#8707: Kind inference fails in data instance definition -------------------------------------+------------------------------------ Reporter: goldfire | Owner: jstolarek Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.7 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: -------------------------------------+------------------------------------ Comment (by goldfire): Quite right. But, your comment about the fact that this works for data instances gives me an idea: what about doing this unification in `kcDataDefn`? Note that the function is used only in the data instance case. (It might be worth renaming to make this now-critical fact more obvious.) There's no knot to worry about there. `kcConDecl` would be rewritten to return the desugared result type so that `kcDataDefn` can access it. Should be rather straightforward. If you agree that this might work, the question now is: is it better to have this (admittedly somewhat delicate) check in addition to the already- delicate handling of GADT return types? Or is it better to have a duplication of all the work in `tc_hs_type`? If we go with duplicating `tc_hs_type`, it's not immediately obvious exactly how the GADT-return- type typechecker would operate, given that the constraints on that type may appear arbitrarily deep in the type, depending on the data instance type patterns. I'm sure there's a way to get it to work, but I have a sneaking suspicion it will look a lot like a call to `unifyType` somewhere. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8707#comment:13 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler