[FFmpeg-devel] [PATCH] avcodec/libmp3lame: properly handle unaligned frame data

[Muhammad Faiz]

On Tue, May 2, 2017 at 1:05 AM, Paul B Mahol <onemda at gmail.com> wrote:
> On 5/1/17, Muhammad Faiz <mfcc64 at gmail.com> wrote:
>> On Tue, May 2, 2017 at 12:45 AM, Paul B Mahol <onemda at gmail.com> wrote:
>>> On 5/1/17, Muhammad Faiz <mfcc64 at gmail.com> wrote:
>>>> On Mon, May 1, 2017 at 11:22 PM, Kyle Swanson <k at ylo.ph> wrote:
>>>>> Hi,
>>>>>
>>>>> On Mon, May 1, 2017 at 3:18 AM, Paul B Mahol <onemda at gmail.com> wrote:
>>>>>> On 4/30/17, Nicolas George <george at nsup.org> wrote:
>>>>>>> Le primidi 11 floreal, an CCXXV, Muhammad Faiz a ecrit :
>>>>>>>> Are you working on these? Because currently I'm not.
>>>>>>>
>>>>>>> There is nothing to work on yet: the message you answer to is raising
>>>>>>> a
>>>>>>> question about the global design of the internal API. That question
>>>>>>> needs an answer before any work can be done, and I can not decide
>>>>>>> alone.
>>>>>>>
>>>>>>
>>>>>> How nice, introducing bug that causes crash and then claiming there is
>>>>>> not such bug.
>>>>>>
>>>>>> Which filters you consider deemed worthy to not crash?
>>>>>> _______________________________________________
>>>>>> ffmpeg-devel mailing list
>>>>>> ffmpeg-devel at ffmpeg.org
>>>>>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>>>>
>>>>> Maybe 383057f8e744efeaaa3648a59bc577b25b055835 should be reverted
>>>>> until API stuff is sorted. This should also be backported to 3.3
>>>>> because these issues are present in that release.
>>>>>
>>>>> Kyle
>>>>
>>>> Of course no. Reverting it will make more bug.
>>>>
>>>> Unless
>>>> https://lists.ffmpeg.org/pipermail/ffmpeg-devel/2017-January/206285.html
>>>> and
>>>> https://lists.ffmpeg.org/pipermail/ffmpeg-devel/2017-January/206284.html
>>>> are also applied.
>>>
>>> One of those are already reviewed, other looks like is not needed at all.
>>> Could you elaborate why it is needed?
>>
>> The code before the patch write to unwritable frame.
>>
>> test-case:
>> ffplay -i lavfi 'aevalsrc=sin(1000*t*t), aformat=sample_fmts=fltp,
>> asplit [a][b];
>> [a] firequalizer=fixed=on, showcqt=s=1280x360 [a1];
>> [b] firequalizer=fixed=on, showcqt=s=1280x360 [b1];
>> [a1][b1] vstack'
>>
>> the data will be corrupted because fixed=on enables partial_buf_size stuff
>>
>> Compare that without fixed=on.
>
> Hmm, so it seems Nicolas blocked [1/2] patch which is mandatory for [2/2] one,
> time without providing alternative solution.

The alternative was 383057f8e744efeaaa3648a59bc577b25b055835, of
course. I approved it at that time.