This is a recent regression:
- the browser hangs almost immediately on launch in release builds
- CHECK hit in RootCompositorFrameSinkImpl::Resize due to a zero size
- CHECK hit in ScopedGpuMemoryBuffer for caps.texture_storage_image
I found the problem here -- my patch https://chromium-review.googlesource.com/c/chromium/src/+/1076531 is completely wrong.
I mis-read "gpu_memory_manager" as "gpu_memory_buffer_manager". That this happened to have the desired effect in tests was near-magic.
But now I have to somehow deal with the fact if you support GPU raster and export texture_storage_image, a bunch of tests will try to use it. Tests that use cc::TestInProcessContextProvider, which CHECK when you try to CreateAnonymousImage.
Somehow cc::TestInProcessContextProvider needs to reach out into either FeatureInfo or GLES2CmdDecoder and tell it "you can't actually use texture_storage_image". I don't see any obvious path between the two of them.
Comment 1 by kylec...@chromium.org
, Jun 5 2018