From: Nathaniel Shead <nathanieloshead@gmail.com>
To: gcc-patches@gcc.gnu.org
Cc: Jason Merrill <jason@redhat.com>
Subject: Re: [PATCH] c++: fix noexcept checking for trivial operations [PR96090]
Date: Mon, 27 Nov 2023 22:07:06 +1100 [thread overview]
Message-ID: <6564785e.630a0220.fe20c.09b1@mx.google.com> (raw)
In-Reply-To: <653db8c5.170a0220.c70e1.602f@mx.google.com>
Ping for https://gcc.gnu.org/pipermail/gcc-patches/2023-October/634626.html.
I've been made aware since constructing this patch of CWG2820, which has
a proposed resolution that would change the result of the testcase
'noexcept(yesthrow_t())' (and similarly for the library builtin), but as
it hasn't yet been accepted I think at least ensuring the builtin
matches the behaviour of the operator is probably still sensible.
On Sun, Oct 29, 2023 at 12:43:28PM +1100, Nathaniel Shead wrote:
> Bootstrapped and regtested on x86_64-pc-linux-gnu.
>
> -- >8 --
>
> This patch stops eager folding of trivial operations (construction and
> assignment) from occurring when checking for noexceptness. This was
> previously done in PR c++/53025, but only for copy/move construction,
> and the __is_nothrow_xible builtins did not receive the same treatment
> when they were added.
>
> To handle `is_nothrow_default_constructible`, the patch also ensures
> that when no parameters are passed we do value initialisation instead of
> just building the constructor call: in particular, value-initialisation
> doesn't necessarily actually invoke the constructor for trivial default
> constructors, and so we need to handle this case as well.
>
> PR c++/96090
> PR c++/100470
>
> gcc/cp/ChangeLog:
>
> * call.cc (build_over_call): Prevent folding of trivial special
> members when checking for noexcept.
> * method.cc (constructible_expr): Perform value-initialisation
> for empty parameter lists.
> (is_nothrow_xible): Treat as noexcept operator.
>
> gcc/testsuite/ChangeLog:
>
> * g++.dg/cpp0x/noexcept81.C: New test.
> * g++.dg/ext/is_nothrow_constructible7.C: New test.
> * g++.dg/ext/is_nothrow_constructible8.C: New test.
>
> Signed-off-by: Nathaniel Shead <nathanieloshead@gmail.com>
> ---
> gcc/cp/call.cc | 17 ++---
> gcc/cp/method.cc | 19 ++++--
> gcc/testsuite/g++.dg/cpp0x/noexcept81.C | 36 +++++++++++
> .../g++.dg/ext/is_nothrow_constructible7.C | 20 ++++++
> .../g++.dg/ext/is_nothrow_constructible8.C | 63 +++++++++++++++++++
> 5 files changed, 141 insertions(+), 14 deletions(-)
> create mode 100644 gcc/testsuite/g++.dg/cpp0x/noexcept81.C
> create mode 100644 gcc/testsuite/g++.dg/ext/is_nothrow_constructible7.C
> create mode 100644 gcc/testsuite/g++.dg/ext/is_nothrow_constructible8.C
>
> diff --git a/gcc/cp/call.cc b/gcc/cp/call.cc
> index c1fb8807d3f..ac02b0633ed 100644
> --- a/gcc/cp/call.cc
> +++ b/gcc/cp/call.cc
> @@ -10231,15 +10231,16 @@ build_over_call (struct z_candidate *cand, int flags, tsubst_flags_t complain)
> /* Avoid actually calling copy constructors and copy assignment operators,
> if possible. */
>
> - if (! flag_elide_constructors && !force_elide)
> + if (!force_elide
> + && (!flag_elide_constructors
> + /* It's unsafe to elide the operation when handling
> + a noexcept-expression, it may evaluate to the wrong
> + value (c++/53025, c++/96090). */
> + || cp_noexcept_operand != 0))
> /* Do things the hard way. */;
> - else if (cand->num_convs == 1
> - && (DECL_COPY_CONSTRUCTOR_P (fn)
> - || DECL_MOVE_CONSTRUCTOR_P (fn))
> - /* It's unsafe to elide the constructor when handling
> - a noexcept-expression, it may evaluate to the wrong
> - value (c++/53025). */
> - && (force_elide || cp_noexcept_operand == 0))
> + else if (cand->num_convs == 1
> + && (DECL_COPY_CONSTRUCTOR_P (fn)
> + || DECL_MOVE_CONSTRUCTOR_P (fn)))
> {
> tree targ;
> tree arg = argarray[num_artificial_parms_for (fn)];
> diff --git a/gcc/cp/method.cc b/gcc/cp/method.cc
> index a70dd5d6adc..3c978e2369d 100644
> --- a/gcc/cp/method.cc
> +++ b/gcc/cp/method.cc
> @@ -2091,6 +2091,7 @@ constructible_expr (tree to, tree from)
> {
> tree expr;
> cp_unevaluated cp_uneval_guard;
> + const int len = TREE_VEC_LENGTH (from);
> if (CLASS_TYPE_P (to))
> {
> tree ctype = to;
> @@ -2098,11 +2099,16 @@ constructible_expr (tree to, tree from)
> if (!TYPE_REF_P (to))
> to = cp_build_reference_type (to, /*rval*/false);
> tree ob = build_stub_object (to);
> - vec_alloc (args, TREE_VEC_LENGTH (from));
> - for (tree arg : tree_vec_range (from))
> - args->quick_push (build_stub_object (arg));
> - expr = build_special_member_call (ob, complete_ctor_identifier, &args,
> - ctype, LOOKUP_NORMAL, tf_none);
> + if (len == 0)
> + expr = build_value_init (ctype, tf_none);
> + else
> + {
> + vec_alloc (args, TREE_VEC_LENGTH (from));
> + for (tree arg : tree_vec_range (from))
> + args->quick_push (build_stub_object (arg));
> + expr = build_special_member_call (ob, complete_ctor_identifier, &args,
> + ctype, LOOKUP_NORMAL, tf_none);
> + }
> if (expr == error_mark_node)
> return error_mark_node;
> /* The current state of the standard vis-a-vis LWG 2116 is that
> @@ -2120,7 +2126,6 @@ constructible_expr (tree to, tree from)
> }
> else
> {
> - const int len = TREE_VEC_LENGTH (from);
> if (len == 0)
> return build_value_init (strip_array_types (to), tf_none);
> if (len > 1)
> @@ -2216,7 +2221,9 @@ is_trivially_xible (enum tree_code code, tree to, tree from)
> bool
> is_nothrow_xible (enum tree_code code, tree to, tree from)
> {
> + ++cp_noexcept_operand;
> tree expr = is_xible_helper (code, to, from, /*trivial*/false);
> + --cp_noexcept_operand;
> if (expr == NULL_TREE || expr == error_mark_node)
> return false;
> return expr_noexcept_p (expr, tf_none);
> diff --git a/gcc/testsuite/g++.dg/cpp0x/noexcept81.C b/gcc/testsuite/g++.dg/cpp0x/noexcept81.C
> new file mode 100644
> index 00000000000..a1481613b5d
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/noexcept81.C
> @@ -0,0 +1,36 @@
> +// { dg-do compile { target c++11 } }
> +// PR c++/96090
> +
> +struct yesthrow_t {
> + yesthrow_t() noexcept(false) = default;
> + yesthrow_t(const yesthrow_t&) noexcept(false) = default;
> + yesthrow_t(yesthrow_t&&) noexcept(false) = default;
> + yesthrow_t& operator=(const yesthrow_t&) noexcept(false) = default;
> + yesthrow_t& operator=(yesthrow_t&&) noexcept(false) = default;
> +};
> +
> +yesthrow_t yes;
> +static_assert(not noexcept(yesthrow_t(static_cast<const yesthrow_t&>(yes))), "");
> +static_assert(not noexcept(yesthrow_t(static_cast<yesthrow_t&&>(yes))), "");
> +static_assert(not noexcept(yes = static_cast<const yesthrow_t&>(yes)), "");
> +static_assert(not noexcept(yes = static_cast<yesthrow_t&&>(yes)), "");
> +
> +// Note: this is value-initialisation, and thus by [dcl.init.general] p9
> +// a trivial non-user-provided non-deleted default constructor is not called.
> +static_assert(noexcept(yesthrow_t()), "");
> +
> +struct nothrow_t {
> + nothrow_t() noexcept(true) = default;
> + nothrow_t(const nothrow_t&) noexcept(true) = default;
> + nothrow_t(nothrow_t&&) noexcept(true) = default;
> + nothrow_t& operator=(const nothrow_t&) noexcept(true) = default;
> + nothrow_t& operator=(nothrow_t&&) noexcept(true) = default;
> +};
> +
> +nothrow_t no;
> +static_assert(noexcept(nothrow_t()), "");
> +static_assert(noexcept(nothrow_t(static_cast<const nothrow_t&>(no))), "");
> +static_assert(noexcept(nothrow_t(static_cast<nothrow_t&&>(no))), "");
> +static_assert(noexcept(no = static_cast<const nothrow_t&>(no)), "");
> +static_assert(noexcept(no = static_cast<nothrow_t&&>(no)), "");
> +
> diff --git a/gcc/testsuite/g++.dg/ext/is_nothrow_constructible7.C b/gcc/testsuite/g++.dg/ext/is_nothrow_constructible7.C
> new file mode 100644
> index 00000000000..b63b13ac52f
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/ext/is_nothrow_constructible7.C
> @@ -0,0 +1,20 @@
> +// { dg-do compile { target c++11 } }
> +// PR c++/100470
> +
> +struct S1{
> + S1(S1&&) noexcept(false);
> +};
> +struct S2{
> + S2(S2&&) noexcept(false) = default;
> +};
> +struct S3{
> + S3(S3&&) noexcept(false){}
> +};
> +struct S4{
> + S4(S4&&) = default;
> +};
> +
> +static_assert(!__is_nothrow_constructible(S1, S1), "");
> +static_assert(!__is_nothrow_constructible(S2, S2), "");
> +static_assert(!__is_nothrow_constructible(S3, S3), "");
> +static_assert( __is_nothrow_constructible(S4, S4), "");
> diff --git a/gcc/testsuite/g++.dg/ext/is_nothrow_constructible8.C b/gcc/testsuite/g++.dg/ext/is_nothrow_constructible8.C
> new file mode 100644
> index 00000000000..f23d48fa888
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/ext/is_nothrow_constructible8.C
> @@ -0,0 +1,63 @@
> +// { dg-do compile { target c++11 } }
> +// PR c++/96090
> +
> +template <typename T>
> +constexpr bool is_nothrow_default_constructible_v
> + = __is_nothrow_constructible(T);
> +template <typename T>
> +constexpr bool is_nothrow_copy_constructible_v
> + = __is_nothrow_constructible(T, const T&);
> +template <typename T>
> +constexpr bool is_nothrow_move_constructible_v
> + = __is_nothrow_constructible(T, T&&);
> +template <typename T>
> +constexpr bool is_nothrow_copy_assignable_v
> + = __is_nothrow_assignable(T, const T&);
> +template <typename T>
> +constexpr bool is_nothrow_move_assignable_v
> + = __is_nothrow_assignable(T, T&&);
> +
> +struct yesthrow_t {
> + yesthrow_t() noexcept(false) = default;
> + yesthrow_t(const yesthrow_t&) noexcept(false) = default;
> + yesthrow_t(yesthrow_t&&) noexcept(false) = default;
> + yesthrow_t& operator=(const yesthrow_t&) noexcept(false) = default;
> + yesthrow_t& operator=(yesthrow_t&&) noexcept(false) = default;
> +};
> +
> +static_assert(not is_nothrow_copy_constructible_v<yesthrow_t>, "");
> +static_assert(not is_nothrow_copy_assignable_v<yesthrow_t>, "");
> +static_assert(not is_nothrow_move_constructible_v<yesthrow_t>, "");
> +static_assert(not is_nothrow_move_assignable_v<yesthrow_t>, "");
> +
> +// Note: by [meta.unary.prop] p9 this should be value-initialisation,
> +// and thus by [dcl.init.general] p9 a trivial non-user-provided
> +// non-deleted default constructor is not called.
> +static_assert(is_nothrow_default_constructible_v<yesthrow_t>, "");
> +
> +struct nothrow_t {
> + nothrow_t() noexcept(true) = default;
> + nothrow_t(const nothrow_t&) noexcept(true) = default;
> + nothrow_t(nothrow_t&&) noexcept(true) = default;
> + nothrow_t& operator=(const nothrow_t&) noexcept(true) = default;
> + nothrow_t& operator=(nothrow_t&&) noexcept(true) = default;
> +};
> +
> +static_assert(is_nothrow_default_constructible_v<nothrow_t>, "");
> +static_assert(is_nothrow_copy_constructible_v<nothrow_t>, "");
> +static_assert(is_nothrow_copy_assignable_v<nothrow_t>, "");
> +static_assert(is_nothrow_move_constructible_v<nothrow_t>, "");
> +static_assert(is_nothrow_move_assignable_v<nothrow_t>, "");
> +
> +struct A { A() noexcept(false) = default; };
> +struct B { B(const B&) noexcept(false) = default; };
> +struct C { C(C&&) noexcept(false) = default; };
> +struct D { D& operator=(const D&) noexcept(false) = default; };
> +struct E { E& operator=(E&&) noexcept(false) = default; };
> +
> +static_assert(is_nothrow_default_constructible_v<A>, ""); // see above
> +static_assert(not is_nothrow_copy_constructible_v<B>, "");
> +static_assert(not is_nothrow_move_constructible_v<C>, "");
> +static_assert(not is_nothrow_copy_assignable_v<D>, "");
> +static_assert(not is_nothrow_move_assignable_v<E>, "");
> +
> --
> 2.42.0
>
next prev parent reply other threads:[~2023-11-27 11:07 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-10-29 1:43 Nathaniel Shead
2023-11-27 11:07 ` Nathaniel Shead [this message]
2023-12-09 20:40 ` Jason Merrill
2023-12-11 2:04 ` Nathaniel Shead
2023-12-11 2:20 ` Jason Merrill
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=6564785e.630a0220.fe20c.09b1@mx.google.com \
--to=nathanieloshead@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=jason@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).