Hi,

On Mon, Nov 28, 2011 at 2:57 PM, Janne Grunau <[email protected]> wrote:
> On Sat, Nov 26, 2011 at 05:06:12PM -0500, Justin Ruggles wrote:
>> Add AV_NUM_DATA_POINTERS to simplify the bump transition.
>> This will allow for supporting more planar audio channels without having to
>> allocate separate pointer arrays.
>> ---
>>  doc/APIchanges       |    4 ++++
>>  libavcodec/avcodec.h |   11 ++++++++---
>>  libavcodec/version.h |    6 ++++--
>>  3 files changed, 16 insertions(+), 5 deletions(-)
>
> [...]
>> diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
>> index 43abcd8..1ea96a6 100644
>> --- a/libavcodec/avcodec.h
>> +++ b/libavcodec/avcodec.h
>> @@ -927,13 +927,18 @@ typedef struct AVPacket {
>>   * sizeof(AVFrame) must not be used outside libav*.
>>   */
>>  typedef struct AVFrame {
>> +#if FF_API_DATA_POINTERS
>> +#define AV_NUM_DATA_POINTERS 4
>> +#else
>> +#define AV_NUM_DATA_POINTERS 8
>> +#endif
>>      /**
>>       * pointer to the picture planes.
>>       * This might be different from the first allocated byte
>>       * - encoding:
>>       * - decoding:
>>       */
>> -    uint8_t *data[4];
>> +    uint8_t *data[AV_NUM_DATA_POINTERS];
>>      int linesize[4];
>
> keeping the four here looks inconsistent. I know that the intention is
> to use only linesize[0] if it's the same for all planes but that should
> be at least documented (can't remember if it's added later).
>
> I would extend linesize to AV_NUM_DATA_POINTERS too.
>
> Not extending it screams for bugs like the following:
>
> for (i = 0; i < nb_channels; i++) {
>        ...
>        memcpy(dst, f.data[i], f.linesize[i]);
>        ...
> }

I agree, but then again this is easily prevented by setting these
linesizes to -1.

The problem is that then we have to create an extended_linesizes*
also, and I really don't think that's right.

Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to