updates on cvs, nfs

[Luka <luka@MIT.EDU>]

i just completed a full clean retest of what would be the new tree
if not for last night's weirdness.  the difference between now and
then is i did the whole thing on local disk this time rather than
building into nfs as last night.

last night's build started going bad in inexplicable ways as soon
as i finished the firewalk build and started the csb huge DB build.
symptoms were random keyboard/cursor grabs that wouldn't go away,
app lockup, window manager lockup, and finally a system hang that
required a reboot.  it sucked.  after the reboot firewalk on the
built wk db simply gave a blank view, and nonsensical hangs after
certain dialog box ui interactions.

today's build has no problems whatsoever.  build procedure was
identical except for the filesystem used to do it on.  intermediate
test of the fresh firewalk binary was fine, and final test of
firewalk on csb huge DB is fine.

i assume the firewalk and DB builds should be fully deterministic
in their outputs (correct in theory at least?).  so unless that
is not the case for unknown weird reason, it seems we have a fine
export/import product (which i will schedule to reproduce maybe
end of the week again), and a bad nfs bug.  some email with seth
indicates this may be a known nfs bug recurring, and i have a
test i can run to try triggering it in a controlled manner if
that is so.  till then, be reasonably distrustful of doing very
huge operations across nfs, and let me know if you discover any
clear failing cases yourself.

this is a pretty complete report of my logic, so if anyone sees
deductive errors, i'm open to criticism.

-pl

p.s. adel: cc'ing you on this in case you recognize the nfs
problem and have any input, and because you'll probably need to
hear about it soon anyway.  otherwise probably nothing for you to
do yet... best for me to find a cleaner characterization first.