[m-dev.] An optimization worth doing.

[Lee Naish]

In message <199702100516.QAA21926 at mundook.cs.mu.OZ.AU> I wrote:
>:- type my23tree1 --->
>        nil;
>        non_empty23t(my23tree1, data, maybe_empty23tree).
>:- type maybe_empty23tree --->
>	yes_its_empty;

Should be
	yes_its_empty_except_it_has_a_right_subtree_of_course(my23tree1);

>	no_its_not_empty(my23tree1, data, my23tree1).

As for whether its a Mercury defficiency or a Mercury implementation
defficiency, Fergus might be right but I suspect some changes to the
Mercury type system would be needed to get the additional
expressiveness.  Pointers at the implementation level tend to be
reflected by something at the language level (even if it is more tame).

Basically what we are talking about is "inlining" type definitions to
avoid pointers.  To me this seems more problematic than inlining
procedures but perhaps its because I'm more open to declarative views of
code than declarative views of data (ie, values rather than a data
structures).  I've certainly be viewing things from a very "data
structures" perspective today.

	lee