On Fri, Dec 21, 2012 at 10:12:38AM +0000, Simon Peyton-Jones wrote:
> I would not use -XMonoLocalBinds for all modules -- that will force you to do
> more work.
> Instead use it just for the offending Pol3_ module, via {-# LANGUAGE
> MonoLocalBinds #-}
>
> Or, probably better, give a type signature inside moduloBasisx, as suggested
> in the http://hackage.haskell.org/trac/ghc/ticket/4361. The signature is
> better because it also makes the program easier to understand!
>
> Simon
>
> [..]
> | > * Alterantively, give a type signature for x, thus (in this case)
> | >
> | > moduloBasisx p = let x :: ()
> | > x = upLinSolvRing p
> | > in ()
Now, I am trying ghc-7.6.1.20121207 built from source on Debian Linux.
-- docon.cabal -----------------------------------------------------
...
extensions:
TypeSynonymInstances UndecidableInstances FlexibleContexts
FlexibleInstances MultiParamTypeClasses OverlappingInstances
RecordWildCards NamedFieldPuns DoAndIfThenElse
.
ghc-options:
-fno-warn-overlapping-patterns -fwarn-unused-binds
-fwarn-unused-matches -fwarn-unused-imports
-XRankNTypes -- new
-O
---------------------------------------------------------------
`make build' fails at Pol3_.hs, at compiling this instance
-- Pol3_.hs ----------------------------------------------------
...
{-# LANGUAGE ScopedTypeVariables #-} -- (1) **
instance forall a. (LinSolvRing (Pol a), CommutativeRing a) => -- (2) **
LinSolvRing (UPol (Pol a))
where
-- gxBasis in P[y], P = a[x1..xn].
-- Map to a[y,x1..xn] apply gxBasis there, return to P:
gxBasis [] = ([], [])
gxBasis fs@(f:_) = (map back gs, mapmap back mt)
where
UPol _ p y dP = f
(o, n) = (pPPO p, genLength $ pVars p)
(toLex, fromLex) = (reordPol $ lexPPO n, reordPol o)
p' = (toLex p) `asTypeOf` p -- (3) **
dP' :: forall a. (LinSolvRing (Pol a), CommutativeRing a) => -- (4) **
Domains1 (Pol a)
dP' = upLinSolvRing p' Map.empty
-- p needs lexPPO reordering, then,
-- its domain bundle needs change too
s' = cToUPol y dP' p'
-- sample for P'[y], P' = a[x1..xn] with lexComp
toOverP' = ct s' . map (\ (a, j) -> (toLex a, j)) . upolMons
fromOverP' = ct f . map (\ (a, j) -> (fromLex a, j)) . upolMons
-- P[y] <--> P'[y]
back = fromOverP' . headVarPol dP
(gs, mt) = gxBasis $ map (fromHeadVarPol . toOverP') fs
-----------------------------------------------------------------
What is newly added:
1) -XRankNTypes to docon.cabal -- in order to allow `forall' to
`instance'.
2) {-# LANGUAGE ScopedTypeVariables #-} -- to support explicit
polymorphic type signatures in the instance implementation,
3) asTypeOf for p',
4) Explicit signature for dP'.
The report is similar as the old one:
------------------------------------------------
Pol3_.hs:328:25:
Could not deduce (a ~ a1)
from the context (CommutativeRing (UPol (Pol a)),
MulMonoid (UPol (Pol a)),
LinSolvRing (Pol a),
CommutativeRing a)
bound by the instance declaration at Pol3_.hs:(313,10)-(314,72)
or from (LinSolvRing (Pol a1), CommutativeRing a1)
bound by the type signature for
dP' :: (LinSolvRing (Pol a1), CommutativeRing a1) =>
Domains1 (Pol a1)
at Pol3_.hs:327:12-71
`a' is a rigid type variable bound by
the instance declaration at Pol3_.hs:313:17
`a1' is a rigid type variable bound by
the type signature for
dP' :: (LinSolvRing (Pol a1), CommutativeRing a1) =>
Domains1 (Pol a1)
at Pol3_.hs:327:12
Expected type: Pol a1
Actual type: Pol a
In the first argument of `upLinSolvRing', namely p'
In the expression: upLinSolvRing p' Map.empty
In an equation for dP': dP' = upLinSolvRing p' Map.empty
Pol3_.hs:331:20:
Could not deduce (EuclideanRing a) arising from a use of dP'
from the context (CommutativeRing (UPol (Pol a)),
MulMonoid (UPol (Pol a)),
LinSolvRing (Pol a),
CommutativeRing a)
bound by the instance declaration at Pol3_.hs:(313,10)-(314,72)
Possible fix:
add (EuclideanRing a) to the context of the instance declaration
In the second argument of `cToUPol', namely dP'
In the expression: cToUPol y dP' p'
In an equation for s': s' = cToUPol y dP' p'
-------------------------------------------------------------
ghc-7.4.1 compiles everything without additions.
Can you, please, advise about options in docon.cabal,
where to set `forall a', signatures, etc ?
Regards,
------
Sergei
_______________________________________________
Glasgow-haskell-bugs mailing list
Glasgow-haskell-bugs@haskell.org
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
