User-defined conversion function
Enables implicit conversion or explicit conversion from a class type to another type.
Syntax
Conversion function is declared like a non-static member function or member function template with no parameters, no explicit return type, and with the name of the form:
operator conversion-type-id
|
(1) | ||||||||
explicit operator conversion-type-id
|
(2) | (since C++11) | |||||||
explicit ( expression ) operator conversion-type-id
|
(3) | (since C++20) | |||||||
conversion-type-id is a type-id except that function and array operators []
or ()
are not allowed in its declarator (thus conversion to types such as pointer to array requires a type alias/typedef or an identity template: see below). Regardless of typedef, conversion-type-id cannot represent an array or a function type.
Although the return type is not allowed in the declaration of a user-defined conversion function, the decl-specifier-seq of the declaration grammar may be present and may include any specifier other than type-specifier or the keyword static
, In particular, besides explicit
, the specifiers inline
, virtual
, constexpr
(since C++11), consteval
(since C++20), and friend
are also allowed (note that friend
requires a qualified name: friend A::operator B();).
When such member function is declared in class X, it performs conversion from X to conversion-type-id:
struct X { // implicit conversion operator int() const { return 7; } // explicit conversion explicit operator int*() const { return nullptr; } // Error: array operator not allowed in conversion-type-id // operator int(*)[3]() const { return nullptr; } using arr_t = int[3]; operator arr_t*() const { return nullptr; } // OK if done through typedef // operator arr_t () const; // Error: conversion to array not allowed in any case }; int main() { X x; int n = static_cast<int>(x); // OK: sets n to 7 int m = x; // OK: sets m to 7 int* p = static_cast<int*>(x); // OK: sets p to null // int* q = x; // Error: no implicit conversion int (*pa)[3] = x; // OK }
Explanation
User-defined conversion function is invoked in the second stage of the implicit conversion, which consists of zero or one converting constructor or zero or one user-defined conversion function.
If both conversion functions and converting constructors can be used to perform some user-defined conversion, the conversion functions and constructors are both considered by overload resolution in copy-initialization and reference-initialization contexts, but only the constructors are considered in direct-initialization contexts.
struct To { To() = default; To(const struct From&) {} // converting constructor }; struct From { operator To() const {return To();} // conversion function }; int main() { From f; To t1(f); // direct-initialization: calls the constructor // Note: if converting constructor is not available, implicit copy constructor // will be selected, and conversion function will be called to prepare its argument // To t2 = f; // copy-initialization: ambiguous // Note: if conversion function is from a non-const type, e.g. // From::operator To();, it will be selected instead of the ctor in this case To t3 = static_cast<To>(f); // direct-initialization: calls the constructor const To& r = f; // reference-initialization: ambiguous }
Conversion function to its own (possibly cv-qualified) class (or to a reference to it), to the base of its own class (or to a reference to it), and to the type void can be defined, but can not be executed as part of the conversion sequence, except, in some cases, through virtual dispatch:
struct D; struct B { virtual operator D() = 0; }; struct D : B { operator D() override { return D(); } }; int main() { D obj; D obj2 = obj; // does not call D::operator D() B& br = obj; D obj3 = br; // calls D::operator D() through virtual dispatch }
It can also be called using member function call syntax:
struct B {}; struct X : B { operator B&() { return *this; }; }; int main() { X x; B& b1 = x; // does not call X::operatorB&() B& b2 = static_cast<B&>(x); // does not call X::operatorB& B& b3 = x.operator B&(); // calls X::operatorB& }
When making an explicit call to the conversion function, conversion-type-id is greedy: it is the longest sequence of tokens that could possibly form a conversion-type-id (including attributes, if any)(since C++11):
& x.operator int * a; // error: parsed as & (x.operator int*) a, // not as & (x.operator int) * a operator int [[noreturn]] (); // error: noreturn attribute applied to a type
The placeholder auto can be used in conversion-type-id, indicating a deduced return type: struct X { operator int(); // OK operator auto() -> short; // error: trailing return type not part of syntax operator auto() const { return 10; } // OK: deduced return type operator decltype(auto)() const { return 10l; } // OK: deduced return type }; Note: a conversion function template is not allowed to have a deduced return type. |
(since C++14) |
Conversion functions can be inherited and can be virtual, but cannot be static. A conversion function in the derived class does not hide a conversion function in the base class unless they are converting to the same type.
Conversion function can be a template member function, for example, std::auto_ptr<T>::operator auto_ptr<Y>
. See member template and template argument deduction for applicable special rules.
Keywords
Defect reports
The following behavior-changing defect reports were applied retroactively to previously published C++ standards.
DR | Applied to | Behavior as published | Correct behavior |
---|---|---|---|
CWG 296 | C++98 | conversion functions could be static | they cannot be declared static |
CWG 2016 | C++98 | conversion functions could not specify return types, but the types are present in conversion-type-id |
return types cannot be specified in the declaration specifiers of conversion functions |
CWG 2175 | C++11 | it was unclear whether the [[noreturn]] in operator int [[noreturn]] (); is parsed as a part of noptr-declarator (of function declarator) or conversion-type-id |
it is parsed as a part of conversion-type-id |