[Mesa-dev] [PATCH v2 1/2] virgl: Fix a strict-aliasing violation in the encoder

[Emil Velikov]

On 8 December 2016 at 00:03, Edward O'Callaghan
<funfunctor at folklore1984.net> wrote:
> On 12/08/2016 02:53 AM, Emil Velikov wrote:
>> On 6 December 2016 at 22:34, Edward O'Callaghan
>> <funfunctor at folklore1984.net> wrote:
>>> On 12/07/2016 12:36 AM, Emil Velikov wrote:
>>>> On 6 December 2016 at 11:30, Edward O'Callaghan
>>>> <funfunctor at folklore1984.net> wrote:
>>>>> As per the C spec, it is illegal to alias pointers to different
>>>>> types. This results in undefined behaviour after optimization
>>>>> passes, resulting in very subtle bugs that happen only on a
>>>>> full moon..
>>>>>
>>>>> Use a memcpy() as a well defined coercion between the double
>>>>> to uint64_t interpretations of the memory.
>>>>>
>>>>> V.2: Use static_assert() instead of assert().
>>>>>
>>>> The lowercase static_assert is a c11 feature. You want to use the all
>>>> caps version.
>>>> See commit 23d1799f7dd5f8d1e8aa9f4efa6b1a4ed45faaa0
>>>
>>> Hi Emil,
>>>
>>> Why? I from that commit it seems only Android would have trouble however
>>> virgl nor svga would want to be enabled on that platform?
>>>
>>> Or is there something else I probably missed?
>>>
>> If there's a valid reason to require X or Y one must clearly point it
>> out, rather than jumping on bandwagons because it's cool/one can/etc.
>> Forcing X and Y [without justification] is disrespectful towards
>> everyone, even fellow developers. It also leads distro maintainers to
>> think that upstream does not give a sh*t about issues they are faced
>> with - old compilers, other.
>>
>> There are [hundreds of] thousands people using mesa many of which with
>> their own quirky requirements. Please consider that for future.
>> Esp. when it's a trivial change as this ;-)
>
> Emil, I understand all that and I had already sent a v3 with the
> required amendment before I replied here. As I am not trying to
> 'force'/'disrespect' anyone or anything, I asked purely to gain a better
> understanding of the reasoning behind requested change.
>
> I don't take things on gospel, no matter how high up the developer is or
> however trivial the change may be, as it is my change it is then my
> responsibility to fully understand any variation to the change I am
> proposing.
>
Pardon for going overboard there, Edward.
We've had enough cases where we (myself including) went ahead for the
new cool stuff and things ended up broken.

On the question:
Barring the point that we build virgl on Android, Vinson (and likely
others) use older versions of GCC and things tend to break every so
often.

It is a bit annoying, yet it's something which can be fixed easily.
Emil