[FFmpeg-devel] Patch review: av_dict: add support for empty values

[Peter Cordes]

I've made some patches, and am finally getting around to sending them
upstream.

I'm trying out gmail for dealing with a high-traffic list like this one,
instead of my usual mutt.  Are the one-line-paragraph long lines it's
probably going to produce an annoyance for anyone?  I didn't notice
anything about line lengths in emails in
https://www.ffmpeg.org/developer.html

I put in pull requests on github for 4 different branches.  Should I send
email to the list for each set of patches?  Should I git-send-email, or is
a mail with a URL for the github pull request enough?  (Or is just a pull
request enough?  developer.html seems to say that sending a pull request is
an alternative to mailing patches to the list.)


Anyway, the av_dict change is the one that requires the most review, so
I'll make this email focus on that set of changes.
https://github.com/FFmpeg/FFmpeg/pull/118

  Currently, -x264-params nocabac:ref=4:ssim  isn't supported.  You have to
say nocabac=1:ref=4:ssim=1  (or cabac=0...).  (and yes, I know about
-preset, but superfast no-cabac is useful... anyway.)

 Also, libx265.c is very minimal, and needs a lot of things passed in
-x265-params.  I didn't fix that, but did make it support keys with no
value.

libx264 has two different ways of passing option strings (-x264-params and
-x264opts), due to avconv gratuitously renaming things.  The insane part is
that the two strings were parsed differently.  One was handled by av_dict,
the other with some custom code that didn't support quoting.

In my x26x-opts branch (which follows the av_dict branch), I add both
option strings to the same av_dict.  (x264opts and then x264-params, in
that order, regardless of order on the command line, same as before).  I
don't think it's ever useful to use the same key multiple times.  Tune only
works as -tune ssim, not in x264-params, and you can use a comma-separated
list for -tune animation,ssim.

There was a previous attempt at keys with no value that didn't get merged:
 http://ffmpeg.org/pipermail/ffmpeg-devel/2013-July/146329.html

I should borrow the idea of adding a new flag for supporting NULL as a
value, instead of the cumbersome empty-string method I went with to
minimize changes to some parts of the code.  Having a flag would make sure
the behaviour didn't change for any callers that weren't expecting it.
(i.e. it would still reject key:key).

So are any of these changes useful?  I assume I'll end up redoing the
commits anyway after a review, so I didn't try to make a final
ready-for-commit version yet.  (For that, I would merge the bugfix changes
into my original buggy commit.)  For review, I'd suggest looking at the
diff of master:av_dict.  The 3 commits after that changing libx264.c and
libx265.c should each stand on their own.  (And actually the libx264.c one
should be split into a couple commits.)

Anyway, I wanted to post first and get some feedback before spending more
time on it.

Thanks,

--
Peter Cordes <peter at cordes.ca>