[FFmpeg-devel] [RFC]Improve adp detection

[Paul B Mahol]

On 10/30/13, Michael Niedermayer <michaelni at gmx.at> wrote:
> On Wed, Oct 30, 2013 at 10:32:13AM +0000, Carl Eugen Hoyos wrote:
>> Michael Niedermayer <michaelni <at> gmx.at> writes:
>>
>> > On Wed, Oct 30, 2013 at 08:39:34AM +0000, Carl Eugen Hoyos wrote:
>> > > Michael Niedermayer <michaelni <at> gmx.at> writes:
>> > >
>> > > > if you print them for the files we have do some of these 4bit
>> > > > values never occur ?
>> > >
>> > > I suspect only 0, 1, 2, 3 actually occur but that does not
>> > > help for the "000000000000000..." case or do I miss something?
>> >
>> > if only 0123 occur then that could be checked for
>>
>> This would break (theoretical) files that the decoder accepts
>> without fixing anything.
>
> the decoder accepts any input, even any that fails the current
> probe function.
>
> Also our decoder is not the official specification of this format,
> just because it accepts something doesnt mean its actually valid
>
> And being valid doesnt mean it actually occurs in the wild
>
> ideally anything valid and anything decodeable should be probed with
> high score but that doesnt work here as /dev/random is decodeable
> and we dont know whats valid as we have no specification, so we can
> only guess here
>
> whats left is to try to detect what actually exists in the wild and
> apply some common sense on top of that
>
> the 4 bits select the predictor, only 4 predictors are defined in
> the decoder. I suspect its unlikely that if a 5th value would occur
> that it would just be treated as one of the 4. But thats a guess of
> course ...

Huh, one byte have 8 bits. 4 bits are predictors and rest of bits
are used in decoding.

Also some bytes are always skipped - so one need to inspect what
real files have there.

>
>
>>
>> > it doesnt fix the 0000000 issue
>>
>> Could that be intended behaviour?
>
> i see nothing that would make the case of all header bytes being 0
> to be non valid
> in combination with the other bytes being 0 too though, 0 headers
> might be non occuring with actual encoders
>
> [...]
> --
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> Into a blind darkness they enter who follow after the Ignorance,
> they as if into a greater darkness enter who devote themselves
> to the Knowledge alone. -- Isha Upanishad
>