[mythtv] Call for testing devel/ffmpeg-resync (FFmpeg 5.1) re PR#610

[John Pilkington]

On 08/09/2022 00:37, Scott Theisen wrote:
> On 9/7/22 17:43, John Pilkington wrote:
>> On 07/09/2022 19:22, Peter Bennett wrote:
>>>
>>> On 9/7/22 06:05, John Pilkington wrote:
>>>> On 06/09/2022 22:46, Peter Bennett wrote:
>>>>>
>>>>> On 9/6/22 14:46, Scott Theisen wrote:
>>>>>> Hello all,
>>>>>>
>>>>>> I have merged FFmpeg 5.1 into our version and copied it into 
>>>>>> MythTV. Peter Bennett has pushed these commits from 
>>>>>> https://github.com/MythTV/mythtv/pull/627 to devel/ffmpeg-resync 
>>>>>> and we would appreciate further testing before it is merged to 
>>>>>> master.
>>>>>>
>>>>>> The deprecated avcodec_encode_video2 has been removed.  I modeled 
>>>>>> the replacement code on the audio version and would appreciate 
>>>>>> some testing of transcoding to make sure my replacement works. 
>>>>>> There is also a similar change to NuppleVideoRecoder, but I don't 
>>>>>> know if anyone actually still uses that.
>>>>>>
>>>>>> FFmpeg has created a new channel layout API and deprecated the old 
>>>>>> one.  This effects downmixing and I would appreciate confirmation 
>>>>>> that it still works as expected on non-SSE2 platforms.
>>>>>>
>>>>>> Those were the main things I wanted tested, but general testing 
>>>>>> would also be appreciated.
>>>>>>
>>>>>>
>>>>>> Klaas,
>>>>>>
>>>>>> Could you test the attached patch on top of devel/ffmpeg-resync to 
>>>>>> test if https://trac.ffmpeg.org/ticket/9532 is still a bug? Apply 
>>>>>> the patch in the mythtv/external/FFmpeg folder and test VDPAU 
>>>>>> decoding with your sample.
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>> Scott Theisen
>>>>>>
>>>>>>
>>>>> I still have that video on my test system. IIRC it used to 
>>>>> pixellate in various places in the video before Klaas's patch. 
>>>>> Using VDPAU. with the ffmpeg-resync version it plays perfectly. 
>>>>> With the ffmpeg-resync version and the revert patch it pixellates 
>>>>> badly for 2-3 seconds at the start of the video but not at any 
>>>>> other place. So the problem may have partially been fixed in ffmpeg.
>>>>>
>>>>> Do you have a version of the revert patch for master so that I can 
>>>>> apply it there as a comparison for how it used to be?
>>>>>
>>>>> Peterdevel/ffmpeg-resync
>>>>
>>>> Yes, Klaas' clip plays for me with no problems in ffmpeg-resync 
>>>> VDPAU 676d7f5, although mythffprobe gives a couple of pages of 
>>>> complaints before the analysis.
>>>>
>>>> And I don't want to distract from the main business of integrating 
>>>> 5.1, but wonder if PR #610 has been overlooked?
>>>>
>>>> Cheers,
>>>>
>>>> John P
>>>> _______________________________________________
>>>>
>>> John
>>>
>>> Have you tested PR #610 ? If it is working correctly I can merge it 
>>> into master, fixes/32 and rebase devel/ffmpeg-resync so that it is 
>>> included there as well.
>>>
>>> Peter
>>
>> Hi Peter:  The tests that I did were described in comments on issue 
>> #548 on 19 July.  They didn't use the commit from the PR, but a bunch 
>> of six patches received as attachments, which Scott later combined to 
>> create the PR after seeing my logfile.  I haven't yet found the magic 
>> to create an acceptable single git patch from the PR that I can use 
>> for a new build of either master or devel/ffmpeg-resync.  Sorry.
>>
>> John
> 
> Pull Request 610 https://github.com/MythTV/mythtv/pull/610 is only one 
> commit.  I trimmed the extra logging that was for debugging only.  That 
> one commit is functionally equivalent to the six patches from 
> https://github.com/MythTV/mythtv/issues/548 .  It should apply cleanly 
> to both master and devel/ffmpeg-resync.
> 
> Scott

Google revealed that the patch would be at

https://github.com/MythTV/mythtv/pull/610.patch

I applied that patch to devel/ffmpeg-resync and have rerun the tests 
from 19 July.

New recordings from DVB-T BBC1 and DVB-T2 BBC1_HD still show their 
correct durations.

A new recording from one of the channels in the 'local' DVB-T mux now 
shows its correct duration.  A similar recording made with the unpatched 
system still shows duration and frame-count both double their true values.

I think that for MPEG2 encoded recordings issue #548 is fixed.

John P