Quoting Martin Storsjö (2016-06-23 21:46:50)
> On Thu, 23 Jun 2016, Anton Khirnov wrote:
>
> > Quoting Martin Storsjö (2016-06-23 13:06:41)
> >> +AVCodec ff_libopenh264_decoder = {
> >> + .name = "libopenh264",
> >> + .type = AVMEDIA_TYPE_VIDEO,
> >> + .id = AV_CODEC_ID_H264,
> >> + .priv_data_size = sizeof(SVCContext),
> >> + .init = svc_decode_init,
> >> + .decode = svc_decode_frame,
> >> + .close = svc_decode_close,
> >> + .long_name = NULL_IF_CONFIG_SMALL("OpenH264"),
> >> + .capabilities = AV_CODEC_CAP_DELAY, // The decoder itself doesn't
> >> have delay, but the BSF might
> >
> > It does call ff_get_buffer(), so it should be flagged as DR I think.
> > Even though it's not "proper" DR.
>
> Hmm, good point. If not setting the flags, the avframe would be allocated
> and filled locally, right?
I think these days the flag makes almost no difference in lavc
internals. Its main purpose is just to signal to the caller that he
shouldn't expect his get_buffer() callback to be used.
>
> I guess I could do that as well (pointing to the buffers returned by the
> decoder); would the refcounting take care of duplicating it in case the
> caller wants to retain it past the next decode call?
No, refcounting cannot magically substitute the data in all the
references the caller might have made. I think the copy is unavoidable
here.
--
Anton Khirnov
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
