Re: [gst-devel] Re: RFC: GNOME 2.0 Multimedia strategy

[Alan Cox <alan redhat com>]

>   The whole debate sounds to me that the current pthread implementation ain't
> good enough. I would rather see it fixed than build the co-thread routines
> within the multimedia libray...

They solve two different problems. Its unfortunate that people muddle the
elegance of co-routines with the ugly great splat called threads.

Co-routines
	-	Require under 1K of code
	-	Leave the code flow expressed explicitly thus avoiding near
		impossible to debug races

PThreads
	-	Require a large library
	-	Require the app does locking
	-	Are not reliable in glibc 2.0 and iffy in places in 2.1

Unless you want the parallelism (and you need to want it badly as most pthread
apps have shared data problems so dont scale anyway) co-routines have a lot
to offer in terms of writing code as multiple state/event loops not one.

If you are going to use pthreads you need to remember that every time a thread
on CPU#1 touches a resource on CPU#0 you will generate a lot of bus traffic,
stalls until the data passes between the processors and actually end up doing
little useful work.