September 7th, 2008

beartato phd

(no subject)

More progress on implementation. Even with the concession that the user has to write like A ^ M for linear application, it's still not really possible to infer heavily higher-order types of simple eta-contracted arguments like as in

o : type.
c : ((o -o o) -o o) -> type.
d : c A.

which is really too bad. The type of the free variable A could be ((o -o o) -o o), but it could be other HLF things, too, or at least so says my current algorithm. Though I thought I had proved some kind of maximality theorem once...