Pereira@SRI-AI@sri-unix.UUCP (10/12/83)
If convenience for the programmer is what you are after, letting
assertz modify running procedures is in at least 50% of the cases
what you need. For example, to compute bottom-up the transitive
closure c(X,Y) of r(U,V), do
close :-
connected(X,Y),
connected(Y,Z),
add_c(Y,Z),
fail.
close.
connected(X,Y) :- r(X,Y).
connected(X,Y) :- c(X,Y).
add_c(X,Y) :-
not_subsumed(c(X,Y)),
assertz(c(X,Y)).
not_subsumed(P) :- "the usual stuff".
Programs of this kind can implement all sorts of bottom-up and mixed
initiative deductive procedures, such as those derived from CF parsing
algorithms like Cocke-Kasami-Young and Earley. What is interesting
about these applications is that the implementation code is not pure
logic, but they use the Prolog machinery to implement another correct
inference procedure, which can be abstracted as the computation of the
transitive closure of some relation (CKY has often been presented this
way).
So, either assertz is allowed to change running procedures, or
two brands of assertz are created to provide the two alternative
behaviors. In any case, trying to abstract frequently used
operations as Richard has done is a good thing.
-- Fernando Pereira