Bdtyktl

Consider this code (demo):

#include <tuple>

#include <type_traits>



struct Ag{int i;int j;};

using  T = std::tuple<int,int>;

using  Ar = int[2];



const Ag ag {};

const T t   {};

const Ar ar {};



void bind_ag(){

    auto [i,j] = ag;

    static_assert(std::is_same_v<decltype((i)),int&>);

    }

void bind_t(){

    auto [i,j] = t;

    static_assert(std::is_same_v<decltype((i)),int&>);

    }

void bind_ar(){

    auto [i,j] = ar;

    static_assert(std::is_same_v<decltype((i)),int&>);       //For GCC

    static_assert(std::is_same_v<decltype((i)),const int&>); //For Clang (and standard?)

    }

A structured binding to a copy of a const c-array are declared const by Clang and non-const by GCC.

The behavior of GCC for c-array is consistent with the behavior observed for aggregate or tuple-like types.

On the other hand from my reading of the standard, I suppose Clang follows what is written. In [dcl.struct.bind]/1 e has type cv A where A is the type of the initializer expression and the cv is the cv-qualifier of the structured binding declaration. And the type of the initializer expression ar is accordingly to [expr.type]/1 const int[2].

What should be expected? My opinion is that Clang follows the standard. On the other hand I feel the intent was that the behaviors for array, aggregate and tuple-like types were equivalent.

The wording of the standard in [dcl.struct.bind] says:

If the assignment-expression in the initializer has array type A and no ref-qualifier is present, e has type cv A and each element is copy-initialized or direct-initialized from the corresponding element of the assignment-expression as specified by the form of the initializer.

We have auto [i,j] = ar;, ar has array type const int[2], and the wording of the standard makes it clear that e has type const int[2]. Thus, per the wording, each binding references the element type - which is const int. Clang is technically correct.

However, as Richard Smith points out in gcc bug 80649:

I think this is a bug in the standard. The cv-qualifiers of the array type should be discarded, as they would be for any normal auto deduction.

That seems right. When you write auto x = y; you'd certainly expect x to not be top-level const, but here we have a situation where it still is. I don't think there's a Core issue open for this yet, but there should be.

In my opinion, GCC is correct (and, as you pointed out, more intuitive), though certainly some people disagree.

The decl-specifier-seq ("auto") has no cv-qualifiers (that is, you did not write const auto), and that should be what is considered as cv when deciding what qualifiers to put on the bindings.

[dcl.struct.bind]/1: [..] Let cv denote the cv-qualifiers in the decl-specifier-seq. [..]

[dcl.struct.bind]/3: If E is an array type with element type T, the number of elements in the identifier-list shall be equal to the number of elements of E. Each v_i is the name of an lvalue that refers to the element i of the array and whose type is T; the referenced type is T. [ Note: The top-level cv-qualifiers of T are cv. — end note ] [..]

Like you, Clang devs have interpreted the auto here as having cv-qualification of const, because the type inferred by it is const Ag. But I see no standard justification for that interpretation.