bug in newwalk() in arla.c

[Alec Wolman]

I was testing out arla using arlad -t and quickly ran into a core dump.
Tracking it down in gdb, I found a bug in the way that newwalk handles symbolic
links (the symlink buffer gets trashed when you have a symlink followed
immediately by another symlink).  After fixing this bug, I discovered that it
still didn't work because arlad -t doesn't handle full path names properly.  In
particular, a symlink that I was traversing uses a full pathname rather than a
relative one, and arlad -t can't deal with the /afs component of the full
path-name.  

As far as I can tell, newwalk() is only used by the test code, so I'm
wondering whether or not its worth the hassle to fix the /afs stuff.
Is there any way to get the test code to use the normal code paths
for directory handling instead of the newwalk() routine?

Alec