[C++ Patch, Java related/RFC] PR 11006

[Paolo Carlini]

Hi,

we have got this very old bug, where we ICE in build_java_class_ref 
because TYPE is an INTEGER_TYPE and we try to use TYPE_FIELDS on it. 
Shall we apply something like the below and resolve it for good? 
Alternately, we could maybe provide the same message we currently 
provide in release-builds, when we get to:

    if (!field)
      {
    error ("can%'t find %<class$%> in %qT", type);
    return error_mark_node;
      }

Thanks!
Paolo.

//////////////////////

Index: cp/init.c
===================================================================
--- cp/init.c   (revision 204353)
+++ cp/init.c   (working copy)
@@ -2461,9 +2461,16 @@ build_new_1 (vec<tree, va_gc> **placement, tree ty
   if (vec_safe_is_empty (*placement) && TYPE_FOR_JAVA (elt_type))
     {
       tree class_addr;
-      tree class_decl = build_java_class_ref (elt_type);
+      tree class_decl;
       static const char alloc_name[] = "_Jv_AllocObject";
 
+      if (!MAYBE_CLASS_TYPE_P (elt_type))
+       {
+         error ("%qT isn%'t a valid Java class type", elt_type);
+         return error_mark_node;
+       }
+
+      class_decl = build_java_class_ref (elt_type);
       if (class_decl == error_mark_node)
        return error_mark_node;
 
Index: testsuite/g++.dg/other/java3.C
===================================================================
--- testsuite/g++.dg/other/java3.C      (revision 0)
+++ testsuite/g++.dg/other/java3.C      (working copy)
@@ -0,0 +1,7 @@
+// PR c++/11006
+
+typedef int* jclass;
+
+void foo () {
+  new __java_boolean;  // { dg-error "valid" }
+}