types

[Olin.Shivers@CENTRO.SOAR.CS.CMU.EDU]

   Date: Thu, 28 Apr 88 03:19:16 PDT
   From: gyro@kestrel.ARPA (Scott W. Layson)
   Besides, the failure-continuation approach would be more suitable for
   a (gasp! heresy) strongly-typed language.  I know T will never become
   one of THOSE, but I'm still curious what the impact would be like...

T might make a nice substrate to build such a system on top of, though.
Suppose you defined a T extension that had a type system like ML's or the
polymorphic lambda calculus.  It would be straightforward to implement a type
inferencer (for ML; inferencing polymorphic lambda calculus is probably
undecidable), and type checker.  This would be trivial to compile into T.

It really craps up the code if you have to do all the declarations with
Lisp syntax. But if  you used ML's type system, you wouldn't have to
put declarations in your code -- they could all be inferred. The full
blown polymorphic lambda calculus, though, is another matter.

Voila. Strictly typed T for those who feel that it's a good thing there
are laws requiring people to wear safety belts, without altering the
T base level at all.
	-Olin

P.s. Even better than the polymorphic lambda calculus type system is
     the one in FX -- then you can be a facist about side effects as 
     well.