Re: Ann: SWI-Prolog 7.1.0

On 28/11/2013, at 1:53 AM, Jan Wielemaker wrote:

> On 11/27/2013 06:25 AM, Richard A. O'Keefe wrote:
> 
> You are just moving the additional tests elsewhere: into functor/3.

No, I'm not "moving" anything anywhere.
The tests have to be in functor/3 ANYWAY.

We used to have such a simple classification scheme:

A Prolog term X is
  - a variable, in which case var(X) succeeds, or
  - not a variable, in which case nonvar(X) succeeds.
    If nonvar(X) succeeds, then functor(X, F, N) *ALWAYS*
    has a solution, where F is a constant and N is a non-negative integer.
    N may be
    - 0, in which case atomic(X) succeeds and F == X.
      This may be further classified as
      - atom(X), an atom
      - number(X), which breaks down into at least two kinds
      - string(X), a string, in some systems
      - some other kind of atomic data, such as (in Interlisp-D)
        references to GUI widgets, Lisp arrays, &c
    - > 0, in which case compound(X) succeeds and atom(F) succeeds.
    
So meta-logical code had the shape

p(X, ...) :-
    (   var(X) ->
    ;   functor(X, F, N),
        do something with F
        do something with arg(I, X, A) for I in 1..N
    ).

That's it.  functor/3 was and remains a *core* metalogical
operation, and functor(X, X, 0) succeeding for *all* atomic(X)
was and remains a *key* non-accidental not-just-a-relic
*highly prized* property that drastically simplifies much
meta-logical code.

I no longer recall if there was a time when functor(42, _, _)
didn't succeed.  I do recall that when I discovered at
Edinburgh that it _did_ succeed I found that it simplified
my meta-logical code so much that it just wasn't funny.

> 
> Changing these relics breaks too much (I tried).

That's nature trying to tell you that making f() a new kind of
term was a bad idea.

Hands up everyone who has written a predicate to find the
least general generalisation of two terms (assuming no cycles)?
It can't just be me.

Guess what predicate that depends on?