#11348: Local open type families instances ignored during type checking -------------------------------------+------------------------------------- Reporter: alexvieth | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 8.0.1-rc1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: GHC rejects | Unknown/Multiple valid program | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by alexvieth):
Yes, exactly. That is much simpler than "meta-nodes", and had the great merit of being correct (see third bullet).
Agreed. My graph approach was all wrong, so I've ditched it. I think it would be ideal if we didn't have to plumb the `FreeVars` past `rnSrcDecls` and onto `tcTyAndClassDecls`, so I've come up with a solution which does essentially what you've described but during dependency analysis. 1. Compute the SCCs for `TyClDecl`s as we do now in HEAD. 2. For all of the `InstDecl`s (I includes class and data family instances. Any harm in doing so?), associate it with its `FreeVars`, intersected with the set of `Name`s of `TyClDecl`s that we just analysed. 3. Fold the list of SCCs, at each step extracting the set of `InstDecl`s for which its `FreeVars` is empty, and then eliminating all of the `Name`s found in that SCC. That set of `InstDecl`s, if non-empty, comes before the current component in the output list. Will upload the patch in a few seconds. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11348#comment:8 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler