#14917: Allow levity polymorhism in binding position -------------------------------------+------------------------------------- Reporter: andrewthad | Owner: (none) Type: feature request | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.2 Resolution: | Keywords: | LevityPolymorphism Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by goldfire): I actually think this is a good idea. Simon's right in that figuring out exactly when you can inline can be tricky. But this logic is already in the compiler (in the inliner!) and so we can perhaps work it into the desugarer (which is where levity- polymorphism errors are issued). It's possible we'll have a hard time producing sensible error messages, but I think we'll be able to surmount that challenge. As to the proof: I'm not concerned. The proof is about Core. The proposed change wouldn't affect Core at all. Core still wouldn't have levity- polymorphic binders. (Unfoldings might, but not actual Core programs that will be compiled.) We can think of this proposal as suggesting "function templates", where these templates are stand-ins for a (perhaps infinite) family of levity-monomorphic functions. The implementation of this would be fiddly, but I don't see any true obstacles to it. And it does seem very silly that users can't reuse their typing for a function as simple as `twice`. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14917#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler