think about how how unboxed tuples interact with rest of planned type/kind/sort system and function types

[cartazio]

https://ghc.haskell.org/trac/ghc/ticket/11723#comment:3 ghc ran into this recently, so its not totally untrodden territory :)

will require some thought, but I do think that It'd be reasonable to not allow variable unboxed tupleness in dependent functions.

A related question is whether unboxed tuples in result (positive) position correspond with an existential/dependent record telescope just as unboxed tuples in arg (negative) position form a dependent universal/pi telescope. On this latter front, thinking about the CPS translation of function application makes me think the answer is "yes, duh", but worth working out. This also does suggest that CPS lowering (perhaps using the ideas of nick rioux, amal ahmed and will bowman as sketched out in http://www.cs.ox.ac.uk/ralf.hinze/WG2.8/33/slides/Amal.pdf might be useful )