On Thu, Jun 02, 2016 at 11:14:48AM -0400, Jason Merrill wrote:
> On Thu, Jun 2, 2016 at 8:13 AM, Jakub Jelinek <ja...@redhat.com> wrote:
> > When cp_fold is called on INDIRECT_REF and ARRAY*_REF and any of the
> > arguments change in the recursive calls, we fail to copy TREE_THIS_VOLATILE
> > flag.
> >
> > PR c++/71372
> > * cp-gimplify.c (cp_fold): For INDIRECT_REF, if the folded
> > expression
> > is INDIRECT_REF or MEM_REF, copy over TREE_READONLY,
> > TREE_SIDE_EFFECTS
> > and TREE_THIS_VOLATILE flags. For ARRAY_REF and ARRAY_RANGE_REF,
> > copy
> > over TREE_READONLY, TREE_SIDE_EFFECTS and TREE_THIS_VOLATILE flags
> > to the newly built tree.
>
> Hmm, maybe we should be using build_indirect_ref and build_array_ref.
Those look too much front-endish to me (having lots of diagnostics etc. in them
for something that should have been already long diagnosed).
> Or perhaps factor the simple build and set flags out of
> cp_build_indirect_ref and call that.
>
> If you don't feel like making that change, the patch is also OK as is.
I've tried to follow what the C FE does here:
if (op0 != orig_op0
&& code == ADDR_EXPR
&& (op1 = get_base_address (op0)) != NULL_TREE
&& INDIRECT_REF_P (op1)
&& TREE_CONSTANT (TREE_OPERAND (op1, 0)))
ret = fold_convert_loc (loc, TREE_TYPE (expr), fold_offsetof_1 (op0));
else if (op0 != orig_op0 || in_init)
ret = in_init
? fold_build1_initializer_loc (loc, code, TREE_TYPE (expr), op0)
: fold_build1_loc (loc, code, TREE_TYPE (expr), op0);
else
ret = fold (expr);
if (code == INDIRECT_REF
&& ret != expr
&& INDIRECT_REF_P (ret))
{
TREE_READONLY (ret) = TREE_READONLY (expr);
TREE_SIDE_EFFECTS (ret) = TREE_SIDE_EFFECTS (expr);
TREE_THIS_VOLATILE (ret) = TREE_THIS_VOLATILE (expr);
}
and similar.
Note, we have also the same
problem in fold-const.c (fold), though we'd hit that more rarely than this, and
there we also should just copy the flags.
Jakub
