#14119: Refactor type patterns -------------------------------------+------------------------------------- Reporter: goldfire | Owner: goldfire Type: task | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: 12564, 13910, | 13938, 14038 Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by goldfire): * owner: (none) => goldfire Comment: After a chat at ICFP, Simon and I agreed that this was worthwhile, but in a very different way than originally proposed: create a new datatype `TypePat`, which would be a ''Core'' type pattern, used in, e.g., `FamInst`. Such a type would have no casts, no type families, and no foralls. Casts can be gotten rid of because the pure unifier (the client of type patterns) never uses them, unless they wrap a variable. And in that case, we can just use a variable with a new, corrected kind, and substitute the new one in for the old one (casting appropriately in the, e.g., type family equation's RHS). This transformation would happen after kind-checking and desugaring. Incidentally, it turns out that the underlying problem that this ticket was originally opened for was fixed by a small fix in the unifier. (The matching substitution must be applied to any cast found in the type pattern.) But that fix might not catch every possible case, while the proposal in this comment should. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14119#comment:9 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler