[Mesa-dev] Can we use anonymous unions?

Marek Olšák maraeo at gmail.com
Wed Aug 19 03:46:37 PDT 2015


On Wed, Aug 19, 2015 at 2:47 AM, Jason Ekstrand <jason at jlekstrand.net> wrote:
> On Tue, Aug 18, 2015 at 5:36 PM, Ian Romanick <idr at freedesktop.org> wrote:
>> On 08/18/2015 10:08 AM, Connor Abbott wrote:
>>> On Tue, Aug 18, 2015 at 9:56 AM, Kenneth Graunke <kenneth at whitecape.org> wrote:
>>>> Hey,
>>>>
>>>> I was thinking about using an anonymous union.  Specifically, something
>>>> like:
>>>>
>>>> struct shader {
>>>>    ...
>>>>    union {
>>>>       struct geometry_shader_info gs;
>>>>       struct tess_eval_shader_info tes;
>>>>       ...
>>>>    };
>>>> };
>>>>
>>>> Are those acceptable in Mesa?  I don't think we've traditionally used
>>>> them, but I'm not sure why.  Apparently they're part of C11, though not
>>>> part of C99.  However, GCC allows them, presumably Clang, and it looks
>>>> like MSVC 2005 supports them in C:
>>>>
>>>> https://msdn.microsoft.com/en-us/library/y9zewe0d%28v=vs.80%29.aspx
>>>>
>>>> We might be able to use these to avoid some of our fun
>>>> thing->base.Base.base.program.Base.base shenanigans...
>>>>
>>>> --Ken
>>>>
>>>> _______________________________________________
>>>> mesa-dev mailing list
>>>> mesa-dev at lists.freedesktop.org
>>>> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
>>>>
>>>
>>> Well, we already use anonymous unions in various places in NIR (for
>>> example, the definition of nir_dest and nir_src), so they should be ok
>>> for core Mesa. IIRC there was a problem with combining anonymous
>>> unions and C99-style initializers, but it doesn't sound like you'll be
>>> too worried about that.
>>
>> Hm... is that why I get thousands of uninitialized field warnings?  Or
>> are NIR_SRC_INIT and NIR_DEST_INIT just missing some things?
>
> That depends on how you look at it.  You could say that they are
> missing things.  However, the correct initialization for either a
> nir_src or a nir_dest is effectively a memset to 0.  Since C (and C++)
> guarantee that missing fields in a structure initializer (or compound
> literal) are initialized to 0 as long as one of them is, we get what
> we want without having to have the entire struct layout full of zeros.
> Also, since there is a union or two in there, it makes everything
> easier.

Yes, using {} (C99 only) or {0} initializers is a common way to
initialize structures to zero. If it causes any warnings, they should
be suppressed.

Marek


More information about the mesa-dev mailing list