Typeable: Ensure that promoted data family instance tycons get bindings

This fixes #13915, where the promoted tycons belonging to data family
instances wouldn't get Typeable bindings, resulting in missing
declarations.

Test Plan: Validate with included testcases

Reviewers: austin, simonpj

Reviewed By: simonpj

Subscribers: simonpj, RyanGlScott, rwbarton, thomie

GHC Trac Issues: #13915

Differential Revision: https://phabricator.haskell.org/D3759

3 files changed, 13 insertions, 9 deletions

diff --git a/compiler/typecheck/TcEnv.hs b/compiler/typecheck/TcEnv.hs
index 935ad3dcb7..12f8a1df4f 100644
--- a/compiler/typecheck/TcEnv.hs
+++ b/compiler/typecheck/TcEnv.hs
@@ -625,15 +625,18 @@ Consider
   data S = MkS (Proxy 'MkT)
 
 Is it ok to use the promoted data family instance constructor 'MkT' in
-the data declaration for S?  No, we don't allow this. It *might* make
-sense, but at least it would mean that we'd have to interleave
-typechecking instances and data types, whereas at present we do data
-types *then* instances.
+the data declaration for S (where both declarations live in the same module)?
+No, we don't allow this. It *might* make sense, but at least it would mean that
+we'd have to interleave typechecking instances and data types, whereas at
+present we do data types *then* instances.
 
 So to check for this we put in the TcLclEnv a binding for all the family
 constructors, bound to AFamDataCon, so that if we trip over 'MkT' when
 type checking 'S' we'll produce a decent error message.
 
+Trac #12088 describes this limitation. Of course, when MkT and S live in
+different modules then all is well.
+
 Note [Don't promote pattern synonyms]
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 We never promote pattern synonyms.
diff --git a/compiler/typecheck/TcRnTypes.hs b/compiler/typecheck/TcRnTypes.hs
index 3992a7e662..6383b57c28 100644
--- a/compiler/typecheck/TcRnTypes.hs
+++ b/compiler/typecheck/TcRnTypes.hs
@@ -1077,7 +1077,8 @@ data PromotionErr
   | ClassPE          -- Ditto Class
 
   | FamDataConPE     -- Data constructor for a data family
-                     -- See Note [AFamDataCon: not promoting data family constructors] in TcRnDriver
+                     -- See Note [AFamDataCon: not promoting data family constructors]
+                     -- in TcEnv.
   | PatSynPE         -- Pattern synonyms
                      -- See Note [Don't promote pattern synonyms] in TcEnv
 
diff --git a/compiler/typecheck/TcTypeable.hs b/compiler/typecheck/TcTypeable.hs
index e7a427f6e1..2fcca7ffc2 100644
--- a/compiler/typecheck/TcTypeable.hs
+++ b/compiler/typecheck/TcTypeable.hs
@@ -170,7 +170,7 @@ mkTypeableBinds
       | tc `elem` [runtimeRepTyCon, vecCountTyCon, vecElemTyCon]
       = False
       | otherwise =
-          (not (isFamInstTyCon tc) && isAlgTyCon tc)
+          isAlgTyCon tc
        || isDataFamilyTyCon tc
        || isClassTyCon tc
 
@@ -243,12 +243,12 @@ todoForTyCons mod mod_id tycons = do
                             }
             | tc     <- tycons
             , tc'    <- tc : tyConATs tc
-              -- If the tycon itself isn't typeable then we needn't look
-              -- at its promoted datacons as their kinds aren't Typeable
-            , Just _ <- pure $ tyConRepName_maybe tc'
               -- We need type representations for any associated types
             , let promoted = map promoteDataCon (tyConDataCons tc')
             , tc''   <- tc' : promoted
+              -- Don't make bindings for data-family instance tycons.
+              -- Do, however, make them for their promoted datacon (see #13915).
+            , not $ isFamInstTyCon tc''
             , Just rep_name <- pure $ tyConRepName_maybe tc''
             , typeIsTypeable $ dropForAlls $ tyConKind tc''
             ]