<Regression> Mozilla-Inbound-Non-PGO - a11y Row Major MozAfterPaint - Ubuntu HW 12.04 x64 - 2.7%

[nob...@cruncher.build.mozilla.org]

Regression: Mozilla-Inbound-Non-PGO - a11y Row Major MozAfterPaint - Ubuntu HW 12.04 x64 - 2.7% increase
--------------------------------------------------------------------------------------------------------
Previous: avg 295.875 stddev 2.112 of 12 runs up to revision 9051a06f0268
New : avg 303.875 stddev 1.884 of 12 runs since revision f5c5742ad004
Change : +8.000 (2.7% / z=3.788)
Graph : http://mzl.la/1gKzcxs

Changeset range: http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=9051a06f0268&tochange=f5c5742ad004

Changesets:
* http://hg.mozilla.org/integration/mozilla-inbound/rev/f3476de5a680
: Simon Sapin <simon...@exyr.org> - Bug 978478 part 1: Add support for repeat() in <line-name-list> (CSS Grid) r=dholbert
: http://bugzilla.mozilla.org/show_bug.cgi?id=978478

* http://hg.mozilla.org/integration/mozilla-inbound/rev/f5c5742ad004
: Simon Sapin <simon...@exyr.org> - Bug 978478 part 2: Add support for repeat() in <track-list> (CSS Grid) r=dholbert
: http://bugzilla.mozilla.org/show_bug.cgi?id=978478

Bugs:
* http://bugzilla.mozilla.org/show_bug.cgi?id=978478 - Support repeat() in CSS Grid templates

[Daniel Holbert]

SimonSapin's patches couldn't have caused this; they simply added the
ability to parse a new syntax for an existing (but preffed off) CSS grid
property, which isn't even implemented in layout yet. (i.e. if you pref
it on, you only get computed style; no useful layout/painting)

So, there's no chance that it affected this MozAfterPaint benchmark.

Looking at the graph ( http://mzl.la/1gKzcxs ) the measurement for
Simon's push (f5c5742ad004) was actually barely within the existing
noise range. In particular, we'd had two previous measurements that were
just as high (304) in the few days before it (one on Mar 31, 2014 14:56,
and one on Mar 27, 2014 05:22). So, that measurement was a spike, but
no more extreme than other recent noise-spikes.

However, things *do* seem to go up to *higher* than the existing noise
range shortly after that. So it appears that something did regress
sometime around there. (which is why the tree-management bot saw fit to
notify us)

The first outside-the-noise point on the graph (with a measurement of
305) was for Matt Woodrow's push d1f8ac35bdd4:

http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?changeset=d1f8ac35bdd4

...which is a bunch of painting-related changes, and seems much more
likely to have impacted this benchmark. Given the height of that
measurement (and the even-higher ones immediatelyafter it - 3x 306, 1x
305.5), it seems like any regression here happened in that push or one
of the ones before it.

Matt, could your push have caused this?

~Daniel

> _______________________________________________
> dev-tree-management mailing list
> dev-tree-...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-tree-management
>

[Matt Woodrow]

I don't see how it could have, those patches should only affect <canvas>
drawing. And I would expect them to only have functional changes for
when we have OMTC too.

- Matt

[Daniel Holbert]

OK, thanks.

I'm going to assume this was a false alarm, given that
(a) no MozAfterPaint regression has been reported on m-c, for when this
range of csets was merged over there.

(b) The graph seems to have returned to its normal level (last few
measurements were all around ~300, which is where it was clustering back
on Mar 27 at least.

~Daniel