diff options
Diffstat (limited to 'compiler/GHC/Tc/Utils')
| -rw-r--r-- | compiler/GHC/Tc/Utils/Env.hs | 3 | ||||
| -rw-r--r-- | compiler/GHC/Tc/Utils/Unify.hs | 42 |
2 files changed, 23 insertions, 22 deletions
diff --git a/compiler/GHC/Tc/Utils/Env.hs b/compiler/GHC/Tc/Utils/Env.hs index e8640a08dc..76ebb79e1e 100644 --- a/compiler/GHC/Tc/Utils/Env.hs +++ b/compiler/GHC/Tc/Utils/Env.hs @@ -628,7 +628,8 @@ tcExtendLocalTypeEnv lcl_env@(TcLclEnv { tcl_env = lcl_type_env }) tc_ty_things -- | @tcCheckUsage name mult thing_inside@ runs @thing_inside@, checks that the -- usage of @name@ is a submultiplicity of @mult@, and removes @name@ from the --- usage environment. See also Note [tcSubMult's wrapper] in TcUnify. +-- usage environment. See also Note [Wrapper returned from tcSubMult] in +-- GHC.Tc.Utils.Unify, which applies to the wrapper returned from this function. tcCheckUsage :: Name -> Mult -> TcM a -> TcM (a, HsWrapper) tcCheckUsage name id_mult thing_inside = do { (local_usage, result) <- tcCollectingUsage thing_inside diff --git a/compiler/GHC/Tc/Utils/Unify.hs b/compiler/GHC/Tc/Utils/Unify.hs index 145520045b..bf933127b8 100644 --- a/compiler/GHC/Tc/Utils/Unify.hs +++ b/compiler/GHC/Tc/Utils/Unify.hs @@ -762,29 +762,29 @@ to a UserTypeCtxt of GenSigCtxt. Why? ambiguity check, but we don't need one for each level within it, and GHC.Tc.Utils.Unify.alwaysBuildImplication checks the UserTypeCtxt. See Note [When to build an implication] --} +Note [Wrapper returned from tcSubMult] +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +There is no notion of multiplicity coercion in Core, therefore the wrapper +returned by tcSubMult (and derived functions such as tcCheckUsage and +checkManyPattern) is quite unlike any other wrapper: it checks whether the +coercion produced by the constraint solver is trivial, producing a type error +is it is not. This is implemented via the WpMultCoercion wrapper, as desugared +by GHC.HsToCore.Binds.dsHsWrapper, which does the reflexivity check. + +This wrapper needs to be placed in the term; otherwise, checking of the +eventual coercion won't be triggered during desugaring. But it can be put +anywhere, since it doesn't affect the desugared code. + +Why do we check this in the desugarer? It's a convenient place, since it's +right after all the constraints are solved. We need the constraints to be +solved to check whether they are trivial or not. Plus there is precedent for +type errors during desuraging (such as the levity polymorphism +restriction). An alternative would be to have a kind of constraint which can +only produce trivial evidence, then this check would happen in the constraint +solver. +-} --- Note [tcSubMult's wrapper] --- ~~~~~~~~~~~~~~~~~~~~~~~~~~ --- There is no notion of multiplicity coercion in Core, therefore the wrapper --- returned by tcSubMult (and derived function such as tcCheckUsage and --- checkManyPattern) is quite unlike any other wrapper: it checks whether the --- coercion produced by the constraint solver is trivial and disappears (it --- produces a type error is the constraint is not trivial). See [Checking --- multiplicity coercions] in TcEvidence. --- --- This wrapper need to be placed in the term, otherwise checking of the --- eventual coercion won't be triggered during desuraging. But it can be put --- anywhere, since it doesn't affect the desugared code. --- --- Why do we check this in the desugarer? It's a convenient place, since it's --- right after all the constraints are solved. We need the constraints to be --- solved to check whether they are trivial or not. Plus there are precedent for --- type errors during desuraging (such as the levity polymorphism --- restriction). An alternative would be to have a kind of constraints which can --- only produce trivial evidence, then this check would happen in the constraint --- solver. tcSubMult :: CtOrigin -> Mult -> Mult -> TcM HsWrapper tcSubMult origin w_actual w_expected | Just (w1, w2) <- isMultMul w_actual = |