Ooops...

[Andrew J Berkley <berkley@wam.umd.edu>]

I left some of my silly code in that patch -- as you
all probably noticed (not).

There are many deadlock conditions in the messmanager code *sigh*.

Locks are needed, but the global locks I put in are
obviously wrong (they were just to check and make
sure that DissolveGroup wasn't occuring while a
message send was in progress).  Since we are going
to replace the MessageManager anyway, I will put
off doing real locking.

Anyway, you can remove all the mlock->lock()/unlock() calls... I should
have that in the tree by the time you get this though :)

Also, I made test_net DissolveGroups when it is done with them. test_net 
should (hopefully) now run continuously without ever having problems (ok,
so it doesn't work great with NUMCONNS > 1).

Andrew