[PATCH 4/5] winevulkan: Implement the concept of faking extension support

[Liam Middlebrook]

> 25 +# List of device extensions to ensure that we support regardless of whether 
>  26 +# the actual driver supports them or not.

I think for an extension to be "faked" it has to be fully unsupported by 
the underlying VK implementation. For simplicity I think it makes most 
sense if this is only used in-practice for platform==win32 extensions.

Theoretically winevulkan could fake a "normal" extension that's just not 
available from the current driver, but that seems like a weird path to 
go down, and I think it would be best to avoid.

Could the comment be updated to reflect the intended scope of faked 
extensions?

It won't matter for this initial implementation/usage, but it's 
interesting to consider if we want conditional support for faked 
extensions in the future.
> 27 +FAKED_EXTENSIONS = [
>  28 +    # Needed to not have an array size of 0.
>  29 +    {"name": "VK_DUMMY_extension", "version": 4}
>  30 +]

I think the generator code should just handle an empty list, rather than 
have a stub put in-place.

It would also be nice to have support for FAKED_DEVICE_EXTENSIONS and 
FAKED_INSTANCE_EXTENSIONS. For now I think it's fine to just implement 
this for device extensions, but I think the naming should be reflective 
of the fact this is only for device extensions.
> 100 +static BOOL wine_vk_device_extension_faked(const char *name)
> 101 +{
> 102 +    unsigned int i;
> 103 +    for (i = 0; i < wine_vk_device_extension_faked_count(); i++)
> 104 +    {
> 105 +        if (strcmp(wine_vk_device_extension_faked_idx(i)->extensionName, name) == 0)
> 106 +            return TRUE;
> 107 +    }
> 108 +    return FALSE;
> 109 +}
> 110 +

I don't have a strong reason as to why, but the _idx() and _count() 
functions feel kind of weird to me. What about having this 
wine_vk_device_extension_faked() as a generated function? It could live 
around the same area of make_vulkan as 
wine_vk_device_extension_supported() and 
wine_vk_instance_extension_supported() are defined. Although then we're 
left with _idx() used when populating the driver-advertised list of 
extensions. I don't really have any better suggestions for what to do 
here, so I'm not too concerned about this part as-is.


Thanks,

Liam Middlebrook

On 7/16/20 12:42 AM, Joshua Ashton wrote:
>