Do not set a WNDPROC on a window created by another thread.

Do not set a WNDPROC on a window created by another thread.

It turns out SetWindowLong succeeds for other threads in the same process but we do not want the WNDPROC to be called from threads other than the thread on which EGL is running on. This was one of the reasons --in-process-gpu was not working with ANGLE in Chrome.
Committed: http://code.google.com/p/angleproject/source/detail?r=677

[nduca, 2011-06-06 22:24:12]

LGTM.

As an aside, this whole subclassing business is starting to be more and more
fragile. E.g. anyone who tries to use ANGLE in a multithreaded renderer would
suddenly start seeing resizing artefacts. Does this argue in favor of checking
the window size at the start of frames instead?

[apatrick1, 2011-06-06 22:25:56]

On 2011/06/06 22:24:12, nduca wrote:
> LGTM.
> 
> As an aside, this whole subclassing business is starting to be more and more
> fragile. E.g. anyone who tries to use ANGLE in a multithreaded renderer would
> suddenly start seeing resizing artefacts. Does this argue in favor of checking
> the window size at the start of frames instead?

Yeah I think that would be more robust.

[dgkoch, 2011-06-07 00:28:22]

LGTM.

(Note that ANGLE is not currently multithreaded safe anyhow).

[apatrick1, 2011-06-07 00:36:13]

On 2011/06/07 00:28:22, dgkoch wrote:
> (Note that ANGLE is not currently multithreaded safe anyhow).

Yeah, that's why this change was necessary. Chrome issues GL calls from the a
spceial thread in the browser process (not the UI thread that created the
window) when run with --in-process-gpu.