(C++) tree inlining patch RFD

[Jason Merrill]

This patch fixes block structure problems with g++.other/inline3.C.  We
were expanding inline the call to operator[] before we saw the SCOPE_STMT
for the for-scope; the for block then ended up nested inside the block for
the inlined function.

I haven't checked this in because I don't really see this as a solution;
the problem is that the way we're handling blocks is extremely fragile.
I'm running into a similar problem in libstdc++/tests/tlist.cc where the
block for a cleanup for a TARGET_EXPR is inserted before the block where
the temporary in question appears.  Mark, do you have any thoughts?

1999-12-15  Jason Merrill  <jason@casey.cygnus.com>

	* tree.c (walk_tree): Walk operand subtrees in forward order.

Index: tree.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/cp/tree.c,v
retrieving revision 1.168
diff -c -p -r1.168 tree.c
*** tree.c	1999/12/15 09:51:24	1.168
--- tree.c	1999/12/15 19:35:22
*************** walk_tree (tp, func, data)
*** 1623,1643 ****
        || TREE_CODE_CLASS (code) == 'r'
        || TREE_CODE_CLASS (code) == 's')
      {
!       int i;
  
        /* Walk over all the sub-trees of this operand.  */
!       i = first_rtl_op (code) - 1;
        /* TARGET_EXPRs are peculiar: operands 1 and 3 can be the same.
  	 But, we only want to walk once.  */
        if (code == TARGET_EXPR
  	  && TREE_OPERAND (*tp, 3) == TREE_OPERAND (*tp, 1))
! 	--i;
!       /* Go through the subtrees.  */
!       while (i >= 0)
! 	{
! 	  WALK_SUBTREE (TREE_OPERAND (*tp, i));
! 	  --i;
! 	}
  
        /* For statements, we also walk the chain so that we cover the
  	 entire statement tree.  */
--- 1623,1641 ----
        || TREE_CODE_CLASS (code) == 'r'
        || TREE_CODE_CLASS (code) == 's')
      {
!       int i, len;
  
        /* Walk over all the sub-trees of this operand.  */
!       len = first_rtl_op (code);
        /* TARGET_EXPRs are peculiar: operands 1 and 3 can be the same.
  	 But, we only want to walk once.  */
        if (code == TARGET_EXPR
  	  && TREE_OPERAND (*tp, 3) == TREE_OPERAND (*tp, 1))
! 	--len;
!       /* Go through the subtrees.  We need to do this in forward order so
!          that the scope of a FOR_EXPR is handled properly.  */
!       for (i = 0; i < len; ++i)
! 	WALK_SUBTREE (TREE_OPERAND (*tp, i));
  
        /* For statements, we also walk the chain so that we cover the
  	 entire statement tree.  */