Nov 28, 2008

we have basic Hindley-Milner with let-polymorphism.
we've tried to add in a simple polymorphic 'object' type, that is
  basically a record type since it supports no subtyping.
trying to infer the class of a method/field reference by introducing a new kind of 'type variable'.
The problem is that I'm not sure if it will work - i.e., if it's even
  possible.  Appel's MCIinML introduces a record type with globally
  unique field names, but mentions that ML solves the problem by using
  globally unique data constructors.  This is tantamount to doing
  something like this:

  "node.tree/right"

instead of

  "node.right"

to specify which class the 'right' attribute access belongs to.

It really seems like my method should work, but maybe I've gone about
it the wrong way?  Think I'm gonna remove the attr_type variable, and
see if I can get away with just doing some type_of rules...

================================================================================

Ok, just refreshed my brain on exactly how SML handles all of this.
It appears that the 'unique name' thing is true of data constructors
(redefinition shadows and hides any previous constructors by that
name), and that they route around the record-field issue by simply
requiring a type annotation.  I think this is do-able, and I can
certainly live with it in order to escape typing hell. 8^)

At some point in the future I may revisit this problem, it certainly
seems like it should be solvable!

================================================================================

Variant types.  I believe this is what a 'sum' type is.  Sum types
will require a run-time tag to distinguish which sub-type has been
chosen.  It'd be *really* cool if I could blend this with the existing
tag system, and then introduce a typecase or pattern match or
something...