Re: question on assert and retract--ground testing.

[spratt]

David,
You piqued my interest, so I went to  my benchmark source and looked at 
the 'ground' implementations. The implementation you provided is very 
near what appears to be the fastest implementation among the four I have.

Years ago (on Fri Feb 28 00:33:49 1992, to be precise), Richard A. 
O'Keefe posted an extensive note on 'ground' checking in the 
comp.lang.prolog newsgroup (as part of a discussion with David Bowen and 
Saumy K. Debray). I adopted RAOK's various forms of definition of ground 
as part of my 'benchmark' program.

Of these various forms (there are four, using univ and functor/arg in 
different ways), the fastest one for gprolog 1.2.13 on my G4 Mac is 
currently:

ground(Term) :-
        (   atomic(Term) -> true
        ;   nonvar(Term),
            Term =.. [_|Args],
            ground_args(Args)
        ).

ground_args([]).
ground_args([H|T]) :-
        ground(H),
        ground_args(T).

This one is two to three times faster than the functor/arg-based 
ground/1 programs. Go figure. (I'd expect functor/arg to be faster than 
univ.) As RAOK points out, it's important to measure actual performance, 
not just speculate. In the case of various ground/1 implementations, 
RAOK found that different implementations were the fastest on different 
systems.

This one differs from David's is_ground/1 in that:
is_ground/1 uses multiple clauses to test atom, number, and var, instead 
of a single clause using the builtin if-then-else in ground/1.
There are no uses of cut in ground/1 or ground_args/1, and particularly 
is_ground_list/1 seems to have a spurious final clause and thus needless 
uses of cut .

-lindsey