Revise tc103, in the light of the story for result type signatures

testsuite/tests/ghc-regress/typecheck/should_fail/tcfail103.hs

 {-# OPTIONS -fglasgow-exts #-}
 
--- !!! Scoped type variables in result signatures
--- This one is a bit crafty
-
 module ShouldCompile where
 import GHC.ST
 import GHC.STRef
 import GHC.Arr
 
--- Note the *pattern* type sig on f, which forces it
--- to be monomorphic; but the separate type sig makes
--- it polymorphic; hence the error.
+-- Another 'escape' example
+
 f:: ST t Int
-f:: ST s Int = do
+f = do
     v <- newSTRef 5
     let g :: ST s Int
-             -- ^ should be in scope
+	-- Implicitly forall a. ST s Int
         g = readSTRef v
     g

testsuite/tests/ghc-regress/typecheck/should_fail/tcfail103.stderr

-
-tcfail103.hs:15:0:
-    Inferred type is less polymorphic than expected
-      Quantified type variable `t' is mentioned in the environment:
-	Scoped type variable `s' = t (bound at: tcfail103.hs:15:7)
-    When trying to generalise the type inferred for `f'
-      Signature type:     forall t. ST t Int
-      Type to generalise: ST t Int
-    In the type signature for `f'
-    When generalising the type(s) for `f'
+
+tcfail103.hs:15:22:
+    Couldn't match expected type `s' (a rigid variable)
+	   against inferred type `t' (a rigid variable)
+      `s' is bound by the type signature for `g' at tcfail103.hs:13:16
+      `t' is bound by the type signature for `f' at tcfail103.hs:10:7
+      Expected type: STRef s Int
+      Inferred type: STRef t t1
+    In the first argument of `readSTRef', namely `v'
+    In the expression: readSTRef v