On Thu, May 29, 2003 at 02:00:06PM -0700, Bill Studenmund wrote:
> Yes, there are still issues, such as SMP-unfriendliness and the main stack
> size issue. They need fixing. But that doesn't mean we don't have a good
> pthreads package now.

I partly agree. What we have now is better than what we had in 1.6. And
the assertions in libpthread make it realy easy to find badly written
apps - wich happen to be out there a lot.

But I'm not able to call the current state "good". The "SMP-unfriendliness"
is a killer. And the interactive feel of threaded apps (sorry, can't realy
quantify this) is very bad, compared to such inferior systems as Win2K (when
running on the same hardware). Of course, Win2K uses all CPUs. Probably
thread priority based scheduling and increased priority for the GUI thread
could help this.

I still think what we did is right. The merge was long overdue. What we
need now is more people getting a clue at the innards of libpthread and help
improving it. I see progress, steady, but slow. We need to speed it up, and/
or stay patient for a while. This is -current, after all.

Now I'll point back at my own nose and notice that we need SMP support on more
archs ;-} and this is taking way too long too.

Martin