[mercury-users] mercury and determinism

[Fergus Henderson]

Henk wrote:

> Any particular reason for the following:
...
> q(X, Y):-
> 	( X = a -> Y=b(0);
> 	  X = b(N), N1 is N + 5, Y = b(N1)
> 	).
...
> little.m:007: In `q(in, out)':
> little.m:007:   error: determinism declaration not satisfied.
> little.m:007:   Declared `det', inferred `semidet'.
> little.m:021:   Unification of `X' and `b(N)' can fail.
...
> Because I'd like to reuse my code in Prolog, I prefer the definition of
> q above the one of p. There could be some serious efficiency-problems
> in some Prolog-implementations  with the first one.

There are a number of alternatives you can use.
Tom already mentioned one of them,

	q(a, b(0).
	q(b(N), b(N1)) :-
		N1 is N + 5.

but there are a couple of others.  
One of them is that you can just add a call to error/1:

	q(X, Y):-
		( X = a -> Y = b(0)
		; X = b(N) -> N1 is N + 5, Y = b(N1)
		; not_reached
		).

	:- pred not_reached is erroneous.
	not_reached :- error("reached not_reached!_).

This has the disadvantage that code will be a slightly less
efficient in Mercury.

Another alternative is to use a cut.

	q(X, Y):-
		( X = a, !, Y=b(0)
		; X = b(N), N1 is N + 5, Y = b(N1)
		).

Mercury doesn't have cuts, but to make it easier to migrate code from
Prolog, there's a predicate '!'/0 which does nothing. 

This technique also has a slight efficiency cost in Mercury.
However, that cost will be eliminated if you compile with intermodule
optimization, since the call to '!'/0 will then be inlined and
optimized away.  This is the technique that I would recommend
for situations where Tom's suggestion is cumbersome.

Thomas Conway wrote:

> I'll put it onto the todo list, but
> I don't expect anything to happen in the near future.

Then again, I think it's not very difficult to implement, so you might
be lucky.

We just need to think carefully about such things before changing the
language.

-- 
Fergus Henderson <fjh at cs.mu.oz.au>   |  "I have always known that the pursuit
WWW: <http://www.cs.mu.oz.au/~fjh>   |  of excellence is a lethal habit"
PGP: finger fjh at 128.250.37.3         |     -- the last words of T. S. Garp.