That was a fascinating post. I have added material and links at
http://mindprod.com/jgloss/thread.html#DEBUGGING
based on your post for future reference.

After reading that I began to wonder if we are all wet on the way
handle locks.  Perhaps the way they should work is you have to get all
your desired locks together and ask for them ALL at ONCE. Then the
locks themselves have a unique internal priority that define the order
you have to grab them.  Attempting to grab a set automatically
releases everything you already have.

That would introduce a basketful of other problems, but perhaps things
might work that way within lock groups.

Another idea is locks within groups have priority numbers. You can't
grab a lock within a group with priority higher than you have already,
at least in assertions on mode.

I have a list of the methods are supposed to be safe according to
notes from Sun I found in various places. See
http://mindprod.com/jgloss/threadsafe.html

It is not very long. The list becomes a sort of contract.  You don't
want people looking at code to find out what IS safe and then trusting
the code to stay that way forever.  You might do that, and hand the
list to Sun to see if they are willing to make a commitment to keeping
some of those methods safe.

That simplifies your task of classifying methods.