FunDeps

[John Meacham]

I was curious if making the restriction that fundeps be acyclic would
cause them to always be expressible by assosiated types (but obviously
not as powerful). my thinking being that every parameter at the
origination of a fundep would stay a class parameter and any parameter
at the target would become an assosiated type.

this rule would be powerful enough to express all of the MTL, and if it
is subsumed by assosiated types, it might be an acceptable compromise.

however, specifying the ambiguity rules is going to be interesting, but
I think we can do so by taking the assosiated types typing rules and
intersecting them with the fundeps rules treating fundep targets as
assosiated types. hopefully the result will be powerful enough and well
specified..

this is just sort of a brainstorming idea and not entirely thought out.

        John

-- 
John Meacham - ⑆repetae.net⑆john⑈