[Mesa-dev] [PATCH 2/2] st/mesa: simplify returning GL_VENDOR

[Ilia Mirkin]

On Wed, Jun 7, 2017 at 4:34 PM, Marek Olšák <maraeo at gmail.com> wrote:
> On Wed, Jun 7, 2017 at 10:28 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>> On Wed, Jun 7, 2017 at 4:22 PM, Marek Olšák <maraeo at gmail.com> wrote:
>>> On Wed, Jun 7, 2017 at 10:12 PM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>>> This creates string lifetime issues... should probably document that
>>>> the pointer returned by get_vendor and get_name are supposed to last
>>>> for the lifetime of the screen.
>>>
>>> Was the lifetime of the string not always guaranteed?
>>>
>>> Most drivers use a statically-declared string or a literal string. I
>>> think the lifetime has always been at least the same as the lifetime
>>> of pipe_screen.
>>
>> Previously it was stored in st_context, so the lifetime only had to be
>> for "a little while", until the next line of code saved it off. The
>> nouveau implementation is, for better or worse, currently:
>>
>> static const char *
>> nouveau_screen_get_name(struct pipe_screen *pscreen)
>> {
>>    struct nouveau_device *dev = nouveau_screen(pscreen)->device;
>>    static char buffer[128];
>>
>>    util_snprintf(buffer, sizeof(buffer), "NV%02X", dev->chipset);
>>    return buffer;
>> }
>>
>> So while the pointer will remain valid, it will potentially change for
>> different screens that are created within the same process space.
>> (Which could happen with EGL, or maybe other odd scenarios.)
>
> I think that should be fixed in drivers. You can, for example, put the
> string into your screen structure.

Fully agreed. And as you probably realize, it's a rare scenario. Just
looking to get some documentation saying that the pointers should
remain valid for the lifetime of the screen.

Cheers,

  -ilia