[Webkit-unassigned] [Bug 224566] REGRESSION: [ wk2 ] http/tests/security/webgl-remote-read-remote-image-allowed-with-credentials.html is flakey crashing

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=224566

--- Comment #12 from Chris Dumez <cdumez at apple.com> ---
(In reply to Chris Dumez from comment #11)
> Could be just flakiness but it looks like reverting both r275921 and r275922
> makes the flaky crash go away. Both patches may impact when we decide to
> launch to GPUProcess so the issue could be timing-related. I'll investigate
> more tomorrow.

I think this has to do with when we decide to initiate the connection to the GPUProcess.

r275921 is mostly an iOS patch but it does one thing on macOS:
- It delays the construction of the RemoteRenderingBackendProxy until we actually need it. It used to get constructed as soon as we constructed the WebPage, now it only gets constructed when the GPUProcess launches or when we start doing canvas rendering.

r275922 is related too. Before r275922, constructing the RemoteRenderingBackendProxy would initiate the GPUProcess connection right away (in the constructor). After r275922, eagerly constructing a RemoteRenderingBackendProxy does not actually initiate the GPUProcess connection. We only launch the GPUProcess when we actually do canvas rendering via the RemoteRenderingBackendProxy.

I think this is why reverting r275921 without reverting r275922 does not fix the crash.

That said, now I have to figure out why launching the GPUProcess when we start canvas drawing instead of on WebPage creation actually causes this flaky crash. Even weirder, it is not a real crash but an XPC error (XPC_ERROR_CONNECTION_INVALID).

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20210415/10751644/attachment-0001.htm>