OK.
On Mon, Aug 26, 2019 at 9:35 AM Marek Polacek <pola...@redhat.com> wrote:
> Now that DECL_MUTABLE_P checks that it got a FIELD_DECL node, One Does Not
> Simply Check DECL_P. For an ARRAY_REF, "elt" can be a VAR_DECL.
>
> Bootstrapped/regtested on x86_64-linux, ok for trunk?
>
> 2019-08-26 Marek Polacek <pola...@redhat.com>
>
> PR c++/91545 - ICE in constexpr store evaluation.
> * constexpr.c (cxx_eval_store_expression): Check FIELD_DECL instead
> of DECL_P.
>
> * g++.dg/cpp0x/pr91545.C: New test.
>
> diff --git gcc/cp/constexpr.c gcc/cp/constexpr.c
> index dbd0dc3b445..6c547d6d179 100644
> --- gcc/cp/constexpr.c
> +++ gcc/cp/constexpr.c
> @@ -3849,7 +3849,7 @@ cxx_eval_store_expression (const constexpr_ctx *ctx,
> tree t,
> {
> tree ob = TREE_OPERAND (probe, 0);
> tree elt = TREE_OPERAND (probe, 1);
> - if (DECL_P (elt) && DECL_MUTABLE_P (elt))
> + if (TREE_CODE (elt) == FIELD_DECL && DECL_MUTABLE_P (elt))
> mutable_p = true;
> if (evaluated
> && modifying_const_object_p (TREE_CODE (t), probe,
> mutable_p)
> diff --git gcc/testsuite/g++.dg/cpp0x/pr91545.C
> gcc/testsuite/g++.dg/cpp0x/pr91545.C
> new file mode 100644
> index 00000000000..2aa1fa18bd5
> --- /dev/null
> +++ gcc/testsuite/g++.dg/cpp0x/pr91545.C
> @@ -0,0 +1,5 @@
> +// PR c++/91545
> +// { dg-do compile { target c++11 } }
> +
> +long a[1];
> +int d, e { d && (a[d] = 0) };
>
