So change lit <$> safeRead n to (\x -> lit x) <$> safeRead n, for example.
Does anyone else find this highly non-obvious? SPJ gives an example https://github.com/ghc-proposals/ghc-proposals/pull/287#issuecomment-5436837... of a context that can indeed distinguish, semantically, a function and its eta expansion. But the fact that, for the benefit of soundness, there must exist contexts where one can be a well-typed program while the other is not, is still mysterious to me. Can a type system with polymorphism really not do better than this? The discussion thread has some more interesting posts, e.g. https://github.com/ghc-proposals/ghc-proposals/pull/287#issuecomment-5443773... https://github.com/ghc-proposals/ghc-proposals/pull/287#issuecomment-1120461... At least, thanks to Jaro, the error index now explains when you should eta-expand. https://errors.haskell.org/messages/GHC-83865/ Olaf