Make LayerTreeImpl::ScrollableArea more robust

Make LayerTreeImpl::ScrollableArea more robust

If the first child of the root scroll layer is not drawn for some reason, its bounds are (correctly) no longer updated due to https://codereview.chromium.org/13939005. The problem is that we currently depend on these bounds to calculate the scrollable area.

In an ideal world, we'd check the bounds of the root scroll layer itself. There was a Blink bug that was giving us (0, 0) bounds for this layer occasionally, so we switched the code to check the bounds of the first child. This patch makes that workaround a bit more robust, but really, we should fix the Blink bug.

R=enne,aelias

BUG=None

[Ian Vollick, 2013-04-20 03:10:08]

+aelias

I've added a unit test. I think that this CL is ready to go, but I still think
it's yucky. Alex, is the Blink bug still present? If so, do you think a fix is
doable in the short term?

[aelias_OOO_until_Jul13, 2013-04-20 05:03:07]

This says BUG=None but I presume this is solving a bug, what is it?  And I don't
understand why looping through makes sense?

I don't think anybody fixed the Blink bug but it doesn't seem like something
that would be intractable, just figuring out where the race happens with enough
logging.  I just filed https://crbug.com/233853 and assigned it to you since you
started looking at this (though feel free to reassign if you don't have the
cycles).

[enne (OOO), 2013-04-22 17:14:40]

If possible, I'd really just prefer to fix this in Blink.

[enne (OOO), 2013-05-28 18:47:44]

Ping?

[aelias_OOO_until_Jul13, 2013-09-09 22:41:33]

Closing this patch to clean up my review queue.