[Webkit-unassigned] [Bug 38557] r58526 introduced a ~30% regression on Dromaeo JS lib

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=38557


anton muhin <antonm at chromium.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |vitalyr at chromium.org




--- Comment #12 from anton muhin <antonm at chromium.org>  2010-05-06 12:40:09 PST ---
[+ Vitaly]

So we carried out couple of experiments.  Current results:

1) speeding up fast checks doesn't help: see patch above, I ran another patch
which uses bit flag when cached node lists are present---regression still
persists;

2) heap profiler (in Chromium) hints that Dromaeo tests retain documents (one
per suite like "DOM Attributes (jQuery)" and those apparently keep node lists;

3) if one runs 'regressed suite' separately, omitting suites before, there is
no regression: http://dromaeo.com/?id=102605,102610

So the best hypothesis for now (which I will check tomorrow) is we've got
'leaked' cached node lists which sit and make subsequent GCs slower.

If that's true, ideally Dromaeo should have cleaned up its garbage (and I'll
experiment with it nulling out globals in suites).

But anyway, esp. as JSC heap is not segregated by DOMWindow yet, we probably
need to collect node lists if they don't have custom properties set.

Thoughts?

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.