[FFmpeg-devel] IMF demuxer ping

[Lynne]

5 Dec 2021, 02:33 by pal at sandflow.com:

> Hi all,
>
> Quick ping re: libavformat/imf demuxer patch set.
>
> All outstanding feedback (thanks!) has been addressed as far as I know.
>
>  What are the next steps?
>
> It would be good to make progress while it is fresh in peoples' minds.
>
> Latest patchset at [1] and tracking PR at [2] (including memory leak checking):
>
> [1] http://ffmpeg.org/pipermail/ffmpeg-devel/2021-November/288173.html
> [2] https://github.com/sandflow/ffmpeg-imf/pull/74
>
> Looking forward to (finally) adding support for IMF demuxing to ffmpeg.
>

There are still many issues.
Put your custom header beneath the standard ffmpeg header. Or
remove it altogether if you're allowed to.
You're not following our code style at all:
All internal functions (except intra-file functions) must have
an ff_<name> prefix. If it's a struct, the prefix is FF<name>.
We do not put brackets on one-line statements, unless it's
as a part of an if {} else { } group where one statement has
more than 2 lines.
We're lenient towards 80-char line limits, where if it looks bad
and hampers readability, we let it go on, but the patches aren't
even trying.
It's all here:
https://ffmpeg.org/developer.html#Coding-Rules-1

"rsrc->edit_rate = av_make_q(0, 0);" <- this means that if
something goes wrong, there will be a division by 0. Timebases
must be initialized by av_make_q(0, 1);

Could you try to reduce the total number of allocations per-packet
by allocating all that you can during init-time, even if it's potentially
unused, and use av_fast_realloc for really necessary allocations during
packet parsing? That way, decoding would be quicker.

Squash the first 3 patches (header, processor and demuxer) patches
together, and move the build stuff from the final patch in with the
first, only keeping the Makefile code for building the tests into the 4th.
Otherwise, you can't build them.

Remove the "@author" and "@file" or other special documentation
commands, doxygen doesn't even use them. Use the copyright
field on top.

Since IMF Is meant for intermediate work, it does allow for weird
non-linear streams with loops, skips and such. We do (or attempt
to) do linearization inside the demuxer, due to bad choices, unless
an option is set, like with mp4/mov. So this demuxer should do
that as well.