The symptoms are of a ui::Layer claiming SetFillsBoundsOpaquely(true) but actually being non-opaque.

Run binary bisect. Regression range: 463863 - 463933

https://chromium.googlesource.com/chromium/src/+log/23724703cfbba977e4628436698c59bdba5c2b05..6ede719ec9e4a968c5eedafda3be94c210568d81

bisected to d85baf0b71c69bbd181aaefc8a803611e03c8eed (https://codereview.chromium.org/2777193002)

Reverting the CL on ToT fixes the issue.

oops, sorry for typo..

We are just few day's away from M59 stable launch, 	capn@ can we get the fix in ASAP please.

I see that capn@ is off is there anyone who can take over the bug please.

Hi all, I just downloaded Chrome dev (current version is 60.0.3112.7) on Linux.
- Opened it -> Settings -> on startup -> Continue where you left off
- Opened tabs
- Closed chrome
- Reopened it

No issue. I can't repro this. I tried with and without --disable-gpu to force SwiftShader usage, still works. Can anyone double check if the issue is still there and, assuming it is, are there any other steps / settings required to reproduce this?
The tabs I opened were Google search result pages. Does it require a certain type of page?
Thanks.

FWIW I was doing this in a Chrome Remote Desktop session, with a developer/Debug build, so it may be Debug and/or software-rendering specific.

If this is caused by the swiftshader change, then it should definitely be software-rendering specific, but using --disable-gpu forces software rendering, so I should be able to repro it using that flag.

Still can't reproduce it. Is there anyone who can still reproduce this issue? If so, can you provide detailed steps? I don't want to close an existing issue, but I will close it if no one answers. Thanks.

Is it possible that this is caused by an asynchronous fallback, where the first renderer gets a channel that it expects works, and then starts using the GPU process / GL to render content (probably ends up getting a swiftshader context), but the UI switches to software rendering, and can't deal with "hardware" (meaning: GL) resources?

--disable-gpu would avoid the asynchronous fallback and possibly why you can't repro.

I can no longer repro this at ToT (61.0.3117.0), FWIW.

I'm not certain how an asynchronous fallback would happen. I think the browser process is the one to take the decision to use SwiftShader, currently, before other processes are started. Can a renderer process request a GPU channel from GpuProcessHost even before GpuDataManagerImplPrivate is initialized? If that's possible, I wasn't aware of it.

Since nobody can reproduce this issue, closing as "won't fix". Please feel free to re-open and re-assign to me if you find a way to reproduce this issue, but make sure to include detailed steps if you do. Thanks.