[FFmpeg-devel] [PATCH] avformat/mov: remove hack breaking creation time parsing

[Marton Balint]

On Sun, 9 Apr 2023, Michael Niedermayer wrote:

> On Sun, Apr 09, 2023 at 07:52:12PM +0200, Marton Balint wrote:
>>
>>
>> On Sun, 9 Apr 2023, Michael Niedermayer wrote:
>>
>>> On Sun, Apr 09, 2023 at 03:49:33PM +0200, Marton Balint wrote:
>>>>
>>>>
>>>> On Sat, 8 Apr 2023, Michael Niedermayer wrote:
>>>>
>>>>> On Sat, Apr 08, 2023 at 08:37:24PM +0200, Marton Balint wrote:
>>>>>> Commit 23eeffcd48a15e73fb2649b712870b6d101c5471 added a hack to support invalid
>>>>>> files where the creation date was encoded as a classic unix timestamp. This
>>>>>> broke however valid files having creation dates before the unix epoch.
>>>>>>
>>>>>> Signed-off-by: Marton Balint <cus at passwd.hu>
>>>>>> ---
>>>>>>  libavformat/mov.c | 3 +--
>>>>>>  1 file changed, 1 insertion(+), 2 deletions(-)
>>>>>
>>>>> This results in:
>>>>> @@ -1,11 +1,11 @@
>>>>> -    creation_time   : 2012-06-20T20:58:31.000000Z
>>>>> -      creation_time   : 2012-06-20T20:58:31.000000Z
>>>>> -      creation_time   : 2012-06-20T20:58:31.000000Z
>>>>> +    creation_time   : 1946-06-20T20:58:31.000000Z
>>>>> +      creation_time   : 1946-06-20T20:58:31.000000Z
>>>>> +      creation_time   : 1946-06-20T20:58:31.000000Z
>>>>>
>>>>> Are you sure that 1946 is the correct creation date and not 2012 ?
>>>>
>>>> If you are referring to the file in ticket #1471, yes, 1946 is consistent
>>>> with what mediainfo shows for creation time. Obviously 1946 was not the
>>>> intended creation time, but that does not warrant us to break files where
>>>> 1946 is the *intended* creation time. Proper way to fix the original issue
>>>> would be to detect the device and software version which produces the
>>>> invalid files, and only apply the hack there. But I don't think that is
>>>> doable here, the file does not seem to contain any device or software
>>>> information.
>>>
>>> what do you mean by intended creation time?
>>> the file format did not exist in 1946. and all the codecs also didnt exist
>>> so when you encounter a file that says its from that time it must be crafted
>>> later and backdated or that bug.
>>> we know the bug is a real thing
>>> do you want to support crafted and backdatred files? if so can you explain
>>> the usecase for that ?
>>
>> http://ffmpeg.org/pipermail/ffmpeg-user/2023-April/056265.html
>>
>> Alternatives I can think of:
>>
>> 1) A -unix_time switch what Anton proposed
>
>> 2) doing strict compliant parsing only if mdat version is 1 so creation time
>> is 64bit. And change our muxer to write mdat version 1 by default, so ffmpeg
>> will be able to read back what it has written...
>
> What do we know about the buggy files that need this correction ?

Not much. Samsung Galaxy Nexus phone and Samsung HMX-S16BP camcorder seems 
affected at least. But not exclusively. I found these files among the 
samples:

Samsung HMX-S16BP:
samples/ffmpeg-bugs/roundup/issue2517/HDV_0112.MP4
samples/ffmpeg-bugs/roundup/issue2517_HDV_0113.MP4

Galaxy Nexus:
Sample in ticket 1471.

Unknown:
samples/ffmpeg-bugs/trac/ticket2095_385 Deadlist Form 2-7-13.mp4
samples/ffmpeg-bugs/trac/ticket3399_VID_20130619_161750_449.mp4
user/aac-input-buffer-exhausted-up_1434137794_VID_20120604_172442.mp4

> Is there any hint/metadata that identifies the muxer/encoder/version ?

In case of HMX-S16BP yes, in other cases no.

>
> Limiting the correction to the cases that need it is a good idea

I guess I will start with limiting workaround for version 1 as a start.

> Iam not sure i feel positive about changing the muxer

Eventually we have to bump the mdhd version, because after 2040 creation 
time will overflow. We should not wait until 2040 to do that, neither we 
should start writing version 1 only after 2040, that would be Y2K problem 
waiting to happen.

Regards,
Marton