OK.
On Tue, Dec 11, 2018 at 1:35 PM Marek Polacek <pola...@redhat.com> wrote:
>
> On Tue, Dec 11, 2018 at 10:26:00AM -0500, Jason Merrill wrote:
> > On 12/10/18 8:48 PM, Marek Polacek wrote:
> > > A template-argument for a non-type template-parameter shall be a converted
> > > constant expression. But an lvalue-to-rvalue conversion applied to a
> > > volatile
> > > glvalue is not allowed to be part of the evaluation of a constant
> > > expression.
> > > So this test should be rejected.
> >
> > It occurred to me after my note on IRC that the
> > potential_constant_expression_1 test we were talking about,
> >
> > if (TREE_THIS_VOLATILE (t) && !DECL_P (t))
> >
> > ought to test want_rval rather than !DECL_P so that we consistently reject
> > the lvalue-to-rvalue conversion, and not other uses of a volatile lvalue.
> > And the diagnostic ought to talk about that rather than "side-effects".
>
> Done.
>
> > It might still be appropriate to change non_const_var_error, but I'd think
> > it could check TREE_THIS_VOLATILE itself, rather than the caller; I don't
> > see a need for the two calls to differ in their handling of volatile
> > variables.
>
> This change was no longer needed...
>
> > Perhaps decl_maybe_constant_var_p should return false for constexpr
> > volatile, as well.
>
> ...neither was this but I did it all the same because I think it's reasonable.
>
> Like I said, not planning to backport it, but I could be easily convinced.
>
> Bootstrapped/regtested on x86_64-linux, ok for trunk?
>
> 2018-12-11 Marek Polacek <pola...@redhat.com>
>
> PR c++/86608 - reading constexpr volatile variable.
> * constexpr.c (potential_constant_expression_1): Check want_rval
> instead of checking if we have a decl.
> * decl2.c (decl_maybe_constant_var_p): Don't consider volatile
> constexpr variables as maybe constant.
>
> * g++.dg/cpp0x/constexpr-volatile2.C: New test.
> * g++.dg/cpp0x/pr65327.C: Add dg-error.
>
> diff --git gcc/cp/constexpr.c gcc/cp/constexpr.c
> index 1c844a8c2ef..44db38029bf 100644
> --- gcc/cp/constexpr.c
> +++ gcc/cp/constexpr.c
> @@ -5476,10 +5476,11 @@ potential_constant_expression_1 (tree t, bool
> want_rval, bool strict, bool now,
> available, so we don't bother with switch tracking. */
> return true;
>
> - if (TREE_THIS_VOLATILE (t) && !DECL_P (t))
> + if (TREE_THIS_VOLATILE (t) && want_rval)
> {
> if (flags & tf_error)
> - error_at (loc, "expression %qE has side-effects", t);
> + error_at (loc, "lvalue-to-rvalue conversion of a volatile lvalue "
> + "%qE with type %qT", t, TREE_TYPE (t));
> return false;
> }
> if (CONSTANT_CLASS_P (t))
> diff --git gcc/cp/decl2.c gcc/cp/decl2.c
> index a8bf28a0cd9..1b3e758b625 100644
> --- gcc/cp/decl2.c
> +++ gcc/cp/decl2.c
> @@ -4313,7 +4313,7 @@ decl_maybe_constant_var_p (tree decl)
> tree type = TREE_TYPE (decl);
> if (!VAR_P (decl))
> return false;
> - if (DECL_DECLARED_CONSTEXPR_P (decl))
> + if (DECL_DECLARED_CONSTEXPR_P (decl) && !TREE_THIS_VOLATILE (decl))
> return true;
> if (DECL_HAS_VALUE_EXPR_P (decl))
> /* A proxy isn't constant. */
> diff --git gcc/testsuite/g++.dg/cpp0x/constexpr-volatile2.C
> gcc/testsuite/g++.dg/cpp0x/constexpr-volatile2.C
> new file mode 100644
> index 00000000000..0def8d73731
> --- /dev/null
> +++ gcc/testsuite/g++.dg/cpp0x/constexpr-volatile2.C
> @@ -0,0 +1,13 @@
> +// PR c++/86608
> +// { dg-do compile { target c++11 } }
> +
> +template<typename T, T v> struct X {};
> +
> +int
> +main ()
> +{
> + static constexpr volatile int a = 3;
> + constexpr volatile int b = 2;
> + return (sizeof(X<decltype(a), a>) // { dg-error "lvalue-to-rvalue
> conversion of a volatile lvalue" }
> + + sizeof(X<decltype(b), b>)); // { dg-error "lvalue-to-rvalue
> conversion of a volatile lvalue" }
> +}
> diff --git gcc/testsuite/g++.dg/cpp0x/pr65327.C
> gcc/testsuite/g++.dg/cpp0x/pr65327.C
> index c6cefaba692..5176b3c3204 100644
> --- gcc/testsuite/g++.dg/cpp0x/pr65327.C
> +++ gcc/testsuite/g++.dg/cpp0x/pr65327.C
> @@ -15,4 +15,4 @@ constexpr volatile int
> bar ()
> {
> return i;
> -}
> +} // { dg-error "lvalue-to-rvalue conversion of a volatile lvalue" }
