[FFmpeg-devel] FATE on BSD

[Reinhard Tartler]

On Sun, Jan 23, 2011 at 15:35:51 (CET), M?ns Rullg?rd wrote:

> Michael Kostylev <michael.kostylev at gmail.com> writes:
>
>>>>>> Though I still cannot use --enable-shared, for some reason the
>>>>>> resulting binaries fall into an infinite loop. The same applies also
>>>>>> to DragonFly and NetBSD, e.g.:
>>>>>> % gdb ./ffmpeg_g
>>>>>> (gdb) r -h
>>>>>> ^C
>>>>>> Program received signal SIGINT, Interrupt.
>>>>>> 0xbb338e24 in ff_av_opt_set_defaults2 ()
>>>>>>    from /home/mik/src/fate/build/tmp/usr/local/lib/libavcodec.so.52
>>>>>> ...
>>>>>> 0xbb338e24 <ff_av_opt_set_defaults2+0>: jmp    0xbb338e24 <ff_av_opt_set_defaults2>
>>>>>
>>>>>Curious.  Relocation problems?
>>>>
>>>> Looks so. Just in case I'll check pkgsrc.
>>
>> Works ok if configured with --disable-symver.
>
> Sounds like a buggy link editor or dynamic loader.
>
>> Normally, configure enables symbol versioning:
>>
>> #define HAVE_SYMVER 1
>> #define HAVE_SYMVER_GNU_ASM 1
>> #define HAVE_SYMVER_ASM_LABEL 0
>
> Which systems are affected?  We can simply have configure disable it
> for those.  Since the syntax is accepted without errors, it's
> difficult to test for it.

I remember android was reported having problems here, which lead to the
introduction of the --disable-symver switch. Let's disable symbol
versioning on Android and BSDs for now.

Note that this causes upgrade problems for binaries that have been
compiled against earlier copies of ffmpeg, like a past release. AFAIUI,
this means that the only hope for these platforms is a major bump of
avcodec and avformat.

-- 
Gruesse/greetings,
Reinhard Tartler, KeyID 945348A4