On Wed, 20 Mar 2013 15:59:32 +0100, Janne Grunau <[email protected]> wrote:
> On 2013-03-12 12:48:35 +0100, Anton Khirnov wrote:
> > Needed e.g. for h264 cropping to work properly.
> > ---
> > libavcodec/utils.c | 23 ++++++++++++++++++++---
> > 1 file changed, 20 insertions(+), 3 deletions(-)
> >
> > diff --git a/libavcodec/utils.c b/libavcodec/utils.c
> > index b4c294e..bec0b3f 100644
> > --- a/libavcodec/utils.c
> > +++ b/libavcodec/utils.c
> > @@ -549,14 +549,19 @@ static void compat_release_buffer(void *opaque,
> > uint8_t *data)
> >
> > int ff_get_buffer(AVCodecContext *avctx, AVFrame *frame, int flags)
> > {
> > + int orig_width = 0, orig_height = 0;
> > int ret;
> >
> > switch (avctx->codec_type) {
> > case AVMEDIA_TYPE_VIDEO:
> > if (!frame->width)
> > - frame->width = avctx->width;
> > + frame->width = FFMAX(avctx->width,
> > avctx->coded_width);
> > + else
> > + orig_width = frame->width;
> > if (!frame->height)
> > - frame->height = avctx->height;
> > + frame->height = FFMAX(avctx->height,
> > avctx->coded_height);
> > + else
> > + orig_height = frame->height;
>
> under which conditions would frame->{height,width} be set? reusing a
> frame? Why are we trusting those values for allocation? Especially
> since a zeroed frame gets height and not MAX(height, coded_height). If
> that frame is reused later we would allocate buffers not large enough
> for the coded dimensions if those are larger.
>
A decoder might want to allocate a slightly larger frame for whatever reason.
AFAIK no decoders do that currently, but Justin requested overriding the
channel_layout in this way and it made sense to me to allow this for all the
other parameters as well. We assume that if the decoder set those values
explicitly, it knows what it's doing.
> > if (frame->format < 0)
> > frame->format = avctx->pix_fmt;
> > if (!frame->sample_aspect_ratio.num)
> > @@ -696,6 +701,11 @@ do {
> > \
> >
> > av_buffer_unref(&dummy_buf);
> >
> > + if (!orig_width)
> > + frame->width = avctx->width;
> > + if (!orig_height)
> > + frame->height = avctx->height;
> > +
>
> see above, also what happens for audio? do we trust that avctx->width
> and height are zero for audio codecs?
I cannot imagine a sane reason why they would be non-zero, but I can wrap this
part in if (video) to be extra safe.
--
Anton Khirnov
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
