[PATCH 0/1] MR1965: d3d12/tests: Fix the expected dxgi adapter refcount check.

[PATCH 1/1] d3d12/tests: Fix the expected dxgi adapter refcount check.

[Francois Gouget]

From: Francois Gouget

The refcount is 0 on Windows 10 1909+ so consider the 1709 and 1809
results to be buggy.

Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=54299
---
 dlls/d3d12/tests/d3d12.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/dlls/d3d12/tests/d3d12.c b/dlls/d3d12/tests/d3d12.c
index 0d3deafda45..e57c98e96e7 100644
--- a/dlls/d3d12/tests/d3d12.c
+++ b/dlls/d3d12/tests/d3d12.c
@@ -885,7 +885,8 @@ static void test_create_device(void)
     hr = D3D12CreateDevice((IUnknown *)adapter, D3D_FEATURE_LEVEL_11_0, &IID_ID3D12Device, (void **)&device);
     ok(hr == S_OK, "Got unexpected hr %#lx.\n", hr);
     refcount = IDXGIAdapter_Release(adapter);
-    ok(refcount >= 1, "Got unexpected refcount %lu.\n", refcount);
+    ok(refcount == 0 || broken(refcount >= 1 /* win10 1709 & 1809 */),
+       "Got unexpected refcount %lu.\n", refcount);
     adapter = NULL;

luid = ID3D12Device_GetAdapterLuid(device);

[Francois Gouget (＠fgouget)]

The mf:mf GitLab CI success-inside-todo is unrelated to this MR. So far I had only seen it on my desktop (see bug 54306) so I'm kind of happy to see I'm not the only one with this failure (although in a way I guess I'm still the only one with this failure :-( ).

[PATCH 0/1] MR1965: d3d12/tests: Fix the expected dxgi adapter refcount check. - approved

[Zebediah Figura (＠zfigura)]

This merge request was approved by Zebediah Figura.

[Alexandre Julliard (＠julliard)]

This now fails on Wine:

```
tools/runtest -q -P wine -T . -M d3d12.dll -p dlls/d3d12/tests/i386-windows/d3d12_test.exe d3d12 && touch dlls/d3d12/tests/i386-windows/d3d12.ok
d3d12.c:888: Test failed: Got unexpected refcount 1.
make: *** [Makefile:16789: dlls/d3d12/tests/i386-windows/d3d12.ok] Error 1
```

[Francois Gouget (＠fgouget)]

On Tue Jan 17 19:35:38 2023 +0000, Alexandre Julliard wrote:
> This now fails on Wine:
> tools/runtest -q -P wine -T . -M d3d12.dll -p
> dlls/d3d12/tests/i386-windows/d3d12_test.exe d3d12 && touch dlls/d3d12/tests/i386-windows/d3d12.ok
> d3d12.c:888: Test failed: Got unexpected refcount 1.
> make: *** [Makefile:16789: dlls/d3d12/tests/i386-windows/d3d12.ok] Error 1

I see.
* test.winehq.org shows that there is no failure in Wine which means it consistently returns refcount >= 1.
* But there was no failure for this test in the 10 TestBot runs. However looking closer it seems that's because the test is systematically skipped there due to 'Failed to create device'.
* There was no failure on the GitLab CI either and no skipped traces. However I believe the 'skipped' traces are silenced there so the tests are probably still not run.

So then adding a todo_wine would probably fix the test on both Windows and Wine.