Re: Possible bug in setarg/3

On 11/14/2013 10:44 AM, Michael Ben Yosef wrote:
> Hi,
>
> I believe I've discovered a bug in setarg/3, or at least a behaviour I
> really don't understand. The following session on Ubuntu 12.04 x86_64
> demonstrates a test case for it:
>
> ---BEGIN---
> $ cat bug.pl
> bug(A, b(A,C)) :-
>          arg(1, A, C),
>          setarg(1, A, x).
>
> bug_new(a(_)).
> $ swipl
> Welcome to SWI-Prolog (Multi-threaded, 64 bits, Version 6.5.2)
> Copyright (c) 1990-2013 University of Amsterdam, VU Amsterdam
> SWI-Prolog comes with ABSOLUTELY NO WARRANTY. This is free software,
> and you are welcome to redistribute it under certain conditions.
> Please visit http://www.swi-prolog.org for details.
>
> For help, use ?- help(Topic). or ?- apropos(Word).
>
> ?- [bug].
> % bug compiled 0.00 sec, 3 clauses
> true.
>
> ?- A = a(_), bug(A, B). % Behaves as expected:
>
> A = a(x),
> B = b(a(x), _G2326).
>
> ?- bug_new(A), bug(A, B). % Behaves unexpectedly:
>
> A = a(x),
> B = b(a(x), x).
>
> ?-
> ---END---
>
> Why is x unified with C in bug/2 in the second query (and not in the first one)?

That is a nice question :-)  It shows one of the many weird behaviours of
setarg/3.  The first arg/3 call unifies the variable in a(_) with the one in
b(_,_).  In Prolog, that means there will be a `reference pointer' from one
to the other.  The direction of this pointer (a->b or b->a) depends on the
age of the term.  So, if it is a->b and we set arg 1 of a to x, b remains
a variable.  If it is b->a and we set arg1 of to x, both have the value x.

I must admit I never realised this.  I guess it means you should typically
ensure that an arg on which you want to do setarg/3 should not be aliased to
anything else.  In other words, generally do not use variables as initial
value for args to use with setarg/3.
	
	Cheers --- Jan