Proper error message for missing constraints in FFI type variables

[tomshackell]

{-# LANGUAGE MagicHash #-}
module Main where

import Java

-- Map.Entry
data {-# CLASS "java.util.Map$Entry" #-} MapEntry k v =
  MapEntry (Object# (MapEntry k v))
  deriving Class

foreign import java unsafe getKey :: Java (MapEntry k v) k
foreign import java unsafe getValue :: Java (MapEntry k v) v

toTuple :: MapEntry k v -> (k, v)
toTuple e = pureJava $ do
    k <- e <.> getKey
    v <- e <.> getValue
    pure (k, v)

main :: IO ()
main = undefined

Eta version: 0.0.9b1

[tomshackell]

From investigating a bit further, I think this is about Eta not liking the fact that getKey and getValue are defined differently in Java.Collections. If I change the code to

{-# LANGUAGE MagicHash, FlexibleContexts #-}
module Main where

import Java

-- Map.Entry
data {-# CLASS "java.util.Map$Entry" #-} MapEntry k v =
  MapEntry (Object# (MapEntry k v))
  deriving Class

foreign import java unsafe "@interface getKey" getKey
  :: (Extends k Object, Extends v Object)
  => Java (MapEntry k v) k

foreign import java unsafe "@interface getValue" getValue
  :: (Extends k Object, Extends v Object)
  => Java (MapEntry k v) v

toTuple :: (Class k, Class v) => MapEntry k v -> (k, v)
toTuple e = pureJava $ do
    k <- e <.> getKey
    v <- e <.> getValue
    pure (k, v)

main :: IO ()
main = undefined

Then it compiles without a panic.

[rahulmutt]

This panic should be ideally converted to a proper error message. Essentially, if you are using any type variables in your import, you must specify a constraint on it. This constraint is essentially a proof that you can cast from a subclass to a parent class and required as the Eta compiler will do the same work that javac does when compiling generic classes. It's the foundation of how we embed subtyping into Eta just for the FFI but not elsewhere in the language.