[FFmpeg-devel] [PATCH 4/5] lavfi: drop the requirement that request_frame returns a frame.
george at nsup.org
Sun Sep 20 19:26:00 CEST 2015
Le decadi 10 fructidor, an CCXXIII, Michael Niedermayer a écrit :
> there may be one problem though
> if the parameters of Frames change and a link "contains" 5 frames
> it is possible that not all these frames have the same size or
> aspect ratio. Currently a link contains dimensions and aspect ratios
> iam not sure what is the best solution or if this even matters ATM
> or maybe is better to ignore for now and deal with at a later point
> when the design is closer to finished and implemented.
> One easy solution would be to split the values for pre and post fifo
> dunno if that is a good or bad idea ...
For filters that are implemented according to the old request_frame() /
filter_frame() design, then updating the link properties when the frame is
dequeued from the head of the FIFO and fed to filter_frame() seems like the
only sane choice anyway.
For filters implemented according to the future new design, we must make
sure they use the frame fields (thanks Paul for reminding us they exist)
instead of the link fields. If the link fields are used somewhere to carry
information not present elsewhere (I suspected get_video_buffer() may be an
example, but it takes w and h as arguments), then I guess it would be the
responsibility of the caller to ensure that the fields have the correct
Anyway, format reconfiguration is not supposed to work at this point; it
works for a few filters by luck/default and with exceptions to bypass
consistency checks, but if we want to make it more standard, it would
probably have to be associated with a flag FILTER_SUPPORT_RECONFIGURATION or
something: checking that everything is done consistently with the fields
should happen before enabling the flag for a particular filter.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 819 bytes
Desc: Digital signature
More information about the ffmpeg-devel