GitLab

After http://crrev/c/2256225 we had two classes of flakes.
1) The impl_destoryed_ CopyOutputResult would be non empty.
2) The callback_count_ would be incorrect.

The cause of 1) was in the change that introduced the flake we called
SetViewportRectAndScale() after RunUntilIdle(), but
SetViewportRectAndScale() was what prevented the CopyOutputResult
returned to the impl callback from being non-empty. So if we
RunUntilIdle() fast enough we would still prevent the drawing and the
test would pass. So the fix is to immediately call
SetViewportRectAndScale() after destroying the main to ensure its
always fast enough.

The cause of 2) was that RunUntilIdle() runs JUST the currently posted
tasks, so there was a race between RunUntilIdle() being called and
during the deconstruction of the layer PostTasking the callback. The
fix is to only wait for the main callback in frame 3 so we can verify
that it will happen before we destroy the impl layer. And wait for the
impl at the end of the test.

Bug: 1096962
Change-Id: Iad066a2316f2e4df3bdf53379c87a4697d6da8bb
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/2294981Reviewed-by: Wez <wez@chromium.org>
Reviewed-by: kylechar <kylechar@chromium.org>
Commit-Queue: Stephen Nusko <nuskos@chromium.org>
Auto-Submit: Stephen Nusko <nuskos@chromium.org>
Cr-Commit-Position: refs/heads/master@{#790371}