Sorry, maybe it I didn't made it clear enough.

Perhaps I'm misunderstanding something, but since one can infer types in GHCI, that implies one can infer types in the GHC API; since Hint wraps the GHC API, that implies one can infer types in Hint, doesn't it? And indeed, there are functions to infer the type of a String; iirc I've even used them myself in mueval.

That's true, inferring the type of a string is not the problem. However, I am not only interested in the top-level type, but also in the type of any arbitrary subexpression. Consider for example following recursive definition: foo [] = [] foo (x:xs) = (:)(x+1)(foo xs) Typing only the first rule one would says its type is '[a] -> [a]'. However, the second rule specialises the type to 'Num a => [a] -> [a]'. So, if I am interested in the type of the 'x' in the pattern of the second rule, I need to be aware of its binding on the right-hand side. Consequently the type of 'x' is 'Num a'. AFAIK, I need Hendley-Milner type inference for that (please correct me if I am wrong) and I was wondering if I really need to implement it by myself, when it is already somewhere inside GHC. Thanks, Martin