Some difficulties to keep track of

[aspiwack]

• In current GHC, constaint resolution is parallel, so that in

  f :: A =o X
  f = (methA, methA)

  both holes are solved independently (in fact, in this case, they are even the same hole!). But we really can't do that, because the use of the A dictionary in one precludes the use of the A dictionary in the other.

• The constraint solver needs to be careful not to duplicate linear constraints (that is, it's not only an issue of constraint generation). e.g.

  instance (A, A) =o B
  
  f :: A =o X
  f = methB -- should fail!

• We need to be able to typecheck all branches of a case/if with one occurrence of each linear constraint in each branch

  f :: A =o Bool -> X
  f b = if b then methA else methA -- ok!
  
  g :: A =o Bool -> X
  g b = if b then methA else noA -- should fail!

• g :: RW s => Array s -o Int
  g arr = do
     .. -- RW s constraint here
     () <- writeArray 0 0 arr
     .. -- RW s constraint might be the same or different
     () <- writeArray 0 0 arr -- - which RW dictionary?
     Return 0
  -- problem: changing =o to => causes an ambiguity error
  -- in functions, changing ->. to -> doesn’t cause an error