#14131: Difference between newtype and newtype instance -------------------------------------+------------------------------------- Reporter: Iceland_jack | Owner: RyanGlScott Type: bug | Status: patch Priority: normal | Milestone: Component: Compiler | Version: 8.2.1 Resolution: | Keywords: TypeFamilies Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: #7938, #9574, | Differential Rev(s): Phab:D3872 #13985 | Wiki Page: | -------------------------------------+------------------------------------- Comment (by RyanGlScott): Replying to [comment:13 simonpj]:
Really? `reportFloatingKvs` seems to report variables mentioned in `typat_tvs`, which doesn't look like the ones on the RHS. I wasn't expecting a test in `tcFamTyPats`, but rather in `tcTyFamInstEqn`.
Ah, I think I managed to confuse you by naming that variable `typat_tvs`. Those are the type variables that are used in type patterns (visible and invisible) //or// the RHS (these are collected during renaming, see `rnFamInstDecl`). We discover in `tcFamTyPats` which variables are actually bound by type patterns, so if there are any used variables that aren't bound, then they're free-floating (and summarily rejected). I suppose I should rename `typat_tvs` to something like `fam_used_tvs` to avoid this confusion.
DO you have a test like the example in comment:8?
Yes, see `testsuite/tests/polykinds/T13985.hs`. In the particular case of type family instances, it tests this: {{{#!hs type instance T = Proxy (Nothing :: Maybe a) }}} Which is rejected for having a free-floating `a`. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14131#comment:14 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler