This wouldn't make much of a difference unless there were a lot of strings being concatenated, which is probably not the most common case. However, it seems like pretty low hanging fruit and I know it makes a big difference in IE6/7.

Also DOM traversing by nextSibling is almost twice faster then index based.

This new implementation suffers from the same inconsistency as the current version, reported in ticket:6827. A corrected version would be

var getText = function (elem) {
	var text = []; 
	for (var node = elem.firstChild; node; node = node.nextSibling) {

		// Get the text from text nodes and CDATA nodes
		if (node.nodeType == 3 || node.nodeType == 4)
			text.push(node.nodeValue);

		// Special case for script nodes
		else if (node.tagName === 'SCRIPT')
			text.push(node.text);

		// Traverse everything else, except comment nodes 
		else if (elem.nodeType !== 8)
			text.push(getText(node));

	} 
	return text.join();
};

Should definitely be using elem.textContent if supported.

Replying to [comment:5 paul.irish]:

Should definitely be using elem.textContent if supported.

We had problems getting a consistent value back from the built-in string methods. The only way to get consistent results across all browsers was to do the traversal.

Naturally, a patch to the contrary would be welcome.

This is a relatively simple change (using arrays instead of concatenation). Should be low-hanging fruit for someone - possibly in 1.7.

I'd like to take a crack at this!

didn't realize that this was related to sizzle... bailing out

I created a pull request that fixes this issue at http://github.com/jquery/sizzle/pull/59.

As noted in the pull request, this change may have a negative impact on performance in Chrome, Firefox 4 and IE8+. It will have a minor performance benefit in IE6 and IE7, Safari (as of version 5.0.5) and Opera (as of 11.10).

http://jsperf.com/ultimate-string-concatenation-tests.

https://github.com/jquery/sizzle/pull/59

I updated the pull request based on feedback from dmethvin and created a jsPerf that measures the performance improvement: http://jsperf.com/sizzle-gettext-bugfix-6863.

Nominating ticket for 1.7 discussion.

+1, Changes actually go to Sizzle I believe - if it can be reasonably achieved

+1, If performance gain is proven... and only if it proven.

+1, Yes, but also +1 on jaubourg's sentiment above !

+0, skeptical on the performance gain.

+1, Worth a look with #3144 as well.

See also this forum post with a link to perf tests there and also linked above:

http://forum.jquery.com/topic/sizzle-gettext-open-pull-request-for-ticket-6863#14737000002377958

Would still like to see #3144 tackled at the same time if possible to reduce some of the quirks of IE6/7/8 text behavior using the code suggested by ioquatix and linked in the fiddle at the bottom. It might make the function even faster since it eliminates recursion for all but XML documents.

+1, Well, certainly seems to be running faster. If it's passing, let's get it in.

+1, perf test shows improvements across all browsers

+1

Thanks for taking this timmywil!

TL;DR: There is a pull already for this in Sizzle's tracker:

http://bugs.jquery.com/ticket/6863#comment:10

Ticket #3144 has an alternate implementation using .textContent/.innerText when possible that may be even faster, and it addresses issues with IE6/7/8 losing spaces.

Update Sizzle. Fixes #3144, #6863.

Changeset: 22fcc7744daded0dc0783d85df3bd88c6dbc4544