CefRenderHandler::OnPaint called with stale pixels

[callum]

Latest Win64 CEF and testing ability of embedded browser in my OSR app to change size - playing a YouTube video as a test.

As I vigorously resize the embedded browser, it normally continues to play correctly but sometimes the updated pixels from the still playing video fail to arrive in my CefRenderHandler::OnPaint override. I capture a hash of the pixels each time and when visually, the video appears to freeze (audio still playing) the hash stays constant.

Resizing again usually puts things back to normal and I see a different hash each frame as expected.

The unpredictable nature suggests a race condition somewhere.

I resize the embedded browser by making the new size available to CefRenderHandler::GetViewRect..) then calling CefBrowser:: GetHost()::WasResized();

Maybe another way to phrase this is : What would cause CefRenderHandler::OnPaint(...) to be called but with bad data ?

[amaitland]

The changes in https://bitbucket.org/chromiumembedded/ ... ce-for-osr were merged late last year.

Rather than saying your using the latest version, please specify the exact version you are using.

[callum]

Version 3.3239.1723.g071d1c1 Windows 64.

[amaitland]

Thanks.

Do you have a sample YouTube video that highlights the problem? I'm guessing that with some videos the issue will be more pronounced than others or will any old video do?

[callum]

The one I have been using to test is https://www.youtube.com/watch?v=h4rZE_J1beA

I feel like it must be something in my code - next step is to write a standalone test that illustrates it happening.

[amaitland]

Looks like a bug, reproduces with `cefclient --multi-threaded-message-loop --off-screen-rendering-enabled`

[callum]

Well that's a surprise - thank you for taking the time to look.

I didn't even think to bother trying cefclient - I'm working in macOS right now but once I'm done, I'll switch my laptop over to Windows and try it.

Actually, I'll try it in macOS too - see if it's OS specific.

Thank you.

[callum]

Wasn't able to repo on macOS (10.13) with same build of cefclient.

Repro's readily on Windows cefclient though as you said (I'm not using multi-threaded message loop).

Also repros on other YouTube videos like this one: https://www.youtube.com/watch?v=r2LpOUwca94 - so it's not something special about the first URL.

Should I file a bug? Happy to if you point me at the right place.

Thank you again - I've been tearing my hair out for the last couple of days with this.

[callum]

https://bitbucket.org/chromiumembedded/ ... -receiving

[salvadordf]

If you are using an external message pump you may need to create a message loop that handles all cases.

These are the comments in /tests/shared/browser/main_message_loop_external_pump.h

Code: Select all

// This MessageLoop implementation simulates the embedding of CEF into an
// existing host application that runs its own message loop. The scheduling
// implementation provided by this class is very simplistic and does not handle
// all cases (for example, nested message loops on Windows will not function
// correctly). See comments in Chromium's platform-specific
// base/message_loop/message_pump_* source files for additional guidance when
// implementing CefBrowserProcessHandler::OnScheduleMessagePumpWork() in your
// application. Run cefclient or ceftests with the
// "--external-message-pump" command-line flag to test this mode.


I originally translated most of that code to Delphi (Windows) to include it in CEF4Delphi but there were some reentrancy problems moving and resizing one of the demos so I had to make some modifications to that code.