Re: [GHC] #15629: "No skolem info" panic (GHC 8.6 only)

11 Sep 2018

      #15629: "No skolem info" panic (GHC 8.6 only)
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler (Type    |              Version:  8.6.1-beta1
  checker)                           |
      Resolution:                    |             Keywords:  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 In a quick hack, I added a `-fno-eager-unifier` flag:

 {{{#!diff
 diff --git a/compiler/main/DynFlags.hs b/compiler/main/DynFlags.hs
 index 9f0ba57..a2f269d 100644
 --- a/compiler/main/DynFlags.hs
 +++ b/compiler/main/DynFlags.hs
 @@ -557,6 +557,7 @@ data GeneralFlag
     | Opt_OptimalApplicativeDo
     | Opt_VersionMacros
     | Opt_WholeArchiveHsLibs
 +   | Opt_NoEagerUnifier
     -- copy all libs into a single folder prior to linking binaries
     -- this should elivate the excessive command line limit restrictions
     -- on windows, by only requiring a single -L argument instead of
 @@ -3017,6 +3018,7 @@ dynamic_flags_deps = [
    , make_ord_flag defGhcFlag "dhex-word-literals"
          (NoArg (setGeneralFlag Opt_HexWordLiterals))

 +
    , make_ord_flag defGhcFlag "ghcversion-file"      (hasArg
 addGhcVersionFile)
    , make_ord_flag defGhcFlag "main-is"              (SepArg setMainIs)
    , make_ord_flag defGhcFlag "haddock"              (NoArg
 (setGeneralFlag Opt_Haddock))
 @@ -4003,7 +4005,8 @@ fFlagsDeps = [
    flagSpec "show-warning-groups"              Opt_ShowWarnGroups,
    flagSpec "hide-source-paths"                Opt_HideSourcePaths,
    flagSpec "show-loaded-modules"              Opt_ShowLoadedModules,
 -  flagSpec "whole-archive-hs-libs"            Opt_WholeArchiveHsLibs
 +  flagSpec "whole-archive-hs-libs"            Opt_WholeArchiveHsLibs,
 +  flagSpec "no-eager-unifier"                 Opt_NoEagerUnifier
    ]
    ++ fHoleFlags

 diff --git a/compiler/typecheck/TcUnify.hs b/compiler/typecheck/TcUnify.hs
 index 045132e..fb937b7 100644
 --- a/compiler/typecheck/TcUnify.hs
 +++ b/compiler/typecheck/TcUnify.hs
 @@ -1348,7 +1348,11 @@ uType_defer t_or_k origin ty1 ty2

  --------------
  uType t_or_k origin orig_ty1 orig_ty2
 -  = do { tclvl <- getTcLevel
 +  = do { no_eager <- goptM Opt_NoEagerUnifier
 +       ; if no_eager
 +         then uType_defer t_or_k origin orig_ty1 orig_ty2
 +         else do
 +       { tclvl <- getTcLevel
         ; traceTc "u_tys" $ vcat
                [ text "tclvl" <+> ppr tclvl
                , sep [ ppr orig_ty1, text "~", ppr orig_ty2]
 @@ -1357,7 +1361,7 @@ uType t_or_k origin orig_ty1 orig_ty2
         ; if isReflCo co
              then traceTc "u_tys yields no coercion" Outputable.empty
              else traceTc "u_tys yields coercion:" (ppr co)
 -       ; return co }
 +       ; return co } }
    where
      go :: TcType -> TcType -> TcM CoercionN
          -- The arguments to 'go' are always semantically identical
 }}}

 However, the panic does //not// reappear if you compile the original
 program with `-fno-eager-unifier`. The only differences I could observe
 were minor variations in error quality. With `-fno-eager-unifier` enabled,
 you don't see any of the "`‘z’ is a rigid type variable bound by`..."
 stuff, for instance.

-- 
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/15629#comment:5
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler