Lifetime of temporaries in new expressions

[Martin von Loewis <martin at mira dot isdn dot cs dot tu-berlin dot de>]

g++ incorrectly inserts a CLEANUP_POINT_EXPR inside each `new'
invocation. As a result, destructors are called too early, as
shown in g++.martin/new1.C.

The correct code is already in init.c. Enabling it, I could not detect
any regression failure in the testsuite. There is some mentioning of
flow problems, but these appear to be fixed.

I propose to activate this code. If somebody can confirm that the
original problem is fixed, I also propose to activate it in the
release branch.

1998-10-20  Martin v. Löwis <loewis@informatik.hu-berlin.de>

        * init.c (build_new_1): Delay cleanup until end of full expression.

Index: init.c
===================================================================
RCS file: /egcs/carton/cvsfiles/egcs/gcc/cp/init.c,v
retrieving revision 1.69
diff -c -3 -p -r1.69 init.c
*** init.c	1998/10/16 03:37:40	1.69
--- init.c	1998/10/20 06:20:49
*************** build_new_1 (exp)
*** 2472,2480 ****
  
  	  if (cleanup)
  	    {
- #if 0
- 	      /* Disable this until flow is fixed so that it doesn't
- 		 think the initialization of sentry is a dead write.  */
  	      tree end, sentry, begin, buf, t = TREE_TYPE (rval);
  
  	      begin = get_target_expr (boolean_true_node);
--- 2472,2477 ----
*************** build_new_1 (exp)
*** 2497,2510 ****
  	      rval = build (COMPOUND_EXPR, t, begin,
  			    build (COMPOUND_EXPR, t, rval,
  				   build (COMPOUND_EXPR, t, end, buf)));
- #else
- 	      /* FIXME: this is a workaround for a crash due to overlapping
- 		 exception regions.  Cleanups shouldn't really happen here.  */
- 	      rval = build1 (CLEANUP_POINT_EXPR, TREE_TYPE (rval), rval);
- 
- 	      rval = build (TRY_CATCH_EXPR, TREE_TYPE (rval), rval, cleanup);
- 	      rval = build (COMPOUND_EXPR, TREE_TYPE (rval), alloc_expr, rval);
- #endif
  	    }
  	}
      }
--- 2494,2499 ----