CVS difference for ais/ai-00167.txt

--- ais/ai-00167.txt	2003/05/24 00:51:33	1.8
+++ ais/ai-00167.txt	2003/06/07 04:58:54	1.9
@@ -569,4 +569,60 @@
 
 ****************************************************************
 
+From: Pascal Leroy
+Date: Thursday, April 24, 2003  10:55 AM
+
+AI 167 singles out assignment_statements and object_declarations as
+constructs that can be used to store the value returned by
+Unchecked_Conversion before testing for validity.  As part of his
+editorial review, Kiyoshi commented:
+
+"Is the special case handling limited to 'an assignment_statement or an
+object declaration'?  For example, in AI-00287, similar special case
+consideration is given also to default_expression, allocator, and formal
+object declaration.  I agree that assignment_statement and object
+declaration cover most of useful cases, but like to make the situation
+clear."
+
+I think that parameter passing should be excluded here, or we are
+killing all optimizations.  So default_expressions for parameters ought
+to be excluded too.  However, I am not sure what the right answer is for
+the other constructs.  Surely at the Padua meeting we didn't look at all
+the constructs that involve an assignment operation, so I am wondering
+if we covered all the useful cases.
+
+****************************************************************
+
+From: Tucker Taft
+Date: Thursday, April 24, 2003  11:59 AM
+
+This seems like a case where we should err on the
+side of limitation, namely, only allow the minimal
+number of constructs we think are needed.
+Assignment statements and object declarations
+are sufficient.  No need to explore other obscure
+alternatives in my view.  But for the record...
+
+    generic formal -- bad idea for shared generics
+       (equivalent to subprogram parameter in that case)
+    parameter/default -- bad, since subprogram body doesn't
+        "see" the unchecked conversion
+    allocator -- obscure
+    aggregate -- obscure
+
+****************************************************************
+
+From: Tucker Taft
+Date: Thursday, April 24, 2003  12:29 PM
+
+I agree, no need to include other constructs.
+
+****************************************************************
+
+From: Robert Dewar
+Date: Friday, April 25, 2003  7:53 AM
+
+I agree with this limitation.
+
+****************************************************************