The wording above is slightly confusing, so I'll try to state it more explicitly. When SurfaceView fullscreen is enabled:
  - There are 2.6x as many browser crashes.
  - There are 0.7x as many renderer crashes (this is within the error bars).
  - There are 1.1x as many GPU crashes (also within the error bars).
  - Cumulative WatchTime is 0.8x (I have not done a variance analysis).

Since this only includes Dev and Canary data, there total counts are quite small (~5k users per day in each group). Since the improvement in browser crashes is clearly significant, it may make sense to raise the size of the test group (currently 10%).

However, it's hard to predict the difference between Dev users and Stable users. It may be better to only do the analysis liberato@ has suggested (which seems pretty safe with Dev users) and wait for Stable to make other decisions.

https://docs.google.com/a/google.com/spreadsheets/d/1uXUmi4ykczLqPTHa4I3Xuvfy2aj_WDZ5HYz9RY2oDBo/edit?usp=sharing

TL;DR: QC crashes with SV are 1.43x of what they are without it, which is the highest of all the GL vendors.  overall rate is 1.25x.

will try to split by device.  i don't seem to be able to get a list out, unfortunately.

SurfaceView disabled crashes: https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20OMIT%20RECORD%20IF%20SUM(custom_data.ChromeCrashProto.experiments.ids%3D%27ded5a5d1-3d47f4f4%27)%20%3D%200&ignore_case=false&enable_rewrite=false&omit_field_name=&omit_field_value=&omit_field_opt=

SurfaceView enabled crashes: https://crash.corp.google.com/browse?q=product.name%3D%27Chrome_Android%27%20OMIT%20RECORD%20IF%20SUM(custom_data.ChromeCrashProto.experiments.ids%3D%27ded5a5d1-3f4a17df%27)%20%3D%200&ignore_case=false&enable_rewrite=false&omit_field_name=&omit_field_value=&omit_field_opt=


By far the largest increase (40x) in crashes is for cc::LayerTreeHostImpl::CreateUIResource().

Should be fixed here: http://crbug.com/643721