#8947: Depending on hint/ghc API fixes the binary version I can use ----------------------------+---------------------------------------------- Reporter: nh2 | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: GHC API | Version: 7.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: GHC rejects valid program Unknown/Multiple | Test Case: Difficulty: Unknown | Blocking: Blocked By: | Related Tickets: | ----------------------------+---------------------------------------------- When I `build-depend` on hint-0.4.0.0 with GHC 7.6.3, I get {{{ % cabal configure Resolving dependencies... cabal: Could not resolve dependencies: trying: pointcloudviewer-0.1.0 (user goal) trying: hint-0.4.0.0/installed-cf3... (dependency of pointcloudviewer-0.1.0) trying: ghc-7.6.3/installed-494... (dependency of hint-0.4.0.0/installed-cf3...) trying: bin-package-db-0.0.0.0/installed-608... (dependency of ghc-7.6.3/installed-494...) trying: GLUtil-0.7.4/installed-223... (dependency of pointcloudviewer-0.1.0) next goal: linear (dependency of GLUtil-0.7.4/installed-223...) rejecting: linear-1.6/installed-300... (conflict: bin-package-db => binary==0.5.1.1/installed-72e..., linear => binary==0.7.1.0/installed-a33...) }}} This fails because `linear` uses a `binary > 0.5.1.1`, but through my use of `hint` which depends on `ghc` (`hint` does not directly depend on `binary`), I pull in a constraint `binary == 0.5.1.1`. I believe this is bad, because depending on the GHC API in any way fixes the binary version I can use, which immediately makes a range of newer libraries unavailable to my program. Probably GHC should somehow hide the fact that binary is used inside, and let the rest of my program use whatever binary it likes. ---- I am aware that this might not be immediately easy, since the GHC API might expose some `binary` types to my program that I could try use with my newer binary version, which of course would be illegal. Maybe it is possible to implement such a check into actual typechecking, and let it pass if I don't actually use types from the old version with functions from the new library. I actually believe I have seen compiler errors in the past where two identical types where not unified because one was from a different version, so maybe infrastructure for this is already in place? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8947 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler