Aaron Sherman wrote:
In reviewing S29 as it stands now, I see that many builtins both receive
and return boxed basic types. This seems like potentially spurious
overhead in some situations, while essential in others, so I wanted to
work out a set of rules for when boxed vs. unboxed types would be used
in core routines (given that all rules will have exceptions).
My thoughts on writing it were:
The boxed version is the specification, in that the language must support them. Think about using a SubType somewhere, and you see why. However, I also fully expected implementations to add an easy optimization of including unboxed equivalents and letting MMD sort it out.
-- Rod Adams
