Bug 87742 – [perf-test] Add a benchmark for querySelector()

• Status: RESOLVED
• Resolution: FIXED
• Component: Tools / Tests

[Kentaro Hara]

Now I am optimizing querySelector() (bug 87625). We want to add a performance test for querySelector().

[Kentaro Hara]

Created attachment 144550 [details]
Patch

[Ryosuke Niwa]

Comment on attachment 144550 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=144550&action=review

We probably want one more case where the node is deep in the tree (i.e. 20-50 levels of nesting).

> PerformanceTests/Parser/query-selector-last.html:5
> +</body>

Why are you closing body here?

> PerformanceTests/Parser/query-selector-last.html:7
> +for (var i = 0; i < 1000; i++) {

Is 1000 enough to make a difference?

> PerformanceTests/Parser/query-selector-last.html:11
> +  var div = document.createElement("div");
> +  div.id = "foo";
> +  div.className = "bar";
> +  document.body.appendChild(div);

Nit: 2-space indentation.

[Kentaro Hara]

(In reply to comment #2)
> We probably want one more case where the node is deep in the tree (i.e. 20-50 levels of nesting).

I am happy to add it if you want, but would there be any big difference between "a node at the tail of the long flat tree" (tested by query-selector-last.html) and "a node deeply nested in the tree"?

> > PerformanceTests/Parser/query-selector-last.html:7
> > +for (var i = 0; i < 1000; i++) {
> 
> Is 1000 enough to make a difference?

Yes. As I described in ChangeLog:

[query-selector-first.html]
        RESULT Parser: query-selector-first= 252.339803014 runs/s
        median= 253.646265594 runs/s, stdev= 2.32610873209 runs/s, min= 248.138957816 runs/s, max= 254.777070064 runs/s

[query-selector-last.html]
        RESULT Parser: query-selector-last= 356.009616076 runs/s
        median= 355.781448539 runs/s, stdev= 0.841519086864 runs/s, min= 354.430379747 runs/s, max= 358.056265985 runs/s

Note that the loop count of query-selector-last.html is 1/10 of query-selector-first.html. (Maybe we should make the loop count the same, to compare the two results at a glance.)

[Kentaro Hara]

Created attachment 144713 [details]
patch for landing

[WebKit Review Bot]

Comment on attachment 144713 [details]
patch for landing

Clearing flags on attachment: 144713

Committed r118899: <http://trac.webkit.org/changeset/118899>