Re: Thoughts from FreeBSD on the upcoming GNOME 2.2

[Thomas Vander Stichele <thomas urgent rug ac be>]

Hi,

> So far, I have to say GNOME has been pretty good for FreeBSD.  We have
> been tracking fairly closely to the official releases.  I'm putting the
> final touches on 2.1.2 as we speak.  However, with the recent addition
> of GStreamer to GNOME 2, I have some doubts.  GStreamer still does not
> work on FreeBSD.  Our pthreads implementation does not allow
> applications to longjump between thread contexts.  Since it looks like
> GStreamer might play an important role in GNOME 2, I thought I'd add
> some input.  If this is not the correct list, please let me know.

This looks like the correct list yeah ;)

We (= gstreamer) know there are issues with pthread code we have.  We have 
been contacted by freebsd people in the past, but most of them were only 
interested in trying out working code, not helping us fix the few issues 
there are with the code we have.  We indeed don't have access to FreeBSD 
machines ourselves to test.

I'm not quite sure what the pthread and libc situation is on freebsd, but 
I've been told that freebsd doesn't use glibc by default, but can use it 
for applications, and in that respect I would expect GStreamer pthread 
code to work when using glibc.  I'm not sure though and haven't found 
anyone I know with freebsd and glibc knowledge to back that up.

In any case, a lot of work has been going on to solve this problem in a 
more generic way, which includes :
a) moving to GThreads and working around it's limitations (we have valid 
reasons for doing the pthread hacks we do, and anyone who thinks he can 
work around it and help us remove limitations so that we don't need to do 
those evil code things is more than welcome to help us out !)
We're currently smack middle in that move to GThreads and fixing issues 
popping up because of it, which is why we need a few testers on various 
platforms to verify we're doing this the right way

b) writing a scheduler (optimal scheduler) that doesn't even need 
cothreads, and thus can do without any sort of GThread or pthread code.

Both are good ways to work around the problem on other platforms, we 
should be able to stabilize a) soon enough, b) will take some more work 
but will be better in many respects.

Our main problem is - we don't have any testers on platforms other than 
ppc and x86.  We've been actively looking for them on other platforms, but 
people seem somehow not inclined to test on other platforms.  We're not 
quite sure how to solve that issue ;) Any ideas here are welcome.

Thomas

 -- 

The Dave/Dina Project : future TV today ! - http://davedina.apestaart.org/
<-*- thomas (dot) apestaart (dot) org -*->
Smelled you on my hand for days
I can't wash away your scent
I'm a dog and you're a bitch
<-*- thomas  (at) apestaart (dot) org -*->
URGent, the best radio on the Internet - 24/7 ! - http://urgent.rug.ac.be/