Bug#789250: bs1770gain: Fail to build on archs with strict alignment requirements

[Peter Belkner]

If you have any doubts whether the alignment of the pointer to the frame 
buffer is given you should note that the buffer is allocated using the 
respective FFmpeg API function "av_frame_get_buffer()" (cf. 
https://www.ffmpeg.org/doxygen/2.6/frame_8c.html).

In "ffsox" the following function in "libffsox-2/ffsox_frame.c" 
implements frame buffer allocation:

    int ffsox_frame_create_cc(frame_t *f, AVCodecContext *cc)
    {
       const AVCodec *codec=cc->codec;
       AVFrame *frame;
       int nb_samples;

       if (ffsox_frame_create(f)<0) {
         DMESSAGE("creating frame");
         goto frame;
       }

       if (NULL!=codec&&(codec->capabilities&CODEC_CAP_VARIABLE_FRAME_SIZE))
         nb_samples=10000;
       else
         nb_samples=cc->frame_size;

       frame=f->frame;
       frame->format=cc->sample_fmt;
       frame->channel_layout=cc->channel_layout;
       frame->channels=cc->channels;
       frame->sample_rate=cc->sample_rate;
       frame->nb_samples=nb_samples;

       if (0<nb_samples&&av_frame_get_buffer(frame,0)<0) {
         DMESSAGE("allocating frame buffer");
         goto buffer;
       }

       return 0;
    // cleanup:
    buffer:
       ffsox_frame_cleanup(f);
    frame:
       return -1;
    }

This code is reminiscent to the FFmpeg example 
"ffmpeg/doc/examples/transcode_aac.c". It's up to the FFmpeg API 
function "av_frame_get_buffer()" to set the "data" field of an "AVFrame" 
to the right alignment and it's not the aim of FFmpeg API client code.



On 24.06.2015 07:23, Peter Belkner wrote:
>> The first build of bs1770gain on the autobuilders showed an error on at
>> least the mipsel architecture, that look like this:
>>
>> ffsox_frame_convert.c: In function 'convert_s16i_s8i':
>> ffsox_frame_convert.c:143:6: error: cast increases required alignment 
>> of target type [-Werror=cast-align]
>>     rp=(R *)p->fr->frame->data[0]; \
>>        ^
>> ffsox_frame_convert.c:157:1: note: in expansion of macro 'CONVERT_II'
>>   CONVERT_II(s16,s8,int16_t,int8_t,CONVERT_INT_INT_II)
>>   ^
>> ffsox_frame_convert.c: In function 'convert_s32i_s8i':
>> ffsox_frame_convert.c:143:6: error: cast increases required alignment 
>> of target type [-Werror=cast-align]
>>     rp=(R *)p->fr->frame->data[0]; \
>>        ^
>> ffsox_frame_convert.c:158:1: note: in expansion of macro 'CONVERT_II'
>>
>> I'm not quite sure how to fix it.
>
> "p->fr->frame->data[0]" is a generic pointer defined in the "AVFrame" 
> structure from the FFmpeg API. It is of type "uint8_t" (cf. e.g. top 
> of https://www.ffmpeg.org/doxygen/2.3/structAVFrame.html) but may 
> point to data of an arbitrary type. The code triggering the error cast 
> the (generic) pointer back to it's real type.
>
> Those kinds of interfaces utilizing generic pointers which has to be 
> cast back their real type are a common C technique and not an error.
>
> To fix the error drop "-Wcast-align" from "CFLAGS" in the respective 
> "Makefile.am".
>
> The fix is available with version 0.4.4 available from 
> http://sourceforge.net/projects/bs1770gain/files/bs1770gain/0.4.4/.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-multimedia-maintainers/attachments/20150624/61becaf1/attachment.html>