rethink and rework subtype?

[hannesm]

connected to issue #37 it worries me a bit to have different implementation for subtype? at compile time and runtime, in addition to that there is a special subtype for type inference.

This needs some research, hopefully with a test library for types (specifically: test without optimizations, test with optimizations, test runtime and compile time type relationship). Somehow subtype? is also connected to loose mode (from dylan/union.dylan [method type-union]):
/* Illegal for loose Mode
subtype?(t1, t2)
=> t2;
subtype?(t2, t1)
=> t1;
*/
instead a new generic method, merge-type is introduced and used here.

[hannesm]

so, there's subtype? in library dylan, which defines the run-time subtype relation. then there is ^subtype? in dfmc-modeling, which contains the compile-time relation. and the type-estimate-subtype? relation in dfmc-typist (used for type inference).

these are inconsistent. take for example:

define method subtype? 
    (c :: <class>, subc :: <subclass>) => (result :: <boolean>)
  c == <class> & subc.subclass-class == <class>
end method;

and

define method ^subtype? 
    (t :: <&type>, subc :: <&subclass>) => (result :: <boolean>)
  ^subtype?(dylan-value(#"<object>"), subc.^subclass-class)
    & ^subtype?(t, dylan-value(#"<class>"))
end method;

of which I can't find a version in dfmc-typist....

Upcoming questions:

• how to define these in dfmc so they get emitted into the Dylan library directly, and we have a consistent relation?
• where is subclass defined and its typing rules?

[hannesm]

oh, and there is another -- subjunctive-subtype? in the dylan library.