[gst-devel] Results from some quick tests of Totem

[Benjamin Otte]

On Sat, 27 Mar 2004, Andy Wingo wrote:

> On Tue, 2004-03-23 at 19:57, Benjamin Otte wrote:
> > I consider opt a broken dead end.
>
> For the record, I don't.
>
:)

> >  opt does things that the current
> > GStreamer model should not allow - namely taking ownership of buffers.
>
> Explain why this is wrong.
>
1) clock sync. If the scheduler takes ownership of too many buffers on
for example the video side, the video sink can starve while the audio
queue is full.
2) Queries (maybe some other events, too). If a query passes over a bufpen
with buffers, it can't work correctly.

> > Another thing that most likely not work with opt is removing elements from
> > a playing pipeline (or a paused one).
>
> Why don't you try it? Perhaps you should go and look at the email I sent
> about a month ago regarding the event spawner. Elements added and
> removed, linked and unlinked, all when the pipeline is PLAYING. Try that
> in basic*.
>
I wasn't talking about the general case, but about special cases.
Like this:
loop1 ! loop2
Loop1 always puts out 2 buffers in one iteration, loop2 pulls 1 buffer on
first iteration and 2 buffer on every iteration following it. Now you try
to unlink them.

Benjamin