Arguments to constructor evaluated in wrong order.

Mon Jun 4 23:09:00 GMT 2001

Mark J. Roberts writes:
> I'm running a build of the 3.0 branch from a couple days
> ago. Methods don't have this problem, it's only constructors,
> afaict.

And you see the problem when the evaluation order of arguments
matters. I have a patch that doesn't trigger regression in all my
tests. I'm going to check this in the trunk and the 3.0 branch:
./A
2001年06月04日 Alexandre Petit-Bianco <apbianco@redhat.com>
	* expr.c (force_evaluation_order): Match wrapped ctor calls, locate
	arguments accordingly.
Index: expr.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/java/expr.c,v
retrieving revision 1.102.2.6
diff -u -p -r1.102.2.6 expr.c

--- expr.c 2001年05月16日 06:42:23 1.102.2.6
+++ expr.c 2001年06月04日 23:04:29
@@ -3268,16 +3268,31 @@ force_evaluation_order (node)
 if (TREE_SIDE_EFFECTS (TREE_OPERAND (node, 1)))
 TREE_OPERAND (node, 0) = save_expr (TREE_OPERAND (node, 0));
 }
- else if (TREE_CODE (node) == CALL_EXPR || TREE_CODE (node) == NEW_CLASS_EXPR)
+ else if (TREE_CODE (node) == CALL_EXPR
+ || TREE_CODE (node) == NEW_CLASS_EXPR
+ || (TREE_CODE (node) == COMPOUND_EXPR
+ && TREE_CODE (TREE_OPERAND (node, 0)) == CALL_EXPR
+ && TREE_CODE (TREE_OPERAND (node, 1)) == SAVE_EXPR)) 
 {
 tree arg, cmp;
 
 if (!TREE_OPERAND (node, 1))
 return node;
 
+ arg = node;
+
+ /* Position arg properly, account for wrapped around ctors. */
+ if (TREE_CODE (node) == COMPOUND_EXPR)
+ arg = TREE_OPERAND (node, 0);
+
+ arg = TREE_OPERAND (arg, 1);
+
+ /* Not having a list of argument here is an error. */ 
+ if (TREE_CODE (arg) != TREE_LIST)
+ abort ();
+
 /* This reverses the evaluation order. This is a desired effect. */
- for (cmp = NULL_TREE, arg = TREE_OPERAND (node, 1); 
- arg; arg = TREE_CHAIN (arg))
+ for (cmp = NULL_TREE; arg; arg = TREE_CHAIN (arg))
 {
 tree saved = save_expr (force_evaluation_order (TREE_VALUE (arg)));
 cmp = (cmp == NULL_TREE ? saved :
