Re: "maplist" for DCG?

On 21/11/2013, at 3:41 AM, Alan Baljeu wrote:

> What you write seems complex.  I don't see what motivates having that 
> much code.  I found a predicate that is apt, but much shorter.  
> 
> for(P,L) --> for_(L,P).
> 
> for_([H|T],P) --> call(P,H),for_(T,P).
> for_([   ],_) --> { true }.
> 

What I wrote was precisely the same as this except for
using the phrase//(N+1) family instead of the call//(N+1) family,
for consistency.

	zic(X) --> [a], X, [b].

will use phrase//1 to interpret X.

The purpose of the two-stage process (first preprocess to produce
a rule body and then call phrase//1 on that) is to avoid the
inefficiency of repeatedly reprocessing the same non-terminals.

In fact the code you present is *precisely* the mapnt//2
that I showed you:

> 
> In the absence of the phrase//(N+1) family,
> 
> mapnt(NT, []) --> [].
> mapnt(NT, [A|As]) --> call(NT, A), mapnt(NT, As).
> 
> will probably do.

(Why would anybody write {true} instead of the normal
indication of an empty RHS, []?)

No, for/4 is NOT the usual name for this.
So far as I know, there _isn't_ a usual name.
And for/4 has fairly often been used for other things,
e.g.,

for(Variable, Lower_Bound, Upper_Bound, Goal) :-
    between(Lower_Bound, Upper_Bound, Variable),
    call(Goal),
    fail ; true.

call/N+1 and phrase//(N+1) are *not* the same thing in this
or any other context.  Contrast

    phrase(., p, [q], [p,q,r], [r])

which should succeed, with

    call(., p, [q], [p,q,r], [r])

which should raise an exception because there is no ./4 predicate.
In this *instance*, where the form that is constructed does not
include any of the special syntax of DCG rule bodies, the two are
equivalent.  But a general-purpose predicate should not presume
special restrictions.

As for the final paragraph, about computing
Term = (f(In1), f(In2), f(In3))
and then inserting that using phrase(Term),
that's precisely what the separate-preprocessing-step version
of mapnt//2 does.