I've been a bit upset by the challenges Template Haskell poses for type inference. For example,
(3 :: Int) == $$(...)
may typecheck when
$$(...) == (3 :: Int)
does not. I don't imagine this problem can be solved in general, but I'm rather curious whether it might be possible to solve for "pure" splices, with types that look like
forall m. Quote m => Code m a
Would it be possible to get full bidirectional inference for these if they were somehow marked specially by the user? For instance, if I wrote
$$$e
that could mean e should be interpreted as having some type
PCode a
where
newtype PCode a = PCode (forall m. Quote m => a)