[mercury-users] Pred defns

[Andrew Bromage]

G'day all.

Dominique de Waleffe wrote:

> Really, I want to :
> - use more than 1 accumulator,
> - escape from passing all of them around to all subgoals, 
> - be able to reference next and current state by name
> - have preference for local effects/transformations 

I think it would help everyone if we put our requirements on the table
rather than concrete proposals.  That way, we can design a system based
on what people want rather than (potentially) ugly syntax.

I want to:

	- Be able to pass variables "in", "out" and "through" a call,
	  the last of the three corresponding to an accumulator pair.

	- Be able to reference the next and current state of each variable
	  by name.

	- Require declarations (at the :- pred declaration) of what
	  named variables will be passed in, out and through.

	- Require declarations at each call of what named variables
	  will be passed in, out and through (i.e. no hidden arguments
	  because that was the problem with EDCGs).

	- Not have to pick an argument position in each call of where
	  the named variables "go". This is the major benefit of EDGCs.

BTW, I like the sound of "registers" on a module-by-module basis, though
not at the source level.  Perhaps a unique passed-in variable or a
destructively updated accumulator could be a hint to the compiler to
try inferring this.

Cheers,
Andrew Bromage