[m-dev.] no tag types

[Peter Schachte]

On Fri, Nov 06, 1998 at 03:21:16PM +1100, Thomas Charles CONWAY wrote:
> 
> Maybe a useful thing would be to make:
> 	:- type foo = int.
> be equivalent to
> 	:- type foo ---> foo(int).

This doesn't seem like a good idea to me because:

	(1) It looks /much/ too much like 

		:- type foo == int.

	    It's already natural and tempting to use `=' for
	    equivalence types.  This would make that mistake more
	    confusing. 

	(2) The syntax doesn't naturally suggest that f/1 is a
	    constructor for type foo.

I think a nicer solution is to put a section in the user manual about
no-tag types so users understand them, and put in the reference manual
a statement that Mercury implementations are required to implement
one-constructor, one-argument types without (additional) tags.  This
latter bit is tricky to specify without tying implementors hands, but
since Mercury requires full compile-time type analysis, we're just
requiring the implementation to implement constructions and
deconstructions of this type as casts.  Relatively speaking, it's not
a lot to demand.

I can think of much more useful extensions to the type system.  For
example, I'd like to be allowed to specify for a type what predicate
to call if a term of that type is used as a goal, possibly with extra
arguments, and what function to call if a term of that type is applied
to some arguments.  E.g., Wouldn't it be nice to be able to use a map
or an array as a function?  This would also let you use a
Boolean-valued function as a predicate.


-- 
Peter Schachte                | The greater part of our happiness depends on
mailto:pets at cs.mu.OZ.AU       | our dispositions and not on our
http://www.cs.mu.oz.au/~pets/ | circumstances.
PGP: finger pets at 128.250.37.3 |     -- Martha Washington