[PATCH] Refine -Waddress-of-packed-member once more

[Jakub Jelinek]

On Thu, Jan 24, 2019 at 06:39:22AM +0000, Bernd Edlinger wrote:
> --- gcc/c-family/c-warn.c	(revision 268195)
> +++ gcc/c-family/c-warn.c	(working copy)
> @@ -2725,14 +2725,18 @@ static tree
>  check_address_or_pointer_of_packed_member (tree type, tree rhs)
>  {
>    bool rvalue = true;
> +  bool indirect = false;
>  
>    if (INDIRECT_REF_P (rhs))
> -    rhs = TREE_OPERAND (rhs, 0);
> +    {
> +      rhs = TREE_OPERAND (rhs, 0);
> +      indirect = true;
> +    }
>  
>    if (TREE_CODE (rhs) == ADDR_EXPR)
>      {
>        rhs = TREE_OPERAND (rhs, 0);
> -      rvalue = false;
> +      rvalue = indirect;
>      }

Given the unfolded *&, I wonder if the above actually shouldn't be
a loop with indirection integral counter instead of a boolean
and simply bump the indirection count on INDIRECT_REF_P and decrease on
ADDR_EXPR, and if indirection count is > 0 after the loop return NULL_TREE?
Say for *&*& or similar, and ***var.field always not warning etc.

Otherwise the patch looks good.

Though you might throw in a few test lines with the intermixed casts and
compound exprs multiple times (why you've added correctly the loop in
there).

	Jakub