Re: Java: Fix for PR 4695

[Bryce McKinlay <bryce at waitaki dot otago dot ac dot nz>]

Alexandre Petit-Bianco wrote:

>>So, I think we need a way to treat them as static but not actually
>>set the ACC_STATIC for the emitted class.
>>
>
>Yes. Is there a PR covering this type of illegal access non-detection?
>
I'll submit one.

>Otherwise, you're patch is good to go, along with the mod I suggested.
>

How about this instead, in create_anonymous_class? It fixes your failing 
test case and causes no regressions. It should enable us to check the 
access to enclosing instance with PURE_INNER_CLASS_DECL_P. As we 
discussed, it should really set another flag instead of ACC_STATIC to be 
correct according to JLS2, but I'll leave that for another patch. Note 
that GCJ doesn't seem to emit the ACC_STATIC bit for classes anyway, so 
this isn't a problem yet.

regards

Bryce.


@@ -3924,6 +3924,7 @@
   char buffer [80];
   tree super = NULL_TREE, itf = NULL_TREE;
   tree id, type_decl, class;
+  int flags;
 
   /* The unqualified name of the anonymous class. It's just a number. */
   sprintf (buffer, "%d", anonymous_class_counter++);
@@ -3947,8 +3948,12 @@
        super = type_name;
     }
 
-  class = create_class (ACC_FINAL, id, super, itf);
+  flags = ACC_FINAL;
 
+  if (ctxp->explicit_constructor_p)
+    flags |= ACC_STATIC; 
+ 
+  class = create_class (flags, id, super, itf);
   /* We didn't know anything about the stuff. We register a dependence. */
   if (!type_decl)
     register_incomplete_type (JDEP_ANONYMOUS, type_name, class, NULL_TREE);