Comments only

[ghc-hetmet.git] / compiler / typecheck / TcExpr.lhs

diff --git a/compiler/typecheck/TcExpr.lhs b/compiler/typecheck/TcExpr.lhs
index 784d466..93d3fe9 100644 (file)
--- a/compiler/typecheck/TcExpr.lhs
+++ b/compiler/typecheck/TcExpr.lhs
@@ -457,12 +457,22 @@ field isn't part of the existential. For example, this should be ok.
   data T a where { MkT { f1::a, f2::b->b } :: T a }
   f :: T a -> b -> T b
   f t b = t { f1=b }
+
 The criterion we use is this:
 
   The types of the updated fields
   mention only the universally-quantified type variables
   of the data constructor
 
+NB: this is not (quite) the same as being a "naughty" record selector
+(See Note [Naughty record selectors]) in TcTyClsDecls), at least 
+in the case of GADTs. Consider
+   data T a where { MkT :: { f :: a } :: T [a] }
+Then f is not "naughty" because it has a well-typed record selector.
+But we don't allow updates for 'f'.  (One could consider trying to
+allow this, but it makes my head hurt.  Badly.  And no one has asked
+for it.)
+
 In principle one could go further, and allow
   g :: T a -> T a
   g t = t { f2 = \x -> x }
@@ -529,7 +539,7 @@ tcExpr expr@(RecordUpd record_expr rbinds _ _ _) res_ty
              con1_flds = dataConFieldLabels con1
              con1_res_ty = mkFamilyTyConApp tycon (mkTyVarTys con1_tvs)
              
-       -- STEP 2
+       -- Step 2
        -- Check that at least one constructor has all the named fields
        -- i.e. has an empty set of bad fields returned by badFields
        ; checkTc (not (null relevant_cons)) (badFieldsUpd rbinds)