For the brave and foolhardy.

*  It's more stable and faster than I expected.  However this is with a R=
adeon=20
7000 and a 600au with 768 Megs.

*  It took 36 hours and gobs of disk to compile all KDE3 packages on this=
=20
600au.

How to compile from pkgsrc:

Please note I'm not familiar with crafting patches.  Any help I can get h=
ere=20
would be appreciated, but I took notes along the way.  *These notes are=20
selfish and inelegant and only work on a current netbsd-alpha.*  My new b=
aby=20
daughter will inform me if I will have the time to learn how to diff this=
=20
stuff properly.

1.  Remove 'Apm' and 'Processor' entries from=20
<work.alpha>/kdebase-3.0.5a/ksysguard/ksysguardd/modules.h
2.  Remove entire 'configure' test for sizeof(size_t) =3D=3D sizeof(unsig=
ned int)=20
in kde[network3,pim3,utils3]  (see pkg/20090)
3.  'configure' (for all kde*) fails to find STRLCAT and STRLCAT_PROTO
=09add #define for both instances in each of <work.alpha>/kde[*]/config.h
4.  <work.alpha>/kdenetwork-3.0.5a/kmail/kmheaders.cpp:
=09add #define <sys/stat.h>
5.  <work.alpha>/kdeutils-3.0.5a/config.h needs work per #3. =20
=09Also in this config.h, if __osf__ is defined (probably from enabling O=
SF1=20
emulation in the kernel), <sys/mount.h> will include, which will cause sy=
ntax=20
errors during the compile.  Comment out the <sys/mount.h> inclusion in=20
config.h and add the statfs define from later in the __osf__ conditional.=
 =20
(or, uh, duh, simply #undef __osf__, I suppose)
6.  ktimer, the final program in kdeutils3, is a lost cause.  It refers t=
o a=20
prefwidgetbase.h file that only exists in some programmer's imaginary pla=
ce. =20
Remove it from the Makefile.  It is the only program in the KDE3 suite fo=
r=20
which I had no solution.

My GCC optimizations were:  -mcpu=3D21164a -Wa,-m21164a -mmemory-latency=3D=
3=20
-fsched-interblock -fsched-spec -fsched-spec-load -fbranch-count-reg -pip=
e

Thanks and good hunting,
Dave