[gst-devel] downstream allocated buffers

[Benjamin Otte]

On Wed, 12 Jan 2005, David Schleef wrote:

> > I would like to see the old GstBufferPool resurected as it allows for
> > this kind of operations.
>
> Um, no.  We have a replacement that actually works.
>
1) I'd like to have a GstBufferPool object in an extension library so
elements that indeed have a pool of buffers they want to provide upstream
can use this instead of having it to implement themselves.

2) The replacement doesn't work good enough yet because it's still not
spec'ed out well enough what you may or may not do with a downstream
allocated buffer (or other special buffers). I think it's still easy
enough to create segfaults and other madness by messing with the right
elements in the right places, though I haven't tried to break
bufferalloc functions v4l(2) or x(v)imagesink recently by setting weird
states on them.

There's also the topic of the different constraints of bufferalloc
functions and how we could implement those. Examples here include filesink
providing mmap'ed buffers that point to a certain predefined location and
alsasink providing mmap'ed memory from its ringbuffers that needs to be
filled until $TIME.

Benjamin