[FFmpeg-devel] [PATCH] avcodec: Remove libfaac, the internal AAC encoder is better

[Rostislav Pehlivanov]

On 10 April 2016 at 17:42, Michael Niedermayer <michael at niedermayer.cc>
wrote:

> On Sun, Apr 10, 2016 at 04:38:35PM +0100, Kieran Kunhya wrote:
> > ---
> >  Changelog              |   1 +
> >  configure              |   6 --
> >  doc/encoders.texi      | 105 ---------------------
> >  doc/ffserver.conf      |   2 +-
> >  doc/general.texi       |   2 +-
> >  doc/muxers.texi        |   4 +-
> >  doc/platform.texi      |   2 +-
> >  libavcodec/Makefile    |   1 -
> >  libavcodec/allcodecs.c |   1 -
> >  libavcodec/libfaac.c   | 248
> -------------------------------------------------
> >  libavcodec/version.h   |   2 +-
> >  11 files changed, 7 insertions(+), 367 deletions(-)
> >  delete mode 100644 libavcodec/libfaac.c
>
> this is not possible currently libfaac is twice as fast than the
> native encoder.
>
> time ./ffmpeg -v 0 -i matrixbench_mpeg2.mpg -vn -c:a libfaac -y test.aac
> real    0m2.828s
> user    0m2.776s
> sys     0m0.048s
>
> time ./ffmpeg -v 0 -i matrixbench_mpeg2.mpg -vn -y test.aac
> real    0m5.908s
> user    0m5.856s
> sys     0m0.048s
>
>
>
FAAC isn't maintained, hasn't had any work done on it in who knows how many
years, nobody but people who don't know that the native encoder/fdk is
better use it (just a few thankfully), isn't particularly stable
(segfaulted a few times when I was comparing it last year) and finally,
it's not good at all.
An argument that it's faster than the native encoder has as much weight as
an argument that libaac_plus was also faster than the native encoder, which
didn't matter as it was eventually removed
The age where we needed a few different AAC encoders because there wasn't
really a single good multipurpose one is gone now. The times have changed
since FAAC was developed (Nokia sponsored at lot of its development, and
you know what they used to make) and so have the computers. What was an
acceptable speed back then for encoding a file at a given quality isn't
necessarily the same now. And considering that fdk-aac can run as slow as
our encoder I'd say we're doing pretty well as far as the balance between
speed and quality goes.

With all this said, I think that we should remove it from git master.
There's plenty of time until the next major release and by then the users
of git master will have plenty of time to decide themselves (and inform in
case they don't know) whether they'd really spend a few more percent of CPU
for a comparatively big jump in quality and stability.
This decision will also probably affect distributions which are wondering
if it's worth maintaining and packaging that library as well. Ubuntu has a
nice collection of fairly critical bugs with the library (wrong duration,
segfaults, invalid bitstream, etc.) and I bet they'd be happy to drop it to
avoid any more piling up filed by uninformed users.