Trying to fix bug crash when eglCreateWindowSurface is called and CreateDevic...

Trying to fix bug crash when eglCreateWindowSurface (or any other code that causes createDevice to be invoked) is called and CreateDevice reports DEVICELOST or NOTAVAILABLE.

This was a Windows XP service pack 3 box.

0x10002545 	[libegl.dll 	- display.cpp:340] 	egl::Display::createDevice()
0x10002912 	[libegl.dll 	- display.cpp:468] 	egl::Display::getDevice()
0x01d6a80f 	[chrome.dll 	- gpu_info_collector_win.cc:29] 	gpu_info_collector::CollectGraphicsInfo(GPUInfo *)
0x01d68eca 	[chrome.dll 	- gpu_thread.cc:104] 	GpuThread::OnEstablishChannel(int)

Previously createDevice only checked for out of memory and the ASSERT(SUCCEEDED(result)) has no effect in release builds.

I simulated the failure in a debugger and discovered a second place where a null dereference would occur in this case.

I don't think this crash is technically correct but it prevents crashes.

[apatrick1, 2010-09-16 22:12:37]

Another one for you Daniel.

Thanks,

Al

[dgkoch, 2010-09-17 20:09:34]

looks good other than the couple comments

http://codereview.appspot.com/2217043/diff/1/src/libEGL/Display.cpp
File src/libEGL/Display.cpp (right):

http://codereview.appspot.com/2217043/diff/1/src/libEGL/Display.cpp#newcode325
src/libEGL/Display.cpp:325: }
you might also want to put an explicit check for D3DERR_DEVICELOST up here as
well, since there is not point in trying to create the other type of device it
the device is really lost.

http://codereview.appspot.com/2217043/diff/1/src/libEGL/Display.cpp#newcode331
src/libEGL/Display.cpp:331: if (FAILED(result))
this seems reasonable.   Perhaps we should also put an assert in here based on
the old conditions (to help tracking issues down in debug builds?).