[GHC] #9190: Iface type variable out of scope: s

[GHC]

#9190: Iface type variable out of scope:  s
-------------------------------------+------------------------------------
        Reporter:  nomeata           |            Owner:
            Type:  bug               |           Status:  new
        Priority:  normal            |        Milestone:
       Component:  Compiler          |          Version:  7.9
      Resolution:                    |         Keywords:
Operating System:  Unknown/Multiple  |     Architecture:  Unknown/Multiple
 Type of failure:  None/Unknown      |       Difficulty:  Unknown
       Test Case:                    |       Blocked By:
        Blocking:                    |  Related Tickets:
-------------------------------------+------------------------------------

Old description:

> I just tried to install criterion with the current GHC HEAD, and
> `statistics` failed with this message:
>
> {{{
> /home/jojo/.cabal/lib/x86_64-linux-ghc-7.9.20140609/math-
> functions-0.1.5.2/Numeric/Sum.hi
> Declaration for R:MVectorsKBNSum:
>   Iface type variable out of scope:  s
> Cannot continue after interface file error
> }}}
>
> I tried to reproduce this with smaller code than `statistics`, but could
> not trigger it.
>
> The interface dump gives me:
> {{{
> 661d69e55c1491128ac1ed3c03f864e2
>   axiom TFCo:R:MVectorsKB2Sum::
>       Data.Vector.Unboxed.Base.MVector s0 Numeric.Sum.KB2Sum
>         = Numeric.Sum.R:MVectorsKB2Sum s0
> a69bb56c3760003bcb7d99cd40c3d843
>   axiom TFCo:R:MVectorsKBNSum::
>       Data.Vector.Unboxed.Base.MVector s0 Numeric.Sum.KBNSum
>         = Numeric.Sum.R:MVectorsKBNSum s0
> }}}
> – not sure if there is a `forall` missing or not. Note that these data
> types are generated using template haskell [http://hdiff.luite.com/cgit
> /math-
> functions/tree/Numeric/Sum.hs?id=06b09eb4f110f99fb51a9ec0ec598fba54ac0986#n123
> source].
>
> In order to reproduce this, run someting like
> `cabal install  --with-compiler=.../inplace/bin/ghc-stage2 --ghc-
> option=-XTypeFamilies .` in `statistics-0.11.0.3` with the attached patch
> applied.

New description:

 I just tried to install criterion with the current GHC HEAD, and
 `statistics` failed with this message:

 {{{
 /home/jojo/.cabal/lib/x86_64-linux-ghc-7.9.20140609/math-
 functions-0.1.5.2/Numeric/Sum.hi
 Declaration for R:MVectorsKBNSum:
   Iface type variable out of scope:  s
 Cannot continue after interface file error
 }}}

 I tried to reproduce this with smaller code than `statistics`, but could
 not trigger it.

 The interface dump gives me:
 {{{
 661d69e55c1491128ac1ed3c03f864e2
   axiom TFCo:R:MVectorsKB2Sum::
       Data.Vector.Unboxed.Base.MVector s0 Numeric.Sum.KB2Sum
         = Numeric.Sum.R:MVectorsKB2Sum s0
 a69bb56c3760003bcb7d99cd40c3d843
   axiom TFCo:R:MVectorsKBNSum::
       Data.Vector.Unboxed.Base.MVector s0 Numeric.Sum.KBNSum
         = Numeric.Sum.R:MVectorsKBNSum s0
 }}}
 – not sure if there is a `forall` missing or note. Note that these data
 types are generated using template haskell [http://hdiff.luite.com/cgit
 /math-
 functions/tree/Numeric/Sum.hs?id=06b09eb4f110f99fb51a9ec0ec598fba54ac0986#n123
 source].

 In order to reproduce this, run someting like
 `cabal install  --with-compiler=.../inplace/bin/ghc-stage2 --ghc-
 option=-XTypeFamilies .` in `statistics-0.11.0.3` with the attached patch
 applied.

--

Comment (by simonpj):

 I tried reproducing this, but the `cabal install` process fell over much
 earlier in `mwc-random`:
 {{{
 System\Random\MWC\Distributions.hs:83:5:
     Illegal equational constraint PrimState m ~ PrimState m
     (Use GADTs or TypeFamilies to permit this)
     When checking that `normalTail'
       has the inferred type `forall (m1 :: * -> *).
                              (PrimMonad m1, PrimState m1 ~ PrimState m) =>
                              Bool -> m1 Double'
     In an equation for `standard':
         standard gen
           = loop
           where ....
 }}}
 This is more fallout from #8883.  (The inferred type also looks ambiguous,
 but it isn't because the type variable `m` is free in the environment.)
 Very similar errors show up in `vector-algorithms`.

 Both are solved by `-XMonoLocalBinds`.  I'm not sure why this didn't
 happen to you.

 Simon

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9190#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler