#8827: Inferring Safe mode with GeneralizedNewtypeDeriving is wrong -------------------------------------+------------------------------------ Reporter: goldfire | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.8.1 Component: Compiler | Version: 7.8.1-rc2 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: 8226, 8745 -------------------------------------+------------------------------------ Comment (by simonpj): `Coercible` is not a regular type class, and its "instances" are not subject to the usual rules of "always-exported". So you can 100% ignore the rules for normal instances. The question is "under what circumstances are the instances of `Coercible` visible?". Or, to make it sound less type-class-ish, you could ask "what rules are used to solve `Coercible` constraints?". Section 3 of the [http://research.microsoft.com/~simonpj/papers/ext-f paper] discusses this in some detail -- I hope everyone in this conversation has read it carefully. Specifically: * The "newtype-unwrapping" instance is available only if the newtype data constructor is visible. This is uncontroversial, just what Safe Haskell wants, and is clearly not what a "normal instance" is like. * The "lifting" instances are controlled by roles. And here there is room for debate about the proper defaults for those roles. For example, we could say that with Safe Haskell the default roles (ie if you don't give an explicit signature) of any data type declared in that module are nominal rather than representational. Simon -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8827#comment:15 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler