Definitions and ODR (One Definition Rule)
Definitions are declarations that fully define the entity introduced by the declaration. Every declaration is a definition, except for the following:
- A function declaration without a function body:
int f(int); // declares, but doesn't define f
- Any declaration with an extern storage class specifier or with a language linkage specifier (such as extern "C") without an initializer:
extern const int a; // declares, but doesn't define a extern const int b = 1; // defines b
- Declaration of a non-inline(since C++17) static data member inside a class definition:
struct S { int n; // defines S::n static int i; // declares, but doesn't define S::i inline static int x; // defines S::x }; // defines S int S::i; // defines S::i
struct S { static constexpr int x = 42; // implicitly inline, defines S::x }; constexpr int S::x; // declares S::x, not a redefinition |
(since C++17) |
- Declaration of a class name (by forward declaration or by the use of the elaborated type specifier in another declaration):
struct S; // declares, but doesn't define S class Y f(class T p); // declares, but doesn't define Y and T (and also f and p)
enum Color : int; // declares, but doesn't define Color |
(since C++11) |
- Declaration of a template parameter:
template<typename T> // declares, but doesn't define T
- A parameter declaration in a function declaration that isn't a definition:
int f(int x); // declares, but doesn't define f and x int f(int x) // defines f and x { return x + a; }
- A typedef declaration:
typedef S S2; // declares, but doesn't define S2 (S may be incomplete)
using S2 = S; // declares, but doesn't define S2 (S may be incomplete) |
(since C++11) |
using N::d; // declares, but doesn't define d
|
(since C++17) |
|
(since C++11) |
- An empty declaration (does not define any entities)
- A using-directive (does not define any entities)
extern template f<int, char>; // declares, but doesn't define f<int, char> |
(since C++11) |
- An explicit specialization whose declaration is not a definition:
template<> struct A<int>; // declares, but doesn't define A<int>
An asm declaration does not define any entities, but it is classified as a definition.
Where necessary, the compiler may implicitly define the default constructor, copy constructor, move constructor, copy assignment operator, move assignment operator, and the destructor.
If the definition of any object results in an object of incomplete type or abstract class type, the program is ill-formed.
Contents |
[edit] One Definition Rule
Only one definition of any variable, function, class type, enumeration type, concept(since C++20) or template is allowed in any one translation unit (some of these may have multiple declarations, but only one definition is allowed).
One and only one definition of every non-inline function or variable that is odr-used (see below) is required to appear in the entire program (including any standard and user-defined libraries). The compiler is not required to diagnose this violation, but the behavior of the program that violates it is undefined.
For an inline function or inline variable(since C++17), a definition is required in every translation unit where it is odr-used.
For a class, a definition is required wherever the class is used in a way that requires it to be complete.
There can be more than one definition in a program of each of the following: class type, enumeration type, inline function, inline variable(since C++17), templated entity (template or member of template, but not full template specialization), as long as all of the following is true:
- each definition appears in a different translation unit
|
(since C++20) |
- each definition consists of the same sequence of tokens (typically, appears in the same header file)
- name lookup from within each definition finds the same entities (after overload-resolution), except that
- constants with internal or no linkage may refer to different objects as long as they are not odr-used and have the same values in every definition
|
(since C++11) |
- overloaded operators, including conversion, allocation, and deallocation functions refer to the same function from each definition (unless referring to one defined within the definition)
- corresponding entities have the same language linkage in each definition (e.g. the include file isn't inside an extern "C" block)
- if a
const
object is constant-initialized in any of the definitions, it is constant-initialized in each definition - the rules above apply to every default argument used in each definition
- if the definition is for a class with an implicitly-declared constructor, every translation unit where it is odr-used must call the same constructor for the base and members
|
(since C++20) |
- if the definition is for a template, then all these requirements apply to both names at the point of definition and dependent names at the point of instantiation
If all these requirements are satisfied, the program behaves as if there is only one definition in the entire program. Otherwise, the program is ill-formed, no diagnostic required.
Note: in C, there is no program-wide ODR for types, and even extern declarations of the same variable in different translation units may have different types as long as they are compatible. In C++, the source-code tokens used in declarations of the same type must be the same as described above: if one .cpp file defines struct S { int x; }; and the other .cpp file defines struct S { int y; };, the behavior of the program that links them together is undefined. This is usually resolved with unnamed namespaces.
[edit] ODR-use
Informally,
- an object is odr-used if its value is read (unless it is a compile time constant) or written, its address is taken, or a reference is bound to it,
- a reference is odr-used if it is used and its referent is not known at compile time,
- a function is odr-used if a function call to it is made or its address is taken.
If an object, a reference or a function is odr-used, its definition must exist somewhere in the program; a violation of that is usually a link-time error.
struct S { static const int x = 0; // static data member // a definition outside of class is required if it is odr-used }; const int& f(const int& r); int n = b ? (1, S::x) // S::x is not odr-used here : f(S::x); // S::x is odr-used here: a definition is required
Formally,
- applying lvalue-to-rvalue conversion to x yields a constant expression that doesn't invoke non-trivial functions
- either x is not an object (that is, x is a reference) or, if x is an object, it is one of the potential results of a larger expression e, where that larger expression is either a discarded-value expression or has the lvalue-to-rvalue conversion applied to it
struct S { static const int x = 1; }; // applying lvalue-to-rvalue conversion // to S::x yields a constant expression int f() { S::x; // discarded-value expression does not odr-use S::x return S::x; // expression where lvalue-to-rvalue conversion // applies does not odr-use S::x }
3) A structured binding is odr-used if it appears as a potentially-evaluated expression.
|
(since C++17) |
A set of potential results of an expression E is a (possibly empty) set of id-expressions that appear within E, combined as follows:
- If E is an id-expression, the expression E is its only potential result.
- If E is a subscript expression (E1[E2]) where one of the operands is an array, the potential results of that operand is included in the set.
- If E is a class member access expression of the form E1.E2 or E1.template E2 naming a non-static data member, the potential results of E1 is included in the set.
- If E is a class member access expression naming a static data member, the id-expression designating the data member is included in the set.
- If E is a pointer-to-member access expression of the form E1.*E2 or E1.*template E2 whose second operand is a constant expression, the potential results of E1 are included in the set.
- If E is an expression in parentheses ((E1)), the potential results of E1 are included in the set.
- If E is a glvalue conditional expression (E1 ? E2 : E3, where E2 and E3 are glvalues), the union of the potential results of E2 and E3 are both included in the set.
- If E is a comma expression (E1, E2), the potential results of E2 are in the set of potential results.
- Otherwise, the set is empty.
struct S { static const int a = 1; static const int b = 2; }; int f(bool x) { return x ? S::a : S::b; // x is a part of the subexpression "x" (to the left of ?), // which applies lvalue-to-rvalue conversion, but applying that conversion to x // does not yield a constant expression, so x is odr-used // S::a and S::b are lvalues, and carry over as "potential results" // to the result of the glvalue conditional // That result is then subject to lvalue-to-rvalue conversion requested // to copy-initialize the return value, therefore S::a and S::b are not odr-used }
- A function is odr-used if it is named by (see below) a potentially-evaluated expression or conversion.
- A virtual member function is odr-used if it is not a pure virtual member function (addresses of virtual member functions are required to construct the vtable).
- A non-placement allocation or deallocation function for a class is odr-used by the definition of a constructor of that class.
- A non-placement deallocation function for a class is odr-used by the definition of the destructor of that class, or by being selected by the lookup at the point of definition of a virtual destructor.
- An assignment operator in a class
T
that is a member or base of another classU
is odr-used by an implicitly-defined copy-assignment or move-assignment functions ofU
. - A constructor (including default constructors) for a class is odr-used by the initialization that selects it.
- A destructor for a class is odr-used if it is potentially invoked.
This section is incomplete Reason: list of all situations where odr-use makes a difference |
[edit] Naming a function
A function is named by an expression or conversion in following cases:
- A function whose name appears as an expression or conversion (including named function, overloaded operator, user-defined conversion, user-defined placement forms of operator new, non-default initialization) is named by that expression if it is selected by overload resolution, except when it is an unqualified pure virtual member function or a pointer-to-member to a pure virtual function.
- An allocation or deallocation function for a class is named by a new expression appearing in an expression.
- A deallocation function for a class is named by a delete expression appearing in an expression.
- A constructor selected to copy or move an object is considered to be named by the expression or conversion even if copy elision takes place. Using a prvalue in some contexts does not copy or move an object, see mandatory elision.(since C++17)
A potentially evaluated expression or conversion odr-uses a function if it names it.
A potentially constant evaluated expression or conversion that names a constexpr function makes it needed for constant evaluation, which triggers definition of a defaulted function or instantiation of a function template specialization, even if the expression is unevaluated. |
(since C++11) |
[edit] 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 261 | C++98 | a deallocation function for a polymorphic class might be odr-used even if there were no relevant new or delete expressions in the program |
supplemented the odr-use cases to cover constructors and destructors |
CWG 678 | C++98 | an entity could have definitions with different language linkages |
the behavior is undefined in this case |
CWG 1472 | C++98 | reference variables which satisfy the requirements for appearing in a constant expression were odr-used even if the lvalue-to-rvalue conversion is immediately applied |
they are not odr-used in this case |
CWG 1614 | C++98 | taking address of a pure virtual function odr-used it | the function is not odr-used |
CWG 1741 | C++98 | constant objects that are immediately lvalue-to-rvalue converted in potentially-evaluated expressions were odr-used |
they are not odr-used |
CWG 1926 | C++98 | array subscript expressions didn't propagate potential results | they propagate |
CWG 2242 | C++98 | it was unclear whether a const object that is onlyconstant-initialized in part of its definitions violates ODR |
ODR is not violated; the object is constant-initialized in this case |
CWG 2300 | C++11 | lambda expressions in different translation units could never have the same closure type |
the closure type can be the same under one definition rule |
CWG 2353 | C++98 | a static data member was not a potential result of a member access expression accessing it |
it is |
CWG 2433 | C++14 | a variable template could not have multiple definitions in a program | it can |
[edit] References
- C++23 standard (ISO/IEC 14882:2024):
- 6.3 One definition rule [basic.def.odr]
- C++20 standard (ISO/IEC 14882:2020):
- 6.3 One definition rule [basic.def.odr]
- C++17 standard (ISO/IEC 14882:2017):
- 6.2 One definition rule [basic.def.odr]
- C++14 standard (ISO/IEC 14882:2014):
- 3.2 One definition rule [basic.def.odr]
- C++11 standard (ISO/IEC 14882:2011):
- 3.2 One definition rule [basic.def.odr]
- C++03 standard (ISO/IEC 14882:2003):
- 3.2 One definition rule [basic.def.odr]
- C++98 standard (ISO/IEC 14882:1998):
- 3.2 One definition rule [basic.def.odr]