GHC says there are unsafe overlapping instances, but only one instance applies

The following code rightfully does not compile:

import Text.Printf

f :: a -> b
f = undefined

g i = f $ printf "" i

However, the error message is a very surprising one:

test.hs:6:1:
    Unsafe overlapping instances for PrintfType (t -> a0)
      arising from the ambiguity check for `g'
    The matching instance is:
      instance [safe] (PrintfArg a, PrintfType r) => PrintfType (a -> r)
        -- Defined in `Text.Printf'
    It is compiled in a Safe module and as such can only
    overlap instances from the same module, however it
    overlaps the following instances from different modules:
    When checking that `g'
      has the inferred type `forall a b t. PrintfType (t -> a) => t -> b'
    Probable cause: the inferred type is ambiguous

What I find surprising is the claim that there are overlapping instances, followed by a list of instances that clearly does not have any overlaps. For comparison, changing the definition of g to be

g = f . printf ""

gives a much more reasonable error:

test.hs:6:9:
    No instance for (PrintfArg a0) arising from a use of `printf'
    The type variable `a0' is ambiguous
    Possible fix: add a type signature that fixes these type variable(s)
    Note: there are several potential instances:
      instance [safe] PrintfArg Char -- Defined in `Text.Printf'
      instance [safe] PrintfArg Double -- Defined in `Text.Printf'
      instance [safe] PrintfArg Float -- Defined in `Text.Printf'
      ...plus 12 others
    In the second argument of `(.)', namely `printf ""'
    In the expression: f . printf ""
    In an equation for `g': g = f . printf ""

Trac metadata

Trac field	Value
Version	7.6.1
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

closed

Thanks. Happily, with HEAD I get

T7857.hs:8:11:
    Could not deduce (PrintfType s0) arising from a use of ‛printf’
    from the context (PrintfArg t)
      bound by the inferred type of g :: PrintfArg t => t -> s
      at T7857.hs:8:1-21
    The type variable ‛s0’ is ambiguous
    Note: there are several potential instances:
      instance [safe] (PrintfArg a, PrintfType r) => PrintfType (a -> r)
        -- Defined in ‛Text.Printf’
      instance [safe] PrintfType (IO a) -- Defined in ‛Text.Printf’
      instance [safe] IsChar c => PrintfType [c]
        -- Defined in ‛Text.Printf’
    In the second argument of ‛($)’, namely ‛printf "" i’
    In the expression: f $ printf "" i
    In an equation for ‛g’: g i = f $ printf "" i

which seems very plausible.

So I propose to close this as fixed; I don't think it's worth messing with the 7.6 branch.

Simon

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

mentioned in commit af58cc6f

Trac metadata

Trac field	Value
Test case	→ typecheck/should_fail/T7857

added Pnormal label