Difference between revisions of "cpp/language/new"

Revision as of 23:47, 3 January 2024

Creates and initializes objects with dynamic storage duration, that is, objects whose lifetime is not necessarily limited by the scope in which they were created.

Syntax

Explanation

The new expression attempts to allocate storage and then attempts to construct and initialize either a single unnamed object, or an unnamed array of objects in the allocated storage. The new-expression returns a prvalue pointer to the constructed object or, if an array of objects was constructed, a pointer to the initial element of the array.

Syntax (1) or (3) is required if type includes parentheses:

new int(*[10])();    // error: parsed as (new int) (*[10]) ()
new (int (*[10])()); // okay: allocates an array of 10 pointers to functions

In addition, type is parsed greedily: it will be taken include every token that can be a part of a declarator:

new int + 1; // okay: parsed as (new int) + 1, increments a pointer returned by new int
new int * 1; // error: parsed as (new int*) (1)

The new-initializer is not optional if

• type is an array of unknown bound,

double* p = new double[]{1, 2, 3}; // creates an array of type double[3]
auto p = new auto('c');            // creates a single object of type char. p is a char*
 
auto q = new std::integral auto(1);         // OK: q is an int*
auto q = new std::floating_point auto(true) // ERROR: type constraint not satisfied
 
auto r = new std::pair(1, true); // OK: r is a std::pair<int, bool>*
auto r = new std::vector;        // ERROR: element type can't be deduced

Dynamic arrays

If type is an array type, all dimensions other than the first must be specified as positive integral constant expression(until C++14)converted constant expression of type std::size_t(since C++14), but (only when using un-parenthesized syntaxes (2) and (4)) the first dimension may be an expression of integral type, enumeration type, or class type with a single non-explicit conversion function to integral or enumeration type(until C++14)any expression convertible to std::size_t(since C++14). This is the only way to directly create an array with size defined at runtime, such arrays are often referred to as dynamic arrays:

int n = 42;
double a[n][5]; // error
auto p1 = new  double[n][5];  // OK
auto p2 = new  double[5][n];  // error: only the first dimension may be non-constant
auto p3 = new (double[n][5]); // error: syntax (1) cannot be used for dynamic arrays

The first dimension of zero is acceptable, and the allocation function is called.

If new-initializer is a braced-init-list, and the first dimension is potentially evaluated and not a core constant expression, the semantic constraints of copy-initializing a hypothetical element of the array from an empty initializer list are checked.

Allocation

The new-expression allocates storage by calling the appropriate allocation function. If type is a non-array type, the name of the function is operator new. If type is an array type, the name of the function is operator new[].

As described in allocation function, the C++ program may provide global and class-specific replacements for these functions. If the new-expression begins with the optional :: operator, as in ::new T or ::new T[n], class-specific replacements will be ignored (the function is looked up in global scope). Otherwise, if T is a class type, lookup begins in the class scope of T.

When calling the allocation function, the new-expression passes the number of bytes requested as the first argument, of type std::size_t, which is exactly sizeof(T) for non-array T.

Array allocation may supply unspecified overhead, which may vary from one call to new to the next, unless the allocation function selected is the standard non-allocating form. The pointer returned by the new-expression will be offset by that value from the pointer returned by the allocation function. Many implementations use the array overhead to store the number of objects in the array which is used by the delete[] expression to call the correct number of destructors. In addition, if the new-expression is used to allocate an array of char, unsigned char, or std::byte(since C++17), it may request additional memory from the allocation function if necessary to guarantee correct alignment of objects of all types no larger than the requested array size, if one is later placed into the allocated array.

Placement new

If placement-args are provided, they are passed to the allocation function as additional arguments. Such allocation functions are known as "placement new", after the standard allocation function void* operator new(std::size_t, void*), which simply returns its second argument unchanged. This is used to construct objects in allocated storage:

// within any block scope...
{
    // Statically allocate the storage with automatic storage duration
    // which is large enough for any object of type “T”.
    alignas(T) unsigned char buf[sizeof(T)];
 
    T* tptr = new(buf) T; // Construct a “T” object, placing it directly into your 
                          // pre-allocated storage at memory address “buf”.
 
    tptr->~T();           // You must **manually** call the object's destructor
                          // if its side effects is depended by the program.
}                         // Leaving this block scope automatically deallocates “buf”.

Note: this functionality is encapsulated by the member functions of the Allocator classes.

new T;      // calls operator new(sizeof(T))
            // (C++17) or operator new(sizeof(T), std::align_val_t(alignof(T))))
new T[5];   // calls operator new[](sizeof(T)*5 + overhead)
            // (C++17) or operator new(sizeof(T)*5+overhead, std::align_val_t(alignof(T))))
new(2,f) T; // calls operator new(sizeof(T), 2, f)
            // (C++17) or operator new(sizeof(T), std::align_val_t(alignof(T)), 2, f)

If a non-throwing allocation function (e.g. the one selected by new(std::nothrow) T) returns a null pointer because of an allocation failure, then the new-expression returns immediately, it does not attempt to initialize an object or to call a deallocation function. If a null pointer is passed as the argument to a non-allocating placement new-expression, which makes the selected standard non-allocating placement allocation function return a null pointer, the behavior is undefined.

Initialization

The object created by a new-expression is initialized according to the following rules.

If type is not an array type, the single object is constructed in the acquired memory area:

• If new-initializer is absent, the object is default-initialized.
• If new-initializer is a parentheses-enclosed expression list, the object is direct-initialized.

If type is an array type, an array of objects is initialized:

• If new-initializer is absent, each element is default-initialized.

• Even if the first dimension is zero, the semantic constraints of default-initializing a hypothetical element still need to be met.

• If new-initializer is a pair of parentheses, each element is value-initialized.

• Even if the first dimension is zero, the semantic constraints of value-initializing a hypothetical element still need to be met.

Initialization failure

If initialization terminates by throwing an exception (e.g. from the constructor), the program looks up a matching deallocation function, then:

• If a suitable deallocation function can be found, the deallocation function is called to free the memory in which the object was being constructed. After that, the exception continues to propagate in the context of the new-expression.
• If no unambiguous matching deallocation function can be found, propagating the exception does not cause the object’s memory to be freed. It is only appropriate if the called allocation function does not allocate memory, otherwise it is likely to result in a memory leak.

The scope of the lookup of the matching deallocation function is performed as follows:

• If the new-expression does not begin with ::, and the allocated type is either a class type T or an array of class type T, a search is performed for the deallocation function’s name in the scope of T.
• Otherwise, or if nothing is found in the T's class scope, the deallocation function’s name is looked up by searching for it in the global scope.

For a non-placement allocation function, the normal deallocation function lookup is used to find the matching deallocation function (see delete-expression).

For a placement allocation function, the matching decllocation function has the same number of parameter, and each parameter type except the first is identical to the corresponding parameter type of the allocation function (after parameter transformations).

• If the lookup finds a single matching deallocation function, that function will be called; otherwise, no deallocation function will be called.
• If the lookup finds a non-placement deallocation function and that function, considered as a placement deallocation function, would have been selected as a match for the allocation function, the program is ill-formed.

In any case, the matching deallocation function (if any) must be non-deleted and(since C++11) accessible from the point where the new-expression appears.

struct S
{
    // Placement allocation function:
    static void* operator new(std::size_t, std::size_t);
 
    // Non-placement deallocation function:
    static void operator delete(void*, std::size_t);
};
 
S* p = new (0) S; // error: non-placement deallocation function matches
                  //        placement allocation function

If a deallocation function is called in a new-expression (due to initialization failure), the arguments passed to that function are determined as follows:

• The first argument is the value (of type void*) returned from the allocation function call.
• Other arguments (only for placement deallocation functions) are the placement-args passed to the placement allocation function.

If the implementation is allowed to introduce a temporary object or make a copy of any argument as part of the call to the allocation function, it is unspecified whether the same object is used in the call to both the allocation and deallocation functions.

Memory leaks

The objects created by new-expressions (objects with dynamic storage duration) persist until the pointer returned by the new-expression is used in a matching delete-expression. If the original value of pointer is lost, the object becomes unreachable and cannot be deallocated: a memory leak occurs.

This may happen if the pointer is assigned to:

int* p = new int(7); // dynamically allocated int with value 7
p = nullptr; // memory leak

or if the pointer goes out of scope:

void f()
{
    int* p = new int(7);
} // memory leak

or due to exception:

void f()
{
    int* p = new int(7);
    g();      // may throw
    delete p; // okay if no exception
} // memory leak if g() throws

To simplify management of dynamically-allocated objects, the result of a new-expression is often stored in a smart pointer: std::auto_ptr (until C++17)std::unique_ptr, or std::shared_ptr(since C++11). These pointers guarantee that the delete expression is executed in the situations shown above.

Keywords

new

Notes

Itanium C++ ABI requires that the array allocation overhead is zero if the element type of the created array is trivially destructible. So does MSVC.

Some implementations (e.g. MSVC before VS 2019 v16.7) require non-zero array allocation overhead on non-allocating placement array new if the element type is not trivially destructible, which is no longer conforming since CWG 2382.

A non-allocating placement array new-expression that creates an array of unsigned char, or std::byte(since C++17) can be used to implicitly create objects on given region of storage: it ends lifetime of objects overlapping with the array, and then implicitly creates objects of implicit-lifetime types in the array.

std::vector offers similar functionality for one-dimensional dynamic arrays.

Defect reports

The following behavior-changing defect reports were applied retroactively to previously published C++ standards.

See also

• constructor
• copy elision
• default constructor
• delete
• destructor
• initialization
  • aggregate initialization
  • default initialization
  • direct initialization
  • list initialization
  • value initialization