Type error suppresses defaulting, causes spurious errors

I have a strong feeling this must have been reported already but I can't find a ticket for it, so here goes.

If a source file contains a type error, like

bad :: Bool
bad = ()

then GHC seems to turn off defaulting for ambiguous types in the remainder of the file. So if the same file contains for instance

square :: Integral i => i -> i
square x = x^2

then GHC reports another type error at the ^2:

test.hs:4:7:
    Couldn't match expected type ‘Bool’ with actual type ‘()’
    In the expression: ()
    In an equation for ‘bad’: bad = ()

test.hs:7:13:
    Could not deduce (Integral b0) arising from a use of ‘^’
    from the context (Integral i)
      bound by the type signature for square :: Integral i => i -> i
      at test.hs:6:11-30
    The type variable ‘b0’ is ambiguous
    Note: there are several potential instances:
      instance Integral Int -- Defined in ‘GHC.Real’
      instance Integral Integer -- Defined in ‘GHC.Real’
      instance Integral GHC.Types.Word -- Defined in ‘GHC.Real’
    In the expression: x ^ 2
    In an equation for ‘square’: square x = x ^ 2

test.hs:7:14:
    Could not deduce (Num b0) arising from the literal ‘2’
    from the context (Integral i)
      bound by the type signature for square :: Integral i => i -> i
      at test.hs:6:11-30
    The type variable ‘b0’ is ambiguous
    Note: there are several potential instances:
      instance Num Double -- Defined in ‘GHC.Float’
      instance Num Float -- Defined in ‘GHC.Float’
      instance Integral a => Num (GHC.Real.Ratio a)
        -- Defined in ‘GHC.Real’
      ...plus three others
    In the second argument of ‘(^)’, namely ‘2’
    In the expression: x ^ 2
    In an equation for ‘square’: square x = x ^ 2

The first error is a legitimate type mismatch and it's all fine, but the other two should be resolved with default instance selection -- and it is, once the type error in bad is fixed. For example, rewriting bad to

bad :: ()
bad = ()

makes *all* errors go away, including the ones on square.

Default instance selection seems to be turned back on when square is made monomorphic, like

squareMono :: Int -> Int
squareMono x = x^2

Note that the type of x is being fixed, not the 2 that GHC complains about in square. The effect of this is that in a file with a large number of ^2, ^3, and such, the error message pointing to the real type error is drowned in all the "could not deduce" messages.

Trac metadata

Trac field	Value
Version	7.8.2
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture