Linux/Alpha show-stopper

["Adam C. Powell, IV" <adam powell nist gov>]

Hello,

This is following up on my message of last Tuesday- every GNOME 0.99 app
and three of the background tasks each take a full load point.

Finally had time to apply my neophyte skills to the problem, and found
that apps like xclock, emacs, gimp (1.0) etc. call __select in libc when
idle.  GNOME apps (tried gnome-help-browser, gnumeric and gnome-stones)
call __syscall_poll in libc when idle.  (Tested by 'ddd ap', run, and
repeat interrupt/cont; GNOME apps are *always* interrupted in
__syscall_poll, others *never* there.)  I think this may be why GNOME
apps take a full load point each and others don't.

A little searching showed poll (which calls __syscall_poll) being called
from gtk_main (via glib's g_main_run).  poll also calls __select.  So
why are GNOME apps always polling, and others sitting in select?

I also found something mysterious about "polling frequency" buried in
gnome-libs/gtk-xmhtml/plc.c.  Could there be some miscalculation of such
a frequency on alpha?  Likewise with panel applets- is their frequency
too slow, so they only update on X events?  (This is a separate problem,
but may be related.  None of the panel applets ever update unless
I mouse over- including the clock, which made me miss an appointment
last week!!!)

Well, this is quickly getting beyond my knowledge, so I'll stop here.  I
have yet to test any non-GNOME gtk+-1.1 apps (I don't have any).  Any
suggestions of which to try?

Also, a friend suggested nicing GNOME apps as a workaround, but that
doesn't solve the problem of gnome-session, gnome-name-service and gmc
loading the CPU.  If I nice gnome-session, then everything I run in that
session is also niced, and only root can use negative nice numbers to
run at higher priority.  So much for that workaround. :-(

IMHO, this is a very serious problem which shouldn't be allowed into the
1.0 release.  I don't use GNOME any more because of it.  My posts to
axp-list@redhat.com saying, "Has anyone else tried 0.99?  Has anyone
else seen this problem?" have not been answered, which I think means
that they are scared off by the problem- and rightly so.  Meanwhile,
someone just yesterday posted availability of alpha rpms for Qt and
KDE...

Hope this helps this problem to get solved.  I'd be glad to help out
with testing, in my limited extra time at work.

Zeen,

-Adam `Cold Fusion' Powell, IV http://www.ctcms.nist.gov/~powell/ ____
USDoC, National Institute of Standards & Technology (NIST)  |\ ||<  |
Center for Theoretical and Computational Materials Science  | \||_> |