#15141 closed bug (fixed)

decideKindGeneralisationPlan is too complicated

Reported by: simonpj Owned by:
Priority: normal Milestone: 8.8.1
Component: Compiler Version: 8.2.2
Keywords: Cc:
Operating System: Unknown/Multiple Architecture: Unknown/Multiple
Type of failure: None/Unknown Test Case:
Blocked By: Blocking:
Related Tickets: Differential Rev(s): Phab:D4971
Wiki Page:

Description (last modified by simonpj)

Consider this program, which originally comes from #14846. I have decorated it with kind signatures so you can see what is going on

{-# LANGUAGE AllowAmbiguousTypes #-}
{-# LANGUAGE ConstraintKinds #-}
{-# LANGUAGE EmptyCase #-}
{-# LANGUAGE FlexibleInstances #-}
{-# LANGUAGE InstanceSigs #-}
{-# LANGUAGE MultiParamTypeClasses #-}
{-# LANGUAGE ScopedTypeVariables #-}
{-# LANGUAGE TypeInType #-}
module T14846 where

import Data.Kind
import Data.Proxy

-- Cat :: * -> *
type Cat ob = ob -> ob -> Type

-- Struct :: forall k. (k -> Constraint) -> *
data Struct :: (k -> Constraint) -> Type where
  S :: Proxy (a::k) -> Struct (cls::k -> Constraint)

-- Structured :: forall k
--               forall (a:k) (cls:k->Constraint) -> Struct k cls
type Structured a cls = (S ('Proxy :: Proxy a) :: Struct cls)

-- AStruct :: forall k (cls : k->Constraint). Struct cls -> Type
data AStruct :: Struct cls -> Type where
  AStruct :: cls a => AStruct (Structured a cls)

-- StructI :: forall k1 k2 (cls :: k2 -> Constraint).
--            k1 -> Struct k2 cls -> Constraint
class StructI xx (structured::Struct (cls :: k -> Constraint)) where
  struct :: AStruct structured

--  Hom :: forall k1 k2 (cls :: k2 -> Constraint).
--         Cat k1 -> Cat (Struct {k2} cls)
data Hom :: Cat k -> Cat (Struct cls) where

-- Category :: forall ob. Cat ob -> Constraint
class Category (cat::Cat ob) where
  i :: StructI xx a => ríki a a

instance forall (riki :: Cat kx)
                (cls  :: kk -> Constraint).
         Category riki => Category (Hom riki :: Cat (Struct cls)) where
  i :: forall xx a. StructI xx a => Hom riki a a
  i = case struct :: AStruct (Structured a cls) of

Focus on the final instance declaration. Do kind inference on the type signature for i. I think we should get

  i :: forall {k1} {k2} {cls1 :: k2 -> Constraint}
       forall (xx :: k1) (a :: Struct {k2} cls1).
       StructI {k1} {k2} xx a => Hom {kx} {k2} riki a a

Nothing about i's type signature constraints the cls to be the same as the in-scope cls. Yes riki is in scope, but its kind is unrelated.

But at present GHC does not kind-generalise i's type (because of kindGeneralisationPlan), so random things in the body of i may influence how its kinds get fixed. And if we changed the signature a bit so that it didn't mention riki any more, suddenly it would be kind-generalised.

This seems incomprehensibly complicated to me.

Moreover, the type of i from the class decl:

class Category (cat::Cat ob) where
  i :: StructI xx a => ríki a a

comes out in its full glory as

  i :: forall k1 k2 (cls :: k2 -> Constraint)
              (xx :: k1) (a :: Struct {k2} cls)
              (riki :: Cat (Struct {k2} cls).
       StructI {k1} {k2} cls xx a => riki a a

So indeed i is expected to be as polymorphic as I expected, and the unexpected lack of generalisation gives rise (in HEAD) to a bogus error:

T14846.hs:51:8: error:
    • Couldn't match type ‘ríki’ with ‘Hom kx k1 cls0 riki’
      ‘ríki’ is a rigid type variable bound by
        the type signature for:
          i :: forall k (cls1 :: k -> Constraint) k3 (xx :: k3) (a :: Struct
                                                                        k cls1) (ríki :: Struct
                                                                                           k cls1
                                                                                         -> Struct
                                                                                              k cls1
                                                                                         -> *).
               StructI k3 k cls1 xx a =>
               ríki a a
        at T14846.hs:51:8-51
      Expected type: ríki a a
        Actual type: Hom kx k1 cls0 riki a0 a0
    • When checking that instance signature for ‘i’
        is more general than its signature in the class
        Instance sig: forall (xx :: k0) (a :: Struct k1 cls0).
                      StructI k0 k1 cls0 xx a =>
                      Hom kx k1 cls0 riki a a
           Class sig: forall k1 (cls :: k1
                                        -> Constraint) k2 (xx :: k2) (a :: Struct
                                                                             k1 cls) (ríki :: Struct
                                                                                              -> Struct
                                                                                              -> *).
                      StructI k2 k1 cls xx a =>
                      ríki a a
      In the instance declaration for
        ‘Category {Struct kk cls} (Hom kx kk cls riki)’

Change History (5)

comment:1 Changed 19 months ago by simonpj

Description: modified (diff)

comment:2 Changed 18 months ago by bgamari


These won't be addressed for GHC 8.6.

comment:3 Changed 17 months ago by goldfire

Differential Rev(s): Phab:D4971
Status: newpatch

My solution to this is now on Phab. Awaiting review before committing.

comment:4 Changed 17 months ago by Richard Eisenberg <rae@…>

In c955a51/ghc:

Remove decideKindGeneralisationPlan

TypeInType came with a new function: decideKindGeneralisationPlan.
This type-level counterpart to the term-level decideGeneralisationPlan
chose whether or not a kind should be generalized. The thinking was
that if `let` should not be generalized, then kinds shouldn't either
(under the same circumstances around -XMonoLocalBinds).

However, this is too conservative -- the situation described in the
motivation for "let should be be generalized" does not occur in types.

This commit thus removes decideKindGeneralisationPlan, always

One consequence is that tc_hs_sig_type_and_gen no longer calls
solveEqualities, which reports all unsolved constraints, instead
relying on the solveLocalEqualities in tcImplicitTKBndrs. An effect
of this is that reporing kind errors gets delayed more frequently.
This seems to be a net benefit in error reporting; often, alongside
a kind error, the type error is now reported (and users might find
type errors easier to understand).

Some of these errors ended up at the top level, where it was
discovered that the GlobalRdrEnv containing the definitions in the
local module was not in the TcGblEnv, and thus errors were reported
with qualified names unnecessarily. This commit rejiggers some of
the logic around captureTopConstraints accordingly.

One error message (typecheck/should_fail/T1633)
is a regression, mentioning the name of a default method. However,
that problem is already reported as #10087, its solution is far from
clear, and so I'm not addressing it here.

This commit fixes #15141. As it's an internal refactor, there is
no concrete test case for it.

Along the way, we no longer need the hsib_closed field of
HsImplicitBndrs (it was used only in decideKindGeneralisationPlan)
and so it's been removed, simplifying the datatype structure.

Along the way, I removed code in the validity checker that looks
at coercions. This isn't related to this patch, really (though
it was, at one point), but it's an improvement, so I kept it.

This updates the haddock submodule.

comment:5 Changed 17 months ago by goldfire

Resolution: fixed
Status: patchclosed

All set now.

Note: See TracTickets for help on using tickets.