattern is zero-width: most regex flavors I'm familiar with do so. Ah well, too bad.

Thanks for the info!

[connec]

A 0-length regex is perfectly valid and can very useful. Iterating through the matches is also not difficult:

// Find positions for commas in a number (e.g., 12,345,678 -> 2, 5)
var s  = '12345678'
  , re = /\B(?=(?:\d{3})+(?!\d))/g
  , m;
while(m = re.exec(s)) {
  console.log('match at', m.index);
  if(m[0].length == 0 && ++re.lastIndex > s.length)
    break;
}

I've made a pull request that implements this, along with some naive highlighting.