[FFmpeg-devel] [PATCH] all: use M_SQRT1_2, M_SQRT2, M_PI

[Paul B Mahol]

On 11/19/15, Ganesh Ajjanagadde <gajjanag at mit.edu> wrote:
> On Thu, Nov 19, 2015 at 12:35 PM, Nicolas George <george at nsup.org> wrote:
>> Le nonidi 29 brumaire, an CCXXIV, Derek Buitenhuis a ecrit :
>>> Paul uses FreeBSD, if I recall. This is probably with glibc.
>>
>> Still, M_PI is meant to have the full precision of the underlying type,
>> and
>> I very much doubt that kind of bug to be present in any common
>> implementation. And let us be realistic, the number of digits in
>> af_dynaudnorm.c is nothing but ludicrous.
>
> Exactly. Can't pull the source for bsd easily, but it likely is
> identical to openlibm since openlibm imported from freebsd:
> https://github.com/JuliaLang/openlibm/blob/master/include/openlibm_math.h
> which has an identical definition to gnu libm for M_PI.
>
> If such a silly, easily fixed bug exists with some libm, it can be
> fixed in a few minutes on their end, and I personally would do my best
> to publicize it to highlight the fundamental brokenness of said libm
> for the benefit of their users.
>

OK, you can change it.