CVS difference for ais/ai-00344.txt

--- ais/ai-00344.txt	2004/12/09 19:55:31	1.14
+++ ais/ai-00344.txt	2005/01/07 03:07:41	1.15
@@ -3084,3 +3084,76 @@
 
 ****************************************************************
 
+From: Randy Brukardt
+Sent: Friday, December 9, 2004  7:25 PM
+
+In writing up the AARM notes for 3.9, I'm beginning to wonder if the definition
+of tags is quite right. The new text of 3.9(4) says:
+
+  The tag of a specific tagged type identifies the full_type_declaration of
+  the type, and for a type extension, is sufficient to uniquely identify
+  the type among all descendants of the same ancestor. If a declaration for
+  a tagged type occurs within a generic_package_declaration, then the
+  corresponding type declarations in distinct instances of the generic
+  package are associated with distinct tags. For a tagged type that is
+  local to a generic package body and with any ancestors also local to
+  the generic body , the language does not specify whether repeated
+  instantiations of the generic body result in distinct tags.
+
+We also have 3.9(26), which gives a permission to raise Tag_Error for tags that
+refer to types that don't exist in the permission. But there is no permission
+to give the wrong answer for a tag representing a type that no longer exists.
+That suggests that "=" could give the wrong answer based on this wording.
+
+Consider:
+
+    procedure Test is
+      T1, T2 : Ada.Tags.Tag;
+
+      function Get_It return Ada.Tags.Tag is
+          type Test_Type is new Ada.Finalization.Controlled with null record;
+      begin
+          return Test_Type'Tag;
+      end Get_It;
+    begin
+      T1 := Get_It;
+      T2 := Get_It;
+      if T1 = T2 then
+         Put_Line ("Violates 3.9(4)! Different type's tags are equal");
+      end if;
+    exception
+      when Tag_Error => null; -- OK, Tag error raised.
+    end Test;
+
+In a typical implementation, the tag would be augmented with a pointer to the
+stack frame of Get_It. And it is very likely that those frames would be the
+same in these two calls. Thus the tags would be identical. The implementation
+could hardly raise Tag_Error in that case, and thus it would get the wrong
+answer.
+
+The implementation could use a serial number to keep the tags different, but
+that would require three word tags (it couldn't be stored at the point of the
+type definition, or it would match in the example above).
+
+This certainly seems to be a pathology that we don't want to make work
+right. It seems to me that the wording of 3.9(4) is a bit too strong. We're
+really only concerned about types that actually exist. (We already gave up in
+streaming going further than that, it seems that should extend to all types.) I
+was going to add a To Be Honest note, but perhaps it needs more than that.
+
+Changing the wording to:
+
+  "...for a type extension, is sufficient to uniquely identify
+  the type among all descendants that currently exist of the same
+  ancestor. ..."
+
+would do the trick.
+
+Should this be a To Be Honest, or should we change the wording?
+
+[Private e-mail discussion decided to use a To Be Honest AARM Note
+rather than changing the wording. I've recorded my mail so that we have
+a record of the pathological example that motivated it. - RLB]
+
+****************************************************************
+