[Ffmpeg-devel] Re: [PATCH] file length handling

[Ronald S. Bultje]

Hi Michael,

On Mon, 18 Dec 2006, Michael Niedermayer wrote:
> On Mon, Dec 18, 2006 at 09:20:49AM -0500, Ronald S. Bultje wrote:
> > -#define LIBAVFORMAT_VERSION_INT ((51<<16)+(6<<8)+0)
> > -#define LIBAVFORMAT_VERSION     51.6.0
> > +#define LIBAVFORMAT_VERSION_INT ((52<<16)+(0<<8)+0)
> > +#define LIBAVFORMAT_VERSION     52.0.0
> >  #define LIBAVFORMAT_BUILD       LIBAVFORMAT_VERSION_INT
>
> how is compatibility broken? if it isnt then theres no need to bump
> the version number up

I'm not sure if URLContext is external API or not, similar for
ByteIOContext. If not, then I'll remove this.

> >      s->seek(s->opaque, s->pos, SEEK_SET);
> >      return size;
> >  }
> > @@ -516,6 +519,7 @@
> >      }
> >      s->is_streamed = h->is_streamed;
> >      s->max_packet_size = max_packet_size;
> > +    s->file_size = h->file_size;
>
> the copying of possible non constant variables is bad, iam not sure how to
> best solve it though, maybe function ptrs would after all be better then a
> variable, dunno
[..]
> > Index: ffmpeg-mpe/libavformat/file.c
> > ===================================================================
> > --- ffmpeg-mpe.orig/libavformat/file.c	2006-12-10 12:54:52.000000000 -0500
> > +++ ffmpeg-mpe/libavformat/file.c	2006-12-18 09:02:53.000000000 -0500
> > @@ -110,6 +110,7 @@
> >  #endif
> >      h->priv_data = (void *)(size_t)fd;
> >      h->is_streamed = 1;
> > +    h->file_size = -1;
>
> checking against 0 and setting to -1 wont work
>
> and the default value should be set in url_open() or another common place
> (of course only if its nor zeroed anyway...)

The default is set to 0. file.c's file:// handler keeps this. The stream
handlers (tcp://, udp://, rtp:// and pipe://) set it to -1, because you
can't retrieve the length (i.e. -1 means error or no length available or
something). So this is intentional. The value of file_size is supposed to
be constant (as is the case for streams and also for http, since you only
get Content-Length once), so I think it should be safe to copy it into
ByteIOContext.

I'll make sure url_open() sets it to zero and remove the increase of the
LIBAVFORMAT_VERSION_INT and send a new patch tonight (not at home right
now...). Please let me know if the above is confusing or if it could be
done better differently.

Ronald