more on functional dependencies and overlapping instances

[Peter Thiemann]

Hi again,

as I said before, the fix was only partial.
I also had to modify instImprove in subst.c, but it is clearly a
hack because it assumes that more specific instances are given first,
so I'm not sending it along. 

[I hope this is the right forum for talking about this:]

I don't know about the current state of the implementation, but the
following looks like a viable approach:

* keep an "improved" tag with each predicate pi
* in instImprove, if pi.improved then terminate immediately with
  result FALSE
* check all instances for improvement candidates (but don't do
  anything, yet)
* if there is exactly one candidate, go for it: improve pi, set the
  flag, and exit
* if there is more than one candidate, find the most specific
* if pi MATCHES the most specific one, go for it: ...
* otherwise, no improvement possible, yet

Doing this properly would require digging much deeper in the code than 
I can afford to right now. I'd appreciate any help (or alpha release:-)
to solve this problem.

And a widely related question: how exactly does the overlapping
instances mechanism deal with specificity? I mean, what
happens if there are two instances which are unifiable, but there is a
third instance that covers the unifier? The feb2000 release definitely 
cannot handle this since instances are added and checked
incrementally.

-Peter