Poor error message with equational type constraints

GHCi 8.0.1 is giving a poor error message when it can't derive a typeclass when there's an equational type constraint involved. A simple example:

{-# LANGUAGE TypeFamilies #-}

class Foo a where
    type FooInner a
    fromInner :: FooInner a -> a

newtype Bar = Bar { fromBar::Char } deriving (Show)
instance Foo Bar where
    type FooInner Bar = Char
    fromInner = Bar

myFunc :: (Foo foo, FooInner foo ~ Char) => String -> foo
myFunc = fromInner . head

Many things work as expected:

ghc> :t myFunc
myFunc :: (FooInner foo ~ Char, Foo foo) => String -> foo
ghc> :t (myFunc "z")
(myFunc "z") :: (FooInner foo ~ Char, Foo foo) => foo
ghc> (myFunc "z") :: Bar
Bar {fromBar = 'z'}

but if I just evaluate the function without the typecast I get an error:

ghc> myFunc "z"
<interactive>:486:1: error:
    • Illegal equational constraint FooInner foo ~ Char
      (Use GADTs or TypeFamilies to permit this)
    • When checking the inferred type
        it :: forall foo. (FooInner foo ~ Char, Foo foo) => foo

Now, there should certainly be an error here: GHC doesn't know the exact type of myFunc so it can't check if it's an instance of Show. However, unless I'm not understanding what's going on, the error should be something like "Could not deduce Show", not "Illegal equational constraint". Even if that is what's going on, the suggestion to "Use GADTs or TypeFamilies to permit this" is clearly wrong, as I am already using TypeFamilies.

Trac metadata

Trac field	Value
Version	8.0.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tbug Trac import labels

Even if that is what's going on, the suggestion to "Use GADTs or TypeFamilies to permit this" is clearly wrong, as I am already using TypeFamilies.

Actually, it's technically not wrong: you don't have TypeFamilies turned on in GHCi which is where the problematic constraint arose.

But I certainly agree that GHC should just give the ambiguous type variable error in preference to the one it gave you, and probably shouldn't bother checking the inferred type of it at all.

Also note that there is another ticket somewhere about having flags set in the current module be in effect in ghci also, which would address this case too.

closed

If you turn on GADTs or TypeFamilies in GHCi, you'll get the error message you were hoping for.

*Test> :set -XGADTs
*Test> myFunc "z"

<interactive>:3:1: error:
    • Couldn't match expected type ‘Char’
                  with actual type ‘FooInner a0’
      The type variable ‘a0’ is ambiguous
    • In the first argument of ‘print’, namely ‘it’
      In a stmt of an interactive GHCi command: print it

I'm closing this ticket. I agree that the original error message is poor, but #11469 (the ticket referred to in ticket:12393#comment:122382) will fix the problem.

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedInvalid
Component	Compiler → GHCi
Related	- → #11469

added GHCi label

added Pnormal label