Re: ddraw: Keep track of interface attached by AddAttachedSurface and detach correct interface when parent is released. (try 3)
Hello, Octavian.
* On Tue, 11 Oct 2011, Octavian Voicu wrote:
try 3:
- fix test failure by moving error check from *_DeleteAttachedSurface to ddraw_surface_delete_attached_surface, after another check (thanks Henri).
try 2:
- use IUnknown_Release(attached_iface) instead of manual dispatch, as suggested by Henri;
- change type of attached_iface from void* to IUnknown* to reflect fact that it's a pointer to an interface;
- add test to show that attaching a surface only attaches a specific interface and return error if DeleteAttachedSurface is called with another interface than the attached one (instead of detaching the surface anyway).
Fixes a bug that causes a stencil buffer leak in the game The Longest Journey. Problem can be seen in the console output log attached to bug #11819. This patch doesn't fix the original bug, but now game exits cleanly.
The root of the problem is that when releasing a surface, its attached surfaces are detached using IDirectDraw7_DeleteAttachedSurface, regardless of what interface was originally attached.
I am troubled by seeing (via test.winehq.org) your ddraw:refcount tests to account only results for "VMware SVGA 3D" adapter/driver (and only those with 3D acceleration disabled), I guess.
But if this driver is 3D-enabled or just present real physical adapter (with no 3d acceleration) then ddraw:refcount contains 6 static failures such as:
refcount.c:214: Test failed: AddAttachedSurface returned 8876000a refcount.c:235: Test failed: AddAttachedSurface returned 8876000a refcount.c:237: Test failed: Got refcount 1, expected 2 refcount.c:262: Test failed: AddAttachedSurface returned 8876000a refcount.c:275: Test failed: AddAttachedSurface returned 8876000a refcount.c:277: Test failed: Got refcount 1, expected 2 refcount: 112 tests executed (0 marked as todo, 6 failures), 0 skipped
This relation can be easily tracked by looking at d3d9:visual or (if they aren't present) at opengl32:opgenl tests.
Eg. failures occuring here:
* Intel(R) 82865G Graphics Controller http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/xp_af-x... http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/xp_af-x...
* NVIDIA GeForce FX 5200 http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/win7_s2... http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/win7_s2...
http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/xp_s2-s... http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/xp_s2-s...
* ATI Radeon HD 4800 Series (telling it has no IDirect3D[7] support for DDraw) http://test.winehq.org/data/a843acef5567bc30901612ce5013da543259b21b/win7_wi... http://test.winehq.org/data/a843acef5567bc30901612ce5013da543259b21b/win7_wi...
* VMware SVGA II http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/2000_fg... http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/2000_fg...
http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/2000_fg... http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/2000_fg...
* VMware SVGA 3D (telling it has not IDirect3D[7] support for DDraw) http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/win7_fg... http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/win7_fg...
All other cases succeed. AFAICS all of them represent the only driver (with 3D support turned off):
* VMware SVGA 3D http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/win7_wt...
visual.c:194: Driver string: "vm3dum.dll" visual.c:195: Description string: "VMware SVGA 3D (Microsoft Corporation - WDDM)" visual.c:197: Device name string: "\.\DISPLAY1" visual.c:199: Driver version 7.14.1.40 visual.c:12627: Tests skipped: Creating the device failed visual: 5 tests executed (0 marked as todo, 0 failures), 1 skipped.
with following results: http://test.winehq.org/data/d0c38ece9db657431559024abb34285cd8b717b6/win7_wt...
refcount: 121 tests executed (0 marked as todo, 0 failures), 0 skipped.
IMHO, it would be nice to conform to real world configuration. Wish you best luck with fixing your tests:)
S.
On 13 October 2011 10:12, Saulius Krasuckas saulius2@ar.fi.lt wrote:
refcount.c:214: Test failed: AddAttachedSurface returned 8876000a refcount.c:235: Test failed: AddAttachedSurface returned 8876000a refcount.c:237: Test failed: Got refcount 1, expected 2 refcount.c:262: Test failed: AddAttachedSurface returned 8876000a refcount.c:275: Test failed: AddAttachedSurface returned 8876000a refcount.c:277: Test failed: Got refcount 1, expected 2 refcount: 112 tests executed (0 marked as todo, 6 failures), 0 skipped
It probably doesn't make sense to attach a depth buffer to a P8 offscreenplain surface. I think it would make more sense to move these tests to the existing tests for AddAttachedSurface() in dsurface.c.
On Thu, Oct 13, 2011 at 11:12 AM, Saulius Krasuckas saulius2@ar.fi.lt wrote:
I am troubled by seeing (via test.winehq.org) your ddraw:refcount tests to account only results for "VMware SVGA 3D" adapter/driver (and only those with 3D acceleration disabled), I guess.
But if this driver is 3D-enabled or just present real physical adapter (with no 3d acceleration) then ddraw:refcount contains 6 static failuDDSCAPS_3DDEVICEres such as:
Hi Saulius,
Thanks for pointing this out. I only tested with testbot and I guess all of the machines use that adapter (at least the default ones do), so that's why I didn't catch the error.
The application that I was debugging used stencil buffers, so I sticked with those -- although I did have a feeling it's not a good idea to test using DDSCAPS_3DDEVICE.
I'm going to write a patch to fix the tests by switching to more general surface types and maybe also move the tests to dsurface.c.
@Henri: I was in doubt between refcount.c and dsurface.c when writing the tests, but refcount.c allowed me to write the tests for both IDirectDrawSurface and IDirectDrawSurface7 in the same function and so reuse the caps for them. Now that I QI the v1 interface, maybe I can still combine them somehow.
Octavian
-
Henri Verbeet -
Octavian Voicu -
Saulius Krasuckas