Fix Canvas2DAllowed test on mac

Fix Canvas2DAllowed test on mac

BUG=106987
TBR=avi
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=113824

[epoger, 2011-12-09 18:19:33]

Avi- I think this should fix it, as you guessed.  Shall I wait for the trybots
or just commit it?  I think just commit it...

[Sergey Ulanov, 2011-12-09 18:39:03]

On 2011/12/09 18:19:33, epoger wrote:
> Avi- I think this should fix it, as you guessed.  Shall I wait for the trybots
> or just commit it?  I think just commit it...

Can you please land this or disable this test ASAP? I think it's fine to TBR
this change because the test is broken anyway.

[Avi (use Gerrit), 2011-12-09 19:11:25]

LGTM

In the future, it is crucial that you do not leave tests red. Either revert, or
disable.

BTW, how was this test failure not caught in your pre-landing tests?

[epoger, 2011-12-09 19:24:12]

On 2011/12/09 19:11:25, Avi wrote:
> LGTM
> 
> In the future, it is crucial that you do not leave tests red. Either revert,
or
> disable.

Sure.  Normally I would have just reverted the CL, but there were other issues
at play here... this CL had to be landed in sync with webkit changes and
deps-rolls in both directions. I had tried to handle things differently as noted
in https://bugs.webkit.org/show_bug.cgi?id=73226 ('make chromium layout_tests
observe use_skia_on_mac.gypi from chromium/build'), but that approach was shot
down by webkit reviewers.

> BTW, how was this test failure not caught in your pre-landing tests?

An excellent question.  This test failure was obscured by a compile failure in
the trybots, but I had thought that the trybots might fail to compile anyway for
this gyp-level change.  So I committed it anyway.

The moral of the story is: trying to make simultaneous, co-dependent changes in
webkit and chromium is tricky.