gpu test: fix flaky failure of Pixel.WebGLYellowTriangleContextRestore.

gpu test: fix flaky failure of Pixel.WebGLYellowTriangleContextRestore.

If we prevent default for webglcontextlost event, we should not restore forcely,
because it causes INVALID_OPERATION exception: "restoreContext: context not lost".

BUG=349195

[dshwang, 2014-03-11 17:28:48]

This CL follows up https://codereview.chromium.org/188073002

[dshwang, 2014-03-11 17:30:56]

https://codereview.chromium.org/194593006/diff/1/content/test/data/gpu/pixel_...
File content/test/data/gpu/pixel_webgl_context_restore.html (right):

https://codereview.chromium.org/194593006/diff/1/content/test/data/gpu/pixel_...
content/test/data/gpu/pixel_webgl_context_restore.html:45: ev.preventDefault();
we prevent default, so we should trust "webglcontextrestored" called
automatically.

[Ken Russell (switch to Gerrit), 2014-03-11 17:35:13]

BTW, this test was reverted this morning, so I think this CL won't apply any
more. Brandon, a couple of hints about why the test might have been flaky.

https://codereview.chromium.org/194593006/diff/1/content/test/data/gpu/pixel_...
File content/test/data/gpu/pixel_webgl_context_restore.html (right):

https://codereview.chromium.org/194593006/diff/1/content/test/data/gpu/pixel_...
content/test/data/gpu/pixel_webgl_context_restore.html:45: ev.preventDefault();
On 2014/03/11 17:30:57, dshwang wrote:
> we prevent default, so we should trust "webglcontextrestored" called
> automatically.

Sorry, the WEBGL_context_lost extension doesn't work that way. See
http://www.khronos.org/registry/webgl/extensions/WEBGL_lose_context/ . However,
there is still a bug in this test which I should have caught during code review.
The timeout should be started in the webglcontextlost event handler, because
otherwise there's a race condition (which was the original problem with the
test).

[bajones, 2014-03-11 17:37:53]

Yes, the original test has been reverted but I will take this into account when
I re-land it.

We can't use this directly for the reasons Ken mentioned but it did help
highlight to me the race condition involved. Thanks!

[dshwang, 2014-03-11 17:40:44]

On 2014/03/11 17:35:13, Ken Russell wrote:
> BTW, this test was reverted this morning, so I think this CL won't apply any
> more. Brandon, a couple of hints about why the test might have been flaky.
> 
>
https://codereview.chromium.org/194593006/diff/1/content/test/data/gpu/pixel_...
> File content/test/data/gpu/pixel_webgl_context_restore.html (right):
> 
>
https://codereview.chromium.org/194593006/diff/1/content/test/data/gpu/pixel_...
> content/test/data/gpu/pixel_webgl_context_restore.html:45:
ev.preventDefault();
> On 2014/03/11 17:30:57, dshwang wrote:
> > we prevent default, so we should trust "webglcontextrestored" called
> > automatically.
> 
> Sorry, the WEBGL_context_lost extension doesn't work that way. See
> http://www.khronos.org/registry/webgl/extensions/WEBGL_lose_context/ .
However,
> there is still a bug in this test which I should have caught during code
review.
> The timeout should be started in the webglcontextlost event handler, because
> otherwise there's a race condition (which was the original problem with the
> test).

Ah, ok. feel free to pick up some code from this CL if needed.
I'll close this CL soon.