Add comments to a very tricky test

testsuite/tests/ghc-regress/typecheck/should_compile/tc163.hs

@@ -7,16 +7,31 @@
 
 module ShouldCompile where
 
-  type M3 a = forall r. (forall b. M3' b -> (b -> M3' a) -> r) -> r
+type M3 a = forall r. (forall b. M3' b -> (b -> M3' a) -> r) -> r
 
-  newtype M3' a
-      = M3' { mkM3' :: M3 a }
+newtype M3' a = M3' { mkM3' :: M3 a }
 
-  bind :: M3 a -> (a -> M3 b) -> M3 b
-  bind m k b = b (M3' m) (\a -> M3' (k a))
+flop :: forall a b. M3' b -> (b -> M3' a) -> Int
+flop = \m' k -> mkM3' m' (\bm k1 -> error "urk")
 
-  observe :: M3 a -> a
-  observe m
+-- Suppose mkM3' has the straightforward type: 
+--     mkM3' :: forall a. M3' a -> M3 a
+-- Then (mkM3' m') :: forall r. (forall b. ...) -> r
+-- If we simply do a subsumption check of this against
+--    alpha -> Int
+-- where alpha is the type inferred for (\bm k1 ...) 
+-- this won't work.
+
+-- But if we give mkM3' the type 
+--     forall a r. M3' a -> (forall b. ...) -> r
+-- everthing works fine.  Very very delicate.
+
+---------------- A more complex case -------------
+bind :: M3 a -> (a -> M3 b) -> M3 b
+bind m k b = b (M3' m) (\a -> M3' (k a))
+
+observe :: M3 a -> a
+observe m
       = m (\m' k -> mkM3' m'
                 (\bm k1 -> observe (bind (mkM3' bm)
                         (\a -> bind (mkM3' (k1 a)) (\a -> mkM3' (k a)))))