Adding Release blocker for this issue.Please undo if not the case.

Thank You!

Sending over to mcnee@ who did the pinch-zoom work for PDFs. I believe this code is all new, we didn't handle pinch-zoom on a trackpad previously, so I don't think this is a release block.

The behaviour in the expected video shows the PDF viewer handling the pinch correctly (which originally landed on Nov 13). So this doesn't appear to be related to the introduction of the handling in the viewer itself, but a regression that happened since the introduction.

In the actual video, the PDF viewer is still handling the pinch, but the scale changes only for the upper left quarter of the PDF content, while the rest gets frozen at a given scale and then snaps back to the correct scale.

The suspected CL mentions the possibility of "pepper getting out of sync with the compositor", so perhaps there's an issue with pepper's SetLayerTransform which we use in the PDF viewer to scale the PDF's bitmap.

The CL moves the texture upload from TextureLayerImpl to the pepper2d host. Along the way it also ensured that the dirty flag is changed when moving between gpu and software compositing modes.

The sync there it refers to is the compositing mode: gpu vs software compositing. This changes when the context is lost, not on pinch or something.

The bug looks to me like its missing damage, like the report to the compositor of what has changed is being scaled by the pinch or something. But I can't see where my CL would change that. The old upload path didn't add any damage: https://chromium-review.googlesource.com/c/chromium/src/+/692999/24/cc/layers/texture_layer_impl.cc

> Note : Issue is not reproducible on Windows(7,8,10) & Linux(14.04 LTS)

Maybe mac is using partial swap but windows/linux aren't, could explain if it is damage related.

This is where damage gets reported fwiw: https://cs.chromium.org/chromium/src/content/renderer/pepper/pepper_plugin_instance_impl.cc?rcl=2b344530d6d9c7898e2cb626880a8e841e5922ce&l=751

I just tried this on ToT on a Mac and wasn't able to reproduce.

The video looks like damage is missing DPI scaling.

Thanks. That was it. I tried with HiDPI and was able to reproduce.

It looks like the |invalidated_rect| set by SetLayerTransform doesn't account for the scaling.
https://cs.chromium.org/chromium/src/content/renderer/pepper/pepper_graphics_2d_host.cc?rcl=874c50528a8520f3c4990d3a464bc172588ada9d&l=841

Mmh, I was wondering why Dana's CL changes that, I think what happens is that it kicks us off the CA compositing, and so we likely behave differently wrt damage: before we would use ResourceProvider to create the resource, which would create a GMB (that we can push to CA), whereas now we create a plain-old-texture, which we can't push to CA, so we fall back to GLRenderer.

It sounds like we want to fix both things.

This will be hard to fix without a mac but I'll give that a try. I will split that off into another bug and leave this one about damage since it has all that context.

https://bugs.chromium.org/p/chromium/issues/detail?id=786140 for the GpuMemoryBuffer thing.

Once 786140 is fixed, the bug will become hidden. In order to fix the damage issue, you can use --disable-mac-overlays to disable the fix I'm landing for that bug.

Oops, I double checked with ccameron and the flags would be these actually:

--disable-zero-copy --disable-native-gpu-memory-buffers 

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/a140876f4c9377e89a8e23b57face1a141325085

commit a140876f4c9377e89a8e23b57face1a141325085
Author: Kevin McNee <mcnee@chromium.org>
Date: Tue Nov 21 15:10:10 2017

Pepper: Account for DPI scaling for damage of SetLayerTransform.

Bug:  785801 
Change-Id: I0bcc31cc9f90205d1c5c0181f25394f48f845605
Reviewed-on: https://chromium-review.googlesource.com/780180
Reviewed-by: Antoine Labour <piman@chromium.org>
Commit-Queue: Kevin McNee <mcnee@chromium.org>
Cr-Commit-Position: refs/heads/master@{#518252}
[modify] https://crrev.com/a140876f4c9377e89a8e23b57face1a141325085/content/renderer/pepper/pepper_graphics_2d_host.cc