[gst-devel] Rowstrides / buffer hierarchy? (was: Re: summit conclusions + plans)

[Ronald S. Bultje]

Hi,

On Sun, 2005-02-20 at 20:55 +0000, Tim Müller wrote:
> Has anything been decided on how to tackle the problem of implicit rowstrides, 
> plane offsets and buffer sizes for image formats? Are they going to be ...
>  (b)  attached to image buffers

This is not possible. You need to have the information available during
negotiation, otherwise you will send out invalid buffers to downstream
elements.

>  (c)  taken care of by centralising format descriptions etc.
>         in a plugins support library that provides the stuff that
>         is now in videotestsrc, gstvideoimage, ffmpegcolorspace,
>         and scattered around the code tree in form of #defines.

Then you still need to put it somewhere. There will of course be utility
functions.

>  (d)  a combination of (b) and (c)
>  (e)  none of the above

No, so that leaves us with one option:

>  (a)  made part of caps (yuk)

Yes, yuck. Anyway, there's various cases that need to work:
A) fixed-stride buffers
B) fixed-byte alignment
Both need to be negotiated correctly. That is hard, because of
dependencies between options in one case and the other case and the fact
that those clash in a very inconvenient way.

In short, someone needs to come up with a really good proposal that
makes both cases work. Until then, stride is implicit to image format
and size like it is now, even though that may have a negative impact on
performance. I have some ideas on this, I'll spam those on-list or in
CVS at some point.

Ronald

-- 
Ronald S. Bultje <rbultje at ronald.bitfreak.net>