[FFmpeg-devel] [PATCH] Implement NewTek NDI support

[Maksym Veremeyenko]

On 23.07.2017 18:07, Nicolas George wrote:

[...]
>> Maybe it's just me, but I am not sure what kind of probing are you referring
>> to. Could you explain in a bit more detail how the changed code should work?
>
> The same way it works for formats without global header, for example
> MPEG-PS: read_header() does not create any stream; read_packet() creates
> the stream when it sees a packet for one that does not exist yet. And
> libavformat takes care of the rest. I think it requires
> AVFMTCTX_NOHEADER.
>
absolutely

>> I guess that is copied from decklink as well. As far as I see, the correct
>> replacement is to set codecpar->field_order in read_header, and that is it,
>> right?
>
> I think so.
>
ok

>> I think it is better if it works with unkown channel layouts as well, as
>> long as the number of channels are supported.
>
> I do not agree, and especially not in the device itself. There are many
> small issues where some component invents a default layout when only a
> count is provided, and then we are unable to propagate a count without a
> layout.
>

i seen no hard limitation about audio channel numbers in SDK, so let's 
leave it for future

>>>> +        av_log(avctx, AV_LOG_ERROR, "Only one video stream is supported!\n");
>>>> +        return AVERROR(ENOTSUP);
>>> I suspect this library exists also for Windows and Macos, so you cannot
>>> use ENOTSUP. EINVAL.
>> Or maybe ENOSYS?
>
> ENOTSUP -> "Operation not supported" / "Not supported"
> ENOSYS  -> "Function not implemented" / "Function not supported"
> EINVAL  -> "Invalid argument"
>
> In this case, the problem is that the avctx argument has several
> streams: EINVAL is really the correct one.

i can use whatever you prefer

-- 
Maksym Veremeyenko