From: Jason Merrill <jason@redhat.com>
To: Marek Polacek <polacek@redhat.com>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH v2] c++: fix broken copy elision with nested TARGET_EXPRs [PR105550]
Date: Fri, 1 Jul 2022 11:48:41 -0400 [thread overview]
Message-ID: <dea0951e-17c6-3d81-b730-afaf28f4418d@redhat.com> (raw)
In-Reply-To: <YrZXHg4qzuzp3anZ@redhat.com>
On 6/24/22 20:30, Marek Polacek wrote:
> On Thu, Jun 02, 2022 at 05:08:54PM -0400, Jason Merrill wrote:
>> On 5/26/22 11:01, Marek Polacek wrote:
>>> In this problem, we are failing to properly perform copy elision with
>>> a conditional operator, so this:
>>>
>>> constexpr A a = true ? A{} : A{};
>>>
>>> fails with:
>>>
>>> error: 'A{((const A*)(&<anonymous>))}' is not a constant expression
>>>
>>> The whole initializer is
>>>
>>> TARGET_EXPR <D.2395, 1 ? TARGET_EXPR <D.2393, {.p=(const struct A *) &<PLACEHOLDER_EXPR struct A>}> : TARGET_EXPR <D.2394, {.p=(const struct A *) &<PLACEHOLDER_EXPR struct A>}>>
>>>
>>> where the outermost TARGET_EXPR is elided, but not the nested ones.
>>> Then we end up replacing the PLACEHOLDER_EXPRs with the temporaries the
>>> TARGET_EXPRs represent, which is precisely what should *not* happen with
>>> copy elision.
>>>
>>> I've tried the approach of tweaking ctx->object, but I ran into gazillion
>>> problems with that. I thought that I would let cxx_eval_constant_expression
>>> /TARGET_EXPR create a new object only when ctx->object was null, then
>>> adjust setting of ctx->object in places like cxx_bind_parameters_in_call
>>> and cxx_eval_component_reference but that failed completely. Sometimes
>>> ctx->object has to be reset, sometimes it cannot be reset, 'this' needed
>>> special handling, etc. I gave up.
>>>> But now that we have potential_prvalue_result_of, a simple but less
>>> elegant solution is the following. I thought about setting a flag on
>>> a TARGET_EXPR to avoid adding ctx.full_expr, but a new flag would be
>>> overkill and using TARGET_EXPR_DIRECT_INIT_P broke things.
>
> Sorry it's taken me so long to get back to this.
>
>> This doesn't seem like a general solution; the same issue would also apply
>> to ?: of TARGET_EXPR when it's a subexpression rather than the full
>> expression, like f(true ? A{} : B{}).
>
> You're right.
>
>> Another simple approach, but more general, would be to routinely strip
>> TARGET_EXPR from the operands of ?: like we do in various other places in
>> constexpr.c.
>
> How about this, then?
>
> Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?
OK.
> -- >8 --
> In this problem, we are failing to properly perform copy elision with
> a conditional operator, so this:
>
> constexpr A a = true ? A{} : A{};
>
> fails with:
>
> error: 'A{((const A*)(&<anonymous>))}' is not a constant expression
>
> The whole initializer is
>
> TARGET_EXPR <D.2395, 1 ? TARGET_EXPR <D.2393, {.p=(const struct A *) &<PLACEHOLDER_EXPR struct A>}> : TARGET_EXPR <D.2394, {.p=(const struct A *) &<PLACEHOLDER_EXPR struct A>}>>
>
> where the outermost TARGET_EXPR is elided, but not the nested ones.
> Then we end up replacing the PLACEHOLDER_EXPRs with the temporaries the
> TARGET_EXPRs represent, which is precisely what should *not* happen with
> copy elision.
>
> I've tried the approach of tweaking ctx->object, but I ran into gazillion
> problems with that. I thought that I would let cxx_eval_constant_expression
> /TARGET_EXPR create a new object only when ctx->object was null, then
> adjust setting of ctx->object in places like cxx_bind_parameters_in_call
> and cxx_eval_component_reference but that failed completely. Sometimes
> ctx->object has to be reset, sometimes it cannot be reset, 'this' needed
> special handling, etc. I gave up.
>
> Instead, this patch strips TARGET_EXPRs from the operands of ?: like
> we do in various other places in constexpr.c.
>
> PR c++/105550
>
> gcc/cp/ChangeLog:
>
> * constexpr.cc (cxx_eval_conditional_expression): Strip TARGET_EXPRs.
>
> gcc/testsuite/ChangeLog:
>
> * g++.dg/cpp1y/nsdmi-aggr16.C: Remove FIXME.
> * g++.dg/cpp1y/nsdmi-aggr17.C: Remove FIXME.
> * g++.dg/cpp0x/constexpr-elision1.C: New test.
> * g++.dg/cpp1y/constexpr-elision1.C: New test.
> ---
> gcc/cp/constexpr.cc | 7 +++
> .../g++.dg/cpp0x/constexpr-elision1.C | 16 ++++++
> .../g++.dg/cpp1y/constexpr-elision1.C | 53 +++++++++++++++++++
> gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr16.C | 5 +-
> gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr17.C | 5 +-
> 5 files changed, 80 insertions(+), 6 deletions(-)
> create mode 100644 gcc/testsuite/g++.dg/cpp0x/constexpr-elision1.C
> create mode 100644 gcc/testsuite/g++.dg/cpp1y/constexpr-elision1.C
>
> diff --git a/gcc/cp/constexpr.cc b/gcc/cp/constexpr.cc
> index 0dc94d9445d..5f7fc6f8f0c 100644
> --- a/gcc/cp/constexpr.cc
> +++ b/gcc/cp/constexpr.cc
> @@ -3507,6 +3507,13 @@ cxx_eval_conditional_expression (const constexpr_ctx *ctx, tree t,
> val = TREE_OPERAND (t, 1);
> if (TREE_CODE (t) == IF_STMT && !val)
> val = void_node;
> + /* A TARGET_EXPR may be nested inside another TARGET_EXPR, but still
> + serve as the initializer for the same object as the outer TARGET_EXPR,
> + as in
> + A a = true ? A{} : A{};
> + so strip the inner TARGET_EXPR so we don't materialize a temporary. */
> + if (TREE_CODE (val) == TARGET_EXPR)
> + val = TARGET_EXPR_INITIAL (val);
> return cxx_eval_constant_expression (ctx, val, lval, non_constant_p,
> overflow_p, jump_target);
> }
> diff --git a/gcc/testsuite/g++.dg/cpp0x/constexpr-elision1.C b/gcc/testsuite/g++.dg/cpp0x/constexpr-elision1.C
> new file mode 100644
> index 00000000000..9e7b9ec3405
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp0x/constexpr-elision1.C
> @@ -0,0 +1,16 @@
> +// PR c++/105550
> +// { dg-do compile { target c++11 } }
> +
> +template <typename, typename> struct pair {
> + constexpr pair(int, int) {}
> +};
> +template <typename _Tp, typename _Compare>
> +pair<const _Tp &, const _Tp &> minmax(const _Tp &__a, const _Tp &__b,
> + _Compare) {
> + return 0 ? pair<const _Tp &, const _Tp &>(__b, __a)
> + : pair<const _Tp &, const _Tp &>(__a, __b);
> +}
> +typedef int value_type;
> +typedef int compare_type;
> +template pair<const value_type &, const value_type &>
> +minmax(const value_type &, const value_type &, compare_type);
> diff --git a/gcc/testsuite/g++.dg/cpp1y/constexpr-elision1.C b/gcc/testsuite/g++.dg/cpp1y/constexpr-elision1.C
> new file mode 100644
> index 00000000000..b225ae29cde
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp1y/constexpr-elision1.C
> @@ -0,0 +1,53 @@
> +// PR c++/105550
> +// { dg-do compile { target c++14 } }
> +
> +struct A {
> + const A *p = this;
> +};
> +
> +struct B {
> + const B *p = this;
> + constexpr operator A() const { return {}; }
> +};
> +
> +constexpr A
> +bar (A)
> +{
> + return {};
> +}
> +
> +constexpr A baz() { return {}; }
> +
> +struct E {
> + A a1 = true ? A{} : A{};
> + A a2 = true ? A{} : B{};
> + A a3 = false ? A{} : B{};
> + A a4 = false ? B{} : B{};
> + A a5 = A{};
> + A a6 = B{};
> + A a7 = false ? B{} : (true ? A{} : A{});
> + A a8 = false ? (true ? A{} : B{}) : (true ? A{} : A{});
> + A a9 = (A{});
> + A a10 = (true, A{});
> + A a11 = bar (A{});
> + A a12 = baz ();
> + A a13 = (A{}, A{});
> +};
> +
> +constexpr E e{};
> +
> +constexpr A a1 = true ? A{} : A{};
> +constexpr A a2 = true ? A{} : B{};
> +constexpr A a3 = false ? A{} : B{};
> +constexpr A a4 = false ? B{} : B{};
> +constexpr A a5 = A{};
> +constexpr A a6 = B{};
> +constexpr A a7 = false ? B{} : (true ? A{} : A{});
> +constexpr A a8 = false ? (true ? A{} : B{}) : (true ? A{} : A{});
> +constexpr A a9 = (A{});
> +constexpr A a10 = (true, A{});
> +constexpr A a11 = bar (A{});
> +//static_assert(a10.p == &a10, ""); // bug, 105619
> +constexpr A a12 = baz ();
> +//static_assert(a11.p == &a11, ""); // bug, 105619
> +constexpr A a13 = (A{}, A{});
> diff --git a/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr16.C b/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr16.C
> index dc6492c1b0b..5e66bdd2370 100644
> --- a/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr16.C
> +++ b/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr16.C
> @@ -40,9 +40,8 @@ struct E {
> A d = true ? (false ? A{} : A{}) : (false ? A{} : A{});
> };
>
> -// FIXME: When fixing this, also fix nsdmi-aggr17.C.
> -constexpr E e; // { dg-bogus "" "PR105550" { xfail *-*-* } }
> -SA (e.a.p == &e.a); // { dg-bogus "" "PR105550" { xfail *-*-* } }
> +constexpr E e;
> +SA (e.a.p == &e.a);
>
> E e1 = { };
>
> diff --git a/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr17.C b/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr17.C
> index fc27a2cdac7..ca9637b37eb 100644
> --- a/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr17.C
> +++ b/gcc/testsuite/g++.dg/cpp1y/nsdmi-aggr17.C
> @@ -56,9 +56,8 @@ struct F {
> A a = true ? A{x} : A{x};
> };
>
> -// FIXME: Doesn't work due to PR105550.
> -//constexpr F f;
> -//SA (f.a.p == &f.a);
> +constexpr F f;
> +SA (f.a.p == &f.a);
> SA (e.x == 42);
> F f2 = { };
> F f3 = { 42 };
>
> base-commit: 113844d68e94f4e9c0e946db351ba7d3d4a1335a
prev parent reply other threads:[~2022-07-01 15:48 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-05-26 15:01 [PATCH] " Marek Polacek
2022-06-02 21:08 ` Jason Merrill
2022-06-25 0:30 ` [PATCH v2] " Marek Polacek
2022-07-01 15:48 ` Jason Merrill [this message]
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=dea0951e-17c6-3d81-b730-afaf28f4418d@redhat.com \
--to=jason@redhat.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=polacek@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).