Re: Ann: SWI-Prolog 7.1.0

Dear Jeff,

On Mon, 25 Nov 2013 13:13:35 +1100
Jeff Schultz <[email protected]> wrote:

> On 24/11/2013 03:49, Jan Wielemaker wrote:
> > On 11/22/2013 10:48 PM, Abramo Bagnara wrote:
> >> ?- A1 = a(), functor(A1, a, 0), functor(A2, a, 0), writeln(A1-A2), A1 = A2.
> >> a()-a
> >> false.
> >>
> >> This behavior is indeed very surprising for me... Is it deliberate?
> >
> > Yes.  I tried changing functor/3 (and =../2), but the compatibility
> > consequences are too involved and often hard to debug.  Therefore I left
> > functor/3 and =../2 alone and added compound_name_arity/3 and
> > compound_name_arguments/3.  You need these if you want to do
> 
> There appear to be four ways a() could be handled by functor/3 and the like:
> 
> (1)  As above.  functor(a(), a, 0) and functor(a, a, 0) succeed, and 
> functor(A, a, 0) succeeds once with A = a.
> 
> (2)  As (1), but functor(A, a, 0) succeeds twice with A = a or A = a().
> 
> (3)  functor(a(), F, N) fails.
> 
> (4)  functor(a(), F, N) raises a domain_error.
> 
> I don't think I like (1) at all.  Currently working code that copies 
> (and modifies) terms will now silently change a() into a in many places.
> 
> (2) introduces non-determinism where we've not had it before, and will 
> be painful to implement efficiently.  (Or at least, I will find it 
> difficult to implement efficiently.)  I think it would be a poor choice.
> 
> (3) will probably cause unnecessary surprise, and I don't like it either.
> 
> (4) is probably the safest.  Any code it breaks will likely be broken by 
> the new a() terms anyway, and the breakage will at least be obvious.
> 
> 
> I'm afraid that I don't see any use for it sufficient for such a large 
> change to the Prolog data model, and it will complicate a lot of my 
> code, but if we're to have it, could we try out (4) please?
> 
> 
>      Jeff Schultz


	There is at least one more way to proceed here, if i am not misinterpreting your alternatives.
	functor/3 could be made to fail/except on atoms. 
	It seems to me more natural for functor/3 to work on compounds.

	This would break some code, but it might regularise the language.
	I am not at all certain myself what is best here, but failing to see this alternative
	altogether does not feel right.
	It is likely that I am (doubly) biased here. I don't think I have much code that
	depends on functor( Atomic, Name, 0 )  and if i have some I would n't mind changing it. 
        Second bias is that code that is not maintained, should not dictate.
        I am probably missing something obvious, so I would appreciate some examples
	in which one would want to use functor/3 with atomic in its first argument.

	Certainly this is not a decision SWI should-have-to/can take on its own, maybe
	people that are involved in standarisation have a view on this and can move this forward.

Nicos Angelopoulos
---
http://stoics.org.uk/~nicos/