Mark WebGLRenderingContext as having lost context upon init failure.

Mark WebGLRenderingContext as having lost context upon init failure.

r196077 added a good test verifying the handling of attempted WebGL context
creation when that fails to create an underlying drawing buffer. This
exposed an Oilpan-specific lifetime issue for the WebGLRenderingContext
object that fails to properly initialize. With Oilpan, that garbage
collected & partially initialized object will not be synchronously deleted,
but be finalized instead when the next GC strikes.

The WebGLRenderingContext object is an ActiveDOMObject attached to its
execution context, so should it be stopped before that GC strikes, it
will be forcibly removed from its (graphics system) context as part
of stopping. The partially constructed object is not prepared for that,
with failing consequences.

To avoid, if the creation of a drawing buffer fails, we explicitly mark
the object as having lost its context.

R=bajones,kbr,haraken
BUG=474665
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=196188

[sof, 2015-05-29 16:21:12]

sigbjornf@opera.com changed reviewers:
+ bajones@chromium.org, kbr@chromium.org, oilpan-reviews@chromium.org

[sof, 2015-05-29 16:21:13]

please take a look.

Takes care of the Oilpan webgl failures


https://storage.googleapis.com/chromium-layout-test-archives/WebKit_Linux_Oil...

[bajones, 2015-05-29 18:16:43]

Good catch, thanks for the fix! LGTM

[sof, 2015-05-29 18:30:19]

On 2015/05/29 18:16:43, bajones wrote:
> Good catch, thanks for the fix! LGTM

Ditto for the test added, framed the problem very well.

(Will wait for owner approval.)

[Ken Russell (switch to Gerrit), 2015-05-29 22:52:25]

Nice. Subtle issue. I wonder if there are going to be more when we turn on
Oilpan for WebGL. LGTM

[haraken, 2015-05-30 00:29:30]

LGTM.

On 2015/05/29 22:52:25, Ken Russell wrote:
> Nice. Subtle issue. I wonder if there are going to be more when we turn on
> Oilpan for WebGL. LGTM

Our plan is:

1) Enable lazy sweeping on trunk.
2) Ship Oilpan for core/animations/.
3) Ship Oilpan for html/canvas/.
4) Ship Oilpan for everything.

We're currently workign on 1) and 2). We'll update performance/memory numbers
before shipping Oilpan for html/canvas/ of course.

[sof, 2015-05-30 05:48:23]

The CQ bit was checked by sigbjornf@opera.com

[commit-bot: I haz the power, 2015-05-30 05:48:28]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1153233005/1

[commit-bot: I haz the power, 2015-05-30 05:51:36]

Committed patchset #1 (id:1) as
https://src.chromium.org/viewvc/blink?view=rev&revision=196188

[sof, 2015-05-30 06:26:34]

On 2015/05/29 22:52:25, Ken Russell wrote:
> Nice. Subtle issue. I wonder if there are going to be more when we turn on
> Oilpan for WebGL. LGTM

It would be prudent to assume so, test coverage (which is excellent for WebGL)
can only get you so far. We're fuzzing Oilpan builds also, but should probably
do more of that.