overlapping orphan instances behave like incoherent without warning/error
Consider these three modules:
module A where
class (Show a) => A a
data A' = A' deriving (Show)
instance A A'
data A'' = A'' deriving (Show)
instance A A''
print_a :: (A a) => a -> IO ()
print_a a = print a
{-# LANGUAGE FlexibleInstances, OverlappingInstances #-}
module B where
import A
data B a = B a deriving (Show)
instance (A a) => A (B a)
{-# LANGUAGE FlexibleInstances, OverlappingInstances #-}
module Main where
import A
import B
instance Show (B A') where
show _ = "kokodak"
instance Show (B A'') where
show _ = "brekeke"
instance A (B A'')
main :: IO ()
main = do
print (B A')
print_a (B A')
putStrLn ""
print (B A'')
print_a (B A'')
Without understanding a thing about dictionaries, I would expect that if this actually compiles (which I now understand it should not), I'd get "kokodak kokodak brekeke brekeke" as output, but I got "kokodak B A' brekeke brekeke" instead.
I figured that even though I redefined Show (B A'), the A (B a) instance was defined in module B and consisted of the original Show dictionary. If I move the Show (B A') instance to module B, ghc complains that the definition of A (B a) depends on the instatiation of a and refuses to compile it, unless I enable IncoherentInstances.
The problem here is that if the Show (B A') instance is orphan, I get the IncoherentInstances behaviour for free without any warning or error, giving me the false feeling that the code is actually OK. Is it possible that ghc gives an error in this case, and may the documentation mention that Overlapping + Orphan => Incoherent?
Trac metadata
| Trac field | Value |
|---|---|
| Version | 6.10.4 |
| Type | Bug |
| TypeOfFailure | OtherFailure |
| Priority | normal |
| Resolution | Unresolved |
| Component | Compiler (Type checker) |
| Test case | |
| Differential revisions | |
| BlockedBy | |
| Related | |
| Blocking | |
| CC | |
| Operating system | |
| Architecture |