[FFmpeg-cvslog] r19375 - in trunk: Changelog libavcodec/Makefile libavcodec/aac.h libavcodec/aaccoder.c libavcodec/aacenc.c libavcodec/aacenc.h libavcodec/aacpsy.c libavcodec/allcodecs.c libavcodec/psymodel.c liba...

[Alex Converse]

On Thu, Jul 9, 2009 at 9:21 AM, Reimar
D?ffinger<Reimar.Doeffinger at gmx.de> wrote:
> On Thu, Jul 09, 2009 at 01:59:13PM +0100, M?ns Rullg?rd wrote:
>> Reimar D?ffinger <Reimar.Doeffinger at gmx.de> writes:
>> > You broke a good deal of FATE configurations. In particular, complex
>> > types do not seem to be generally working.
>> >
>> > From the gcc 4.3.3/Solaris 10 build:
>> > /home/fate/source/libavcodec/iirfilter.c: In function 'ff_iir_filter_init_coeffs':
>> > /home/fate/source/libavcodec/iirfilter.c:83: error: '_Imaginary_I' undeclared (first use in this function)
>> > /home/fate/source/libavcodec/iirfilter.c:83: error: (Each undeclared identifier is reported only once
>> > /home/fate/source/libavcodec/iirfilter.c:83: error: for each function it appears in.)
>> > /home/fate/source/libavcodec/iirfilter.c:56: warning: unused variable 'size'
>>
>> Quoth the spec:
>>
>> ? The macro I expands to either _Imaginary_I or _Complex_I. If
>> ? _Imaginary_I is not defined, I shall expand to _Complex_I.
>>
>> It follows that the system headers are buggy.
>
> Oh, I don't doubt some or even most of it is the system's fault.
>
>> > From the NetBSD, DragonFly BSD and AVR Linux builds:
>> > aaccoder.c:(.text+0x2ba2): undefined reference to `fminf'
>> > aaccoder.c:(.text+0x2bbc): undefined reference to `fmaxf'
>>
>> Can't FFMIN and FFMAX be used? ?They only differ in the case of NaN
>> arguments.
>>
>> > From the FreeBSD build:
>> > /usr/home/mik/src/fate/build/libavcodec/libavcodec.a(aaccoder.o)(.text+0x998):
>> > In function `search_for_quantizers_fast':
>> > : undefined reference to `log2'
>>
>> Yet another standard C function they've missed/ignored...
>
> Nevertheless IMO FFmpeg should build by default on more than OSX, Linux and OpenBSD -
> looking at FATE that seems to be what it is currently reduced to.
> Note that these systems are actually doing quite good in comparison, e.g. MIPSEL OpenWRT
> lacks far more math functions.
> I don't know how high the costs of avoiding these are, in the worst case the aac encoder
> should be disabled for those systems where it doesn't work.

whatever happened to libbrokenos?

Anyway, is:

#ifndef log2
#define log2(x) (log(x) * (1/M_LN2))
#endif

acceptable?