summaryrefslogtreecommitdiff
path: root/compiler/GHC/Core
diff options
context:
space:
mode:
authorSimon Peyton Jones <simon.peytonjones@gmail.com>2023-01-28 22:12:35 +0000
committerMarge Bot <ben+marge-bot@smart-cactus.org>2023-02-01 12:31:23 -0500
commit955a99ea28a0d06de67f0595d366450281aab0c0 (patch)
tree43ee724bf62c7e57403c8e60ab724c7758233afa /compiler/GHC/Core
parent9f95db54e38b21782d058043abe42fd77abfb9ad (diff)
downloadhaskell-955a99ea28a0d06de67f0595d366450281aab0c0.tar.gz
Treat existentials correctly in dubiousDataConInstArgTys
Consider (#22849) data T a where MkT :: forall k (t::k->*) (ix::k). t ix -> T @k a Then dubiousDataConInstArgTys MkT [Type, Foo] should return [Foo (ix::Type)] NOT [Foo (ix::k)] A bit of an obscure case, but it's an outright bug, and the fix is easy.
Diffstat (limited to 'compiler/GHC/Core')
-rw-r--r--compiler/GHC/Core/Opt/WorkWrap/Utils.hs25
1 files changed, 17 insertions, 8 deletions
diff --git a/compiler/GHC/Core/Opt/WorkWrap/Utils.hs b/compiler/GHC/Core/Opt/WorkWrap/Utils.hs
index f599975355..488eeaf5f4 100644
--- a/compiler/GHC/Core/Opt/WorkWrap/Utils.hs
+++ b/compiler/GHC/Core/Opt/WorkWrap/Utils.hs
@@ -707,7 +707,7 @@ Worker/wrapper will unbox
* is an algebraic data type (not a newtype)
* is not recursive (as per 'isRecDataCon')
* has a single constructor (thus is a "product")
- * that may bind existentials
+ * that may bind existentials (#18982)
We can transform
> data D a = forall b. D a b
> f (D @ex a b) = e
@@ -1272,16 +1272,25 @@ also unbox its components. That is governed by the `usefulSplit` mechanism.
-}
-- | Exactly 'dataConInstArgTys', but lacks the (ASSERT'ed) precondition that
--- the 'DataCon' may not have existentials. The lack of cloning the existentials
--- compared to 'dataConInstExAndArgVars' makes this function \"dubious\";
--- only use it where type variables aren't substituted for!
+-- the 'DataCon' may not have existentials. The lack of cloning the
+-- existentials this function \"dubious\"; only use it where type variables
+-- aren't substituted for! Why may the data con bind existentials?
+-- See Note [Which types are unboxed?]
dubiousDataConInstArgTys :: DataCon -> [Type] -> [Type]
dubiousDataConInstArgTys dc tc_args = arg_tys
where
- univ_tvs = dataConUnivTyVars dc
- ex_tvs = dataConExTyCoVars dc
- subst = extendSubstInScopeList (zipTvSubst univ_tvs tc_args) ex_tvs
- arg_tys = map (GHC.Core.Type.substTy subst . scaledThing) (dataConRepArgTys dc)
+ univ_tvs = dataConUnivTyVars dc
+ ex_tvs = dataConExTyCoVars dc
+ univ_subst = zipTvSubst univ_tvs tc_args
+ (full_subst, _) = substTyVarBndrs univ_subst ex_tvs
+ arg_tys = map (substTy full_subst . scaledThing) $
+ dataConRepArgTys dc
+ -- NB: use substTyVarBndrs on ex_tvs to ensure that we
+ -- substitute in their kinds. For example (#22849)
+ -- Consider data T a where
+ -- MkT :: forall k (t::k->*) (ix::k). t ix -> T @k a
+ -- Then dubiousDataConInstArgTys MkT [Type, Foo] should return
+ -- [Foo (ix::Type)], not [Foo (ix::k)]!
findTypeShape :: FamInstEnvs -> Type -> TypeShape
-- Uncover the arrow and product shape of a type