Re: [PATCH] Disallow empty "or" type specifier

[felix . winkelmann]

> It certainly shouldn't be *, nor should it be an error... technically an 
> empty union should be a null set, which would correspond to either a 
> non-extant type or to no return/value at all...

Well, the problem is: there is no non-extant type, no "bottom": we simply use
the Scheme type system (slightly extended) that clearly describes what types
values can have and in Scheme every value has a type, there are no programs
that are unsound on the type level - they may break at run-time, but are still
valid Scheme code. There are meta-types like "undefined", which is the same
as the union of all possible types, but explicitly declared as undefined by
the standard. CHICKEN uses other meta types like "*" for convenience, and
"or" and "forall" for limiting the possible set of types a value can have, but
there are still very much concrete types for concrete values.

The number of return values on the other hand is not a type-related concept, it 
is 
an operational thing (the number of arguments to pass to the continuation), so 
I 
would not mix these two issues.

There can be no "(or)"-typed value in a Scheme system, there can be the absence 
of
a return (like calling a continuation), but that is, I believe, the area of
effect systems.


felix