On 20 March 2014 12:28, Stefan Dösinger stefan@codeweavers.com wrote:
- if (!(caps.ddraw_caps.caps & WINEDDCAPS_3D))
- {
WARN("Created a wined3d object without 3D support.\n");wined3d_decref(d3d8->wined3d);goto done;- }
I don't think that really matches Windows. Going by the testbot results, I think creating a D3D object should succeed, and creating a (hardware) device should fail.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Am 2014-03-20 12:46, schrieb Henri Verbeet:
I don't think that really matches Windows. Going by the testbot results, I think creating a D3D object should succeed, and creating a (hardware) device should fail.
Do we care enough about this? The behavior of my patch matches the behavior if OpenGL is not available. To change this we'd have to put this check in every d3d object method (or rather its wined3d counterpart) that accepts a device type. A quick search through our bug database didn't bring up any application that fails with the current handling and would otherwise do something reasonable if we returned an error during device creation.
On 20 March 2014 13:28, Stefan Dösinger stefandoesinger@gmail.com wrote:
Am 2014-03-20 12:46, schrieb Henri Verbeet:
I don't think that really matches Windows. Going by the testbot results, I think creating a D3D object should succeed, and creating a (hardware) device should fail.
Do we care enough about this? The behavior of my patch matches the behavior if OpenGL is not available.
Perhaps not quite enough to fix all the device type checks right now, but I certainly don't think we should add extra code to make the behaviour worse either.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Am 2014-03-20 14:06, schrieb Henri Verbeet:
Perhaps not quite enough to fix all the device type checks right now, but I certainly don't think we should add extra code to make the behaviour worse either.
So you'd prefer to just ignore the possibly set WINED3D_NO3D flag in d3d8/9/dxgi? The result of that is (at least in the tests) a crash later. I'm fine with that, just mentioning it.
On 20 March 2014 14:44, Stefan Dösinger stefandoesinger@gmail.com wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Am 2014-03-20 14:06, schrieb Henri Verbeet:
Perhaps not quite enough to fix all the device type checks right now, but I certainly don't think we should add extra code to make the behaviour worse either.
So you'd prefer to just ignore the possibly set WINED3D_NO3D flag in d3d8/9/dxgi? The result of that is (at least in the tests) a crash later. I'm fine with that, just mentioning it.
Which test is that?
Am 20.03.2014 16:45 schrieb "Henri Verbeet" hverbeet@gmail.com:
On 20 March 2014 14:44, Stefan Dösinger stefandoesinger@gmail.com wrote:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Am 2014-03-20 14:06, schrieb Henri Verbeet:
Perhaps not quite enough to fix all the device type checks right now, but I certainly don't think we should add extra code to make the behaviour worse either.
So you'd prefer to just ignore the possibly set WINED3D_NO3D flag in d3d8/9/dxgi? The result of that is (at least in the tests) a crash later. I'm fine with that, just mentioning it.
Which test is that?
I think it was some operator error by me. I could not reproduce it when I tested before sending v3. It was the d3d8 visual test. It crashed during device creation inside wined3d. I suspect I had a mismatched d3d8 and wined3d build.
-
Henri Verbeet -
Stefan Dösinger