Wrong result of "constructible" traits for unbound arrays

[bolshakov-a]

These asserts fail in Clang (and GCC as well) even in the C++20 mode:

static_assert(__is_constructible(int[], int, int));
static_assert(__is_nothrow_constructible(int[], int, int));

However, it seems to be wrong because [meta.unary.prop] p. 9 states:

The predicate condition for a template specialization is_constructible<T, Args...> shall be satisfied if and only if the following variable definition would be well-formed for some invented variable t:
T t(declval<Args>()...);

and the following compiles (C++20 and later):

#include <type_traits>

template< class T >
std::add_rvalue_reference_t<T> declval() noexcept;

using T = int[];

T t(declval<int>(), declval<int>());

It looks correct according to [dcl.init.general] p. 16.5:

... If the destination type is an array of unknown bound, it is defined as having k elements.

Is it actually a bug, or am I missing something?

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Andrey Ali Khan Bolshakov (bolshakov-a)

These asserts fail in Clang (and GCC as well) even in the C++20 mode: ```cpp static_assert(__is_constructible(int[], int, int)); static_assert(__is_nothrow_constructible(int[], int, int)); ``` However, it seems to be wrong because [[meta.unary.prop] p. 9](http://eel.is/c++draft/meta#unary.prop-9) states: > The predicate condition for a template specialization is_constructible<T, Args...> shall be satisfied if and only if the following variable definition would be well-formed for some invented variable t: > `T t(declval<Args>()...);`

and the following compiles (C++20 and later):

#include <type_traits>

template< class T >
std::add_rvalue_reference_t<T> declval() noexcept;

using T = int[];

T t(declval<int>(), declval<int>());

Is it actually a bug, or am I missing something?

[cor3ntin]

maybe a LWG issue @ldionne

my guess is the wording predates paren list initialization

semi related https://cplusplus.github.io/LWG/lwg-active.html#2077

[bolshakov-a]

my guess is the wording predates paren list initialization

It has been not the case since C++20 (P0960).

semi related https://cplusplus.github.io/LWG/lwg-active.html#2077

Honestly, I don't understand the problem with is_destructible. The standard clearly states that the trait may evaluate to true only when the type is a complete object type if not a reference, whereas arrays of unknown bound are always incomplete ([basic.types] p.5), regardless of whether the element type is complete or not. So, std::is_destructible_v<A[]> should be false, and it looks quite reasonable.

[lichray]

These holds

static_assert(__is_constructible(int[2], int, int));
static_assert(__is_nothrow_constructible(int[2], int, int));

But https://eel.is/c++draft/meta#unary.prop-9 allows T to be arrays of unknown bound 🤔

[bolshakov-a]

maybe a LWG issue

It turned out that it is already open!
https://wg21.link/lwg3486