Re: [PATCH 2/2] d3d11/tests: Add additional test for occlusion query.

[Henri Verbeet]

On 20 May 2017 at 14:12, Józef Kucia  wrote:
> - for (i = 0; i < 500; ++i)
> - {
> -
>    if ((hr = ID3D11DeviceContext_GetData(context, query, NULL, 0, 0)) != S_FALSE)
> -
>        break;
> -
>    Sleep(10);
> - }
> - ok(hr == S_OK, "Got unexpected hr %#x.\n", hr);
> - hr = ID3D11DeviceContext_GetData(context, query, &data, sizeof(data), 0);
> - ok(hr == S_OK, "Got unexpected hr %#x.\n", hr);
> - todo_wine ok(data.dword[0] == 0x1000 && !data.dword[1],
> -
>        "Got unexpected query result 0x%08x%08x.\n", data.dword[1], data.dword[0]);

This fails about half of the time with csmt enabled:

d3d11.c:4872: Test succeeded inside todo block: Got unexpected
query result 0x0000000000001000.

[PATCH 2/2] d3d11/tests: Add additional test for occlusion query.

[Józef Kucia]

On Mon, May 22, 2017 at 11:47 AM, Henri Verbeet  wrote:
> On 20 May 2017 at 14:12, Józef Kucia  wrote:
> > - for (i = 0; i < 500; ++i)
> > - {
> > -
> >    if ((hr = ID3D11DeviceContext_GetData(context, query, NULL, 0, 0)) != S_FALSE)
> > -
> >        break;
> > -
> >    Sleep(10);
> > - }
> > - ok(hr == S_OK, "Got unexpected hr %#x.\n", hr);
> > - hr = ID3D11DeviceContext_GetData(context, query, &data, sizeof(data), 0);
> > - ok(hr == S_OK, "Got unexpected hr %#x.\n", hr);
> > - todo_wine ok(data.dword[0] == 0x1000 && !data.dword[1],
> > -
> >        "Got unexpected query result 0x%08x%08x.\n", data.dword[1], data.dword[0]);
>
> This fails about half of the time with csmt enabled:
> d3d11.c:4872: Test succeeded inside todo block: Got unexpected
>
> query result 0x0000000000001000.

It's a race condition. A query may result may be retrieved in
wined3d_cs_run() before the swapchain is destroyed. Can we get the fix
committed without a test case? We've got "wrong thread" FIXME()
messages for this issue anyway.

[PATCH 2/2] d3d11/tests: Add additional test for occlusion query.

[Henri Verbeet]

On 22 May 2017 at 18:13, Józef Kucia  wrote:
> It's a race condition. A query may result may be retrieved in
> wined3d_cs_run() before the swapchain is destroyed. Can we get the fix
> committed without a test case? We've got "wrong thread" FIXME()
> messages for this issue anyway.

I think it's useful to have the test to show that it doesn't crash,
but perhaps checking the resulting data could be inside an "if (0)"
with an appropriate comment.