tools/benchmarks: Fix failures in image decoding measurements.

tools/perf/benchmarks/image_decoding.py

 # Copyright 2013 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 from core import perf_benchmark
 from telemetry import benchmark
 
 from measurements import image_decoding
 import page_sets
 
 
 @benchmark.Owner(emails=['cblume@chromium.org', 'reveman@chromium.org'])
 class ImageDecodingToughImageCases(perf_benchmark.PerfBenchmark):
   test = image_decoding.ImageDecoding
   # TODO: Rename this page set to tough_image_cases.py
   page_set = page_sets.ImageDecodingMeasurementPageSet
 
   @classmethod
   def Name(cls):
     return 'image_decoding.image_decoding_measurement'
 
   @classmethod
   def ShouldDisable(cls, possible_browser):
     # crbug.com/667501
     return possible_browser.platform.GetDeviceTypeName() == 'Nexus 7'
 
   def GetExpectations(self):
     return page_sets.ImageDecodingMeasurementStoryExpectations()
+
+  def CustomizeBrowserOptions(self, options):
+    options.AppendExtraBrowserArgs([
+        # Disable asynchronous decodes in the renderer since these test
+        # rely on images have been decoded between consecutive
+        # requestAnimationFrames.
+        '--disable-checker-imaging'

[nednguyen, 2017/06/28 19:28:11]

Chris: I hope the overhauled version of image decode benchmark will not require
this flag? Otherwise, I worry that we may drift away from testing what we ship
to the users.

[Khushal, 2017/06/28 20:29:55]

If the only thing this benchmark wants to measure is the performance of the
decoding algorithm, then this flag should not have an impact. It only impacts
how these decodes are scheduled.

[cblume, 2017/06/28 21:15:28]

I think our concern is if the existing perf was thrown off, the new, overhauled
version will be thrown off as well.

And to add to the concern, I don't think the new version has a way to specify
flags like this.

It might force us to either 1.) find a better way to get the decoder time to the
page, or 2.) abandon this test (which has been a real consideration for a while
now)

+    ])