On Sun, 12 Jun 2016, Nicolas George wrote:
Le quartidi 24 prairial, an CCXXIV, Marton Balint a écrit :
Maybe I am missing something, but as far as I see the alloc is due to the
alloc in ff_bsf_get_packet. We could create a similar function
ff_bsf_get_packet2(AVBSFContext *ctx, AVPacket *pkt) which does not allocate
a packet but simply moves the ref to pkt. Then this can be called with the
output packet pointer, so no allocation will be requried. And this function
can be used for performance sensitive places instead of the original.
Indeed, maybe it is as simple as that.
I think we need to support both ways to configure the bsf list. What the tee
muxer, or ffmpeg.c would use, is the text based configuration, therefore
it's worth adding the capability of parsing such lists.
That is exactly what I was suggesting.
So what we need, on top of what I proposed, is an additional API what you
proposed (av_bsf_list_append), and which can be used to add filters to the
list, after we allocated the list filter, but before we initialized it.
No, definitely no. It must be the other way around. Building a text API on
top of a programmatic is fairly common, building a programmatic API on top
of a text API is bad design.
Bad wording on my part, I did not mean "on top of" API wise :). Agreed,
surely the text parsing can use the av_bsf_list_append functions to do the
actual work.
Thanks,
Marton
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
