Reland "Revert "Skip MISSED frames in browser compositor""

This reverts commit 1374013c and
disables some media tests that flake due to a race that is exposed
by the revert.

Reason for revert: causes jank https://crbug.com/947587
The original reason for landing this CL does not apply anymore. It is
no longer the case that viz skips a BeginFrame if an undrawn frame
exists. See https://crrev.com/c/1521967

Original change's description:
> Skip MISSED frames in browser compositor
>
> This CL fixes a dropped frame issue when a scroll transitions from
> dragging to a momentum "fling" phase. When the user is scrolling and
> lifts their finger with velocity, the scroll continues according to a
> decelerating animation curve. When this happens, the FlingController
> receives a FlingStart instead of a ScrollUpdate and it uses that to
> create an animation curve.
>
> When this happens the browser process registers itself as an animation
> observer so that it begins to receive BeginFrame signals. However, the
> display service issues a MISSED BeginFrame for the frame in which the
> animation was registered. This will frequently miss submission in the
> current frame, so it submit next frame and won't receive a BeginFrame
> then. This leads to a frame without the browser ticking the fling
> animation. See https://crbug.com/882907#c50 for a diagram and more
> complete explanation of the issue.
>
> The browser process shouldn't require processing MISSED frames so in
> this CL we simply drop MISSED frames if we're in the browser compositor
> (commit_to_active_tree is true). Longer term, the display scheduler
> should make these decisions but this is an immediate fix for the fling
> issue.
>
> Bug: 882907
> Change-Id: Ieb6d14051dc05c5e177da1920dbd27684f941d29
> Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1517164
> Commit-Queue: David Bokan <bokan@chromium.org>
> Reviewed-by: Sunny Sachanandani <sunnyps@chromium.org>
> Cr-Commit-Position: refs/heads/master@{#640172}

TBR=bokan@chromium.org,sunnyps@chromium.org


Bug: 947587,882907,963141
Change-Id: Id23b586f89de67f9e77212ee265f4ad2db5a6fe7
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1613363Reviewed-by: Saman Sami <samans@chromium.org>
Reviewed-by: David Bokan <bokan@chromium.org>
Commit-Queue: Saman Sami <samans@chromium.org>
Cr-Commit-Position: refs/heads/master@{#660014}

cc/scheduler/scheduler.cc

@@ -933,17 +933,6 @@ bool Scheduler::ShouldDropBeginFrame(const viz::BeginFrameArgs& args) const {
     return true;
   }
 
-  // We shouldn't be handling missed frames in the browser process (i.e. where
-  // commit_to_active_tree is set) since we don't know if we should create a
-  // frame at this time. Doing so leads to issues like crbug.com/882907. This
-  // early-out is a short term fix to keep fling animations smooth.
-  // TODO(bokan): In the long term, the display compositor should decide
-  // whether to issue a missed frame; it is tracked in
-  // https://crbug.com/930890.
-  if (args.type == viz::BeginFrameArgs::MISSED &&
-      settings_.commit_to_active_tree)
-    return true;
-
   return false;
 }
 

cc/scheduler/scheduler_unittest.cc

@@ -639,26 +639,6 @@ TEST_F(SchedulerTest, VideoNeedsBeginFrames) {
   EXPECT_FALSE(scheduler_->begin_frames_expected());
 }
 
-// As a short term fix for https://crbug.com/882907, we should skip MISSED
-// frames from the browser compositor.
-// TODO(bokan): In the long term, the display compositor should decide
-// whether to issue a missed frame; it is tracked in
-// https://crbug.com/930890.
-TEST_F(SchedulerTest, BrowserCompositorSkipsMissedBeginFrames) {
-  scheduler_settings_.commit_to_active_tree = true;
-  SetUpScheduler(EXTERNAL_BFS);
-  scheduler_->SetNeedsBeginMainFrame();
-
-  task_runner_->AdvanceMockTickClock(viz::BeginFrameArgs::DefaultInterval());
-  viz::BeginFrameArgs args =
-      fake_external_begin_frame_source_->CreateBeginFrameArgs(
-          BEGINFRAME_FROM_HERE, task_runner_->GetMockTickClock());
-  args.type = viz::BeginFrameArgs::MISSED;
-
-  fake_external_begin_frame_source_->TestOnBeginFrame(args);
-  EXPECT_FALSE(client_->IsInsideBeginImplFrame());
-}
-
 TEST_F(SchedulerTest, RequestCommit) {
   SetUpScheduler(EXTERNAL_BFS);
 

third_party/blink/web_tests/TestExpectations

@@ -5578,6 +5578,11 @@ crbug.com/962726 [ Linux Win ] external/wpt/event-timing/timingconditions.html [
 crbug.com/937170 [ Linux Win7 ] external/wpt/IndexedDB/interleaved-cursors-large.html [ Pass Timeout ]
 crbug.com/937991 [ Win7 Release ] http/tests/devtools/cache-storage/cache-data.js [ Pass Timeout ]
 
+# These video tests seem to take a screenshot too early.
+crbug.com/963141 [ Mac Win ] media/video-object-fit.html [ Pass Failure ]
+crbug.com/963141 [ Mac Win ] media/video-object-fit-change.html [ Pass Failure ]
+crbug.com/963141 [ Mac Win ] media/controls/video-overlay-cast-light-rendering.html [ Pass Failure ]
+
 # Disable tests in order to land
 # https://webrtc-review.googlesource.com/c/src/+/135948
 crbug.com/943975 external/wpt/webrtc/RTCDataChannel-send.html [ Pass Failure ]