It is strange that adding -fallow-incoherent-instances should *stop* it compiling. This does not happen in the HEAD. I can see why it happens in 6.2 You are on very very very thin ice with this program. You have instance Confuse a => Typeable a f :: Typeable a => ... f x = ....stuff that needs Typeable .... At any time the compiler might use the instance decl to reduce the constraint Typeable a to Confuse a The only thing *stopping* it doing so is the desire to preserve coherence, which is why it does though before you add the flag. When you add the flag -fallow-incoherent-instances, the compiler feels free to reduce Typeable a to Confuse a... but then it finds that the type signature for f only provides Typeable. Sorry.. but if you have all of undecidable instances overlapping instances incoherent instances then life is pretty wild. Simon | -----Original Message----- | From: glasgow-haskell-users-bounces@haskell.org [mailto:glasgow-haskell-users- | bounces@haskell.org] On Behalf Of Christian Maeder | Sent: 30 November 2004 11:48 | To: GHC Users Mailing List | Subject: -fallow-incoherent-instances | | The attached module does not compile and yields the following error: | | InCoherentInst.hs:17: | Could not deduce (Confuse a) from the context (Typeable a) | arising from use of `breakFn' at InCoherentInst.hs:17 | Probable fix: | Add (Confuse a) to the type signature(s) for `addGeneralFallOut' | In the first argument of `GeneralBreakFn', namely `breakFn' | In the definition of `addGeneralFallOut': | addGeneralFallOut = let | breakFn a = throwDyn (GeneralFallOutExcep a) | in GeneralBreakFn breakFn | | | The same source compiles ok without -fallow-incoherent-instances (or | with -fno-allow-incoherent-instances). | | If, furthermore, the "confusing instance" is commented out, the source | even compiles without extensions. | | I don't know if this is a bug, possibly related to the import of | Typeable stuff. I don't need a fix. I only want to point out that | globally switching on the option -fallow-incoherent-instances is likely | to break existing code, currently (ghc 6.2.2). | | Cheers Christian