On Fri, Nov 29, 2019 at 2:07 PM Matteo Bruni matteo.mystral@gmail.com wrote:
On Fri, Nov 29, 2019, 13:05 Sven Baars sven.wine@gmail.com wrote:
On 29-11-2019 11:57, Matteo Bruni wrote:
elseok(!desc.Stream,·"(%u):·got·unexpected·Stream·%u.\n",·i,·desc.Stream);Hi Matteo,
This is not true according to the tests:
Oh indeed, there is something weird going on (e.g. https://test.winehq.org/data/ec9eddf86b144cc0fcf975a9cc864cf08ba84eaf/win10_...). That suggests that somehow the "missing" field is changed by the implementation? AFAICS the local variable is supposed to be completely initialized to 0 by the test. Also of note, it looks like only the 64-bit tests are affected.
Okay, as it turns out on 64-bit sizeof(D3D10_SIGNATURE_PARAMETER_DESC) == sizeof(D3D11_SIGNATURE_PARAMETER_DESC) == 32 i.e. the structs have actually the same size, the d3d10 version has just more padding at the end. I thought that the d3d10 version would always end before the offset of the Stream field in the d3d11 structure so testing that the implementation (especially our implementation) doesn't write out of the struct would be a useful test. But that's clearly not the case and checking the contents of the padding isn't a particularly sensible thing to do. I'll drop those checks from the d3d10 tests; Sven, thanks again for the patch and keep looking after the fix in case I still miss something :D
P.S.: It's mildly annoying that pahole doesn't work with PE files, it would have been useful in this case.