#8827: Inferring Safe mode with GeneralizedNewtypeDeriving is wrong -------------------------------------+------------------------------------ Reporter: goldfire | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.10.1 Component: Compiler | Version: 7.9 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): OK, so we've essentially decided (for 7.8) * Keep the status quo (as presented in the paper) for 7.8 * But GND and `Data.Coerce` are both out of bounds for Safe Haskell That's ok with me. Subsequent to 7.8, the Davids (or other Safe Haskell folk) may want to propose a concrete plan for getting them back in. Or maybe it just doesn't matter because Safe Haskell clients don't care (enough) about GND or `Coercible`. FWIW I am keen to avoid any solution based on direct user control of `Coercible` instances. We have roles (we need them regardless to guarantee type-soundness); what we want can be expressed through roles or role signatures; adding ''another'' mechanism of control that does almost the same thing should be avoided if at all possible. Indeed, I don't think in terms of instance declarations for `Coercible` at all; instead it's just a matter of what rules are available for solving `Coercible` constraints. To take an analogy, implicit parameters are internally implemented as type-class constraints, with some special rules. They share some stuff in common with type class constraints, but are best thought of separately. Same with `Coercible`. Simon -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8827#comment:36 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler