Sat Jul 9 07:47:06 EDT 2005 rsc
    I applied a different set of changes.  I fixed -P.
    I changed the notefs process to postnote once the
    remote side closes /dev/cpunote, so the server
    need not write the "finished" note.  This seemed
    cleaner and it only requires running a new cpu client.
    
    I'm not convinced that the so-called grid home directory
    logic is correct or that it belongs in cpu.  I don't know
    what the right answer is yet.  I log into machines hwere
    I don't have a home directory and I'm pretty happy with
    the current behavior.  I don't want none's home directory
    bound on top of mine just because I don't have a profile.
    What if I'm logging in to run /sys/lib/newuser?!
    
    Instead of testing if(dirstat(home)==nil) you could just
    do if(access(home, AEXIST) >= 0), which doesn't leak.
    Instead of explicitly opening the notepg file, you can
    use postnote(PNGROUP, exportfspid, "kill").