figured out.

physical or virtual, as indicated by the isvirtnode flag.

Previously, any error in assign wrapper would cause the experiment to
swap out because the "DB had been modified" ... well I have isolated
all of the changes that are made, and errors in assign_wrapper proper
no longer do that. tbswap now restores the experiment back the way it
was. Not that errors after assign_wrapper (like in os_setup) are still
a problem.

In addition, rather then kill off all of the vlans, leave them in
place and then do a comparison after assign wrapper, removing obsolete
and modified vlans only. I have made use of the obsolete vlans table
for this by having snmpit track its changes in that table. There is a
bunch of new code in Lan.pm for doing the comparisons.

into Experiment.pm.
Also a few little protogeni additions that might get whacked later.

Add SelectOS() method to encapsulate the os_select goo.

lets checkpoint.

Also some changes to cross machine login to trac to avoid repeated
cookie generation.

This turned out to be Firefox doing the right thing, I guess. We had
used the same names for two sets of loading indicators and Firefox was
(properly?) clearing the first set but not the second. I think Safari
was doing the same thing, not that I think about it.

Note that remote sites will have the option of referencing Emulab
directly by using a redirect on their ops node.

added wiki as a cname to ops.

that fails!

experiments.

experiments. Running out of options letters!

only).  Also add a util script (meant to be used from a user's desktop)
that contacts the elab and plab xmlrpc servers for the config info, and
writes all that junk to the right places on a dongle or dongle img.

Add item for Frisbee ISO.

DAVE: see my note re the client gui version number.