What are inline namespaces for?

C++11 allows inline namespaces, all members of which are also automatically in the enclosing namespace. I cannot think of any useful application of this -- can somebody please give a brief, succinct example of a situation where an inline namespace is needed and where it is the most idiomatic solution?

(Also, it is not clear to me what happens when a namespace is declared inline in one but not all declarations, which may live in different files. Isn't this begging for trouble?)

Inline namespaces are a library versioning feature akin to symbol versioning, but implemented purely at the C++11 level (ie. cross-platform) instead of being a feature of a specific binary executable format (ie. platform-specific).

It is a mechanism by which a library author can make a nested namespace look and act as if all its declarations were in the surrounding namespace (inline namespaces can be nested, so "more-nested" names percolate up all the way to the first non-inline namespace and look and act as if their declarations were in any of the namespaces in between, too).

As an example, consider the STL implementation of vector. If we had inline namespaces from the beginning of C++, then in C++98 the header <vector> might have looked like this:

namespace std {

#if __cplusplus < 1997L // pre-standard C++
    inline
#endif

    namespace pre_cxx_1997 {
        template <class T> __vector_impl; // implementation class
        template <class T> // e.g. w/o allocator argument
        class vector : __vector_impl<T> { // private inheritance
            // ...
        };
    }
#if __cplusplus >= 1997L // C++98/03 or later
                         // (ifdef'ed out b/c it probably uses new language
                         // features that a pre-C++98 compiler would choke on)
#  if __cplusplus == 1997L // C++98/03
    inline
#  endif

    namespace cxx_1997 {

        // std::vector now has an allocator argument
        template <class T, class Alloc=std::allocator<T> >
        class vector : pre_cxx_1997::__vector_impl<T> { // the old impl is still good
            // ...
        };

        // and vector<bool> is special:
        template <class Alloc=std::allocator<bool> >
        class vector<bool> {
            // ...
        };

    };

#endif // C++98/03 or later

} // namespace std

Depending on the value of __cplusplus, either one or the other vector implementation is chosen. If your codebase was written in pre-C++98 times, and you find that the C++98 version of vector is causing trouble for you when you upgrade your compiler, "all" you have to do is to find the references to std::vector in your codebase and replace them by std::pre_cxx_1997::vector.

Come the next standard, and the STL vendor just repeats the procedure again, introducing a new namespace for std::vector with emplace_back support (which requires C++11) and inlining that one iff __cplusplus == 201103L.

OK, so why do I need a new language feature for this? I can already do the following to have the same effect, no?

namespace std {

    namespace pre_cxx_1997 {
        // ...
    }
#if __cplusplus < 1997L // pre-standard C++
    using namespace pre_cxx_1997;
#endif

#if __cplusplus >= 1997L // C++98/03 or later
                         // (ifdef'ed out b/c it probably uses new language
                         // features that a pre-C++98 compiler would choke on)

    namespace cxx_1997 {
        // ...
    };
#  if __cplusplus == 1997L // C++98/03
    using namespace cxx_1997;
#  endif

#endif // C++98/03 or later

} // namespace std

Depending on the value of __cplusplus, I get either one or the other of the implementations.

And you'd be almost correct.

Consider the following valid C++98 user code (it was permitted to fully specialize templates that live in namespace std in C++98 already):

// I don't trust my STL vendor to do this optimisation, so force these 
// specializations myself:
namespace std {
    template <>
    class vector<MyType> : my_special_vector<MyType> {
        // ...
    };
    template <>
    class vector<MyOtherType> : my_special_vector<MyOtherType> {
        // ...
    };
    // ...etc...
} // namespace std

This is perfectly valid code where the user supplies its own implementation of a vector for a set of type where she apparently knows a more efficient implementation than the one found in (her copy of) the STL.

But: When specializing a template, you need to do so in the namespace it was declared in. The Standard says that vector is declared in namespace std, so that's where the user rightfully expects to specialize the type.

This code works with a non-versioned namespace std, or with the C++11 inline namespace feature, but not with the versioning trick that used using namespace <nested>, because that exposes the implementation detail that the true namespace in which vector was defined was not std directly.

There are other holes by which you could detect the nested namespace (see comments below), but inline namespaces plug them all. And that's all there is to it. Immensely useful for the future, but AFAIK the Standard doesn't prescribe inline namespace names for its own standard library (I'd love to be proven wrong on this, though), so it can only be used for third-party libraries, not the standard itself (unless the compiler vendors agree on a naming scheme).

http://www.stroustrup.com/C++11FAQ.html#inline-namespace (a document written by and maintained by Bjarne Stroustrup, who you'd think should be aware of most motivations for most C++11 features.)

According to that, it is to allow versioning for backward-compatibility. You define multiple inner namespaces, and make the most recent one inline. Or anyway, the default one for people who don't care about versioning. I suppose the most recent one could be a future or cutting-edge version which is not yet default.

The example given is:

// file V99.h:
inline namespace V99 {
	void f(int);	// does something better than the V98 version
	void f(double);	// new feature
	// ...
}

// file V98.h:
namespace V98 {
	void f(int);	// does something
	// ...
}

// file Mine.h:
namespace Mine {
#include "V99.h"
#include "V98.h"
}

#include "Mine.h"
using namespace Mine;
// ...
V98::f(1);	// old version
V99::f(1);	// new version
f(1);		// default version

I don't immediately see why you don't put using namespace V99; inside namespace Mine, but I don't have to entirely understand the use-case in order to take Bjarne's word for it on the committee's motivation.

In addition to all the other answers.

Inline namespace can be used to encode ABI information or Version of the functions in the symbols. It is due to this reason they are used to provide backward ABI compatibility. Inline namespaces let you inject information into the mangled name (ABI) without altering the API because they affect linker symbol name only.

Consider this example:

Suppose you write a function Foo that takes a reference to an object say bar and returns nothing.

Say in main.cpp

struct bar;
void Foo(bar& ref);

If you check your symbol name for this file after compiling it into an object.

$ nm main.o
T__ Z1fooRK6bar 

> The linker symbol name may vary but it will surely encode the name of function and argument types somewhere.

Now, it could be that bar is defined as:

struct bar{
   int x;
#ifndef NDEBUG
   int y;
#endif
};

Depending upon Build type, bar can refer to two different types/layouts with same linker symbols.

To prevent such behavior we wrap our struct bar into an inline namespace, where depending upon the Build type the linker symbol of bar will be different.

So, we could write:

#ifndef NDEBUG
inline namespace rel { 
#else
inline namespace dbg {
#endif
struct bar{
   int x;
#ifndef NDEBUG
   int y;
#endif
};
}

Now if you look at the object file of each object you build one using release and other with debug flag. You will find that linker symbols include inline namespace name as well. In this case

$ nm rel.o
T__ ZROKfoo9relEbar
$ nm dbg.o
T__ ZROKfoo9dbgEbar

> Linker Symbol names may be different.

Notice presence of rel and dbg in the symbol names.

Now, if you try to link debug with release mode or vise-versa you will get a linker error as contrary to runtime error.

I actually discovered another use for inline namespaces.

With Qt, you get some extra, nice features using Q_ENUM_NS, which in turn requires that the enclosing namespace has a meta-object, which is declared with Q_NAMESPACE. However, in order for Q_ENUM_NS to work, there has to be a corresponding Q_NAMESPACE in the same file⁽¹⁾. And there can only be one, or you get duplicate definition errors. This, effectively, means that all of your enumerations have to be in the same header. Yuck.

Or... you can use inline namespaces. Hiding enumerations in an inline namespace causes the meta-objects to have different mangled names, while looking to users like the additional namespace doesn't exist⁽²⁾.

So, they're useful for splitting stuff into multiple sub-namespaces that all look like one namespace, if you need to do that for some reason. Of course, this is similar to writing using namespace inner in the outer namespace, but without the DRY violation of writing the name of the inner namespace twice.

---

1. It's actually worse than that; it has to be in the same set of braces.

2. Unless you try to access the meta-object without fully qualifying it, but the meta-object is hardly ever used directly.

So to sum up the main points, using namespace v99 and inline namespace were not the same, the former was a workaround to version libraries before a dedicated keyword (inline) was introduced in C++11 which fixed the problems of using using, whilst providing the same versioning functionality. Using using namespace used to cause problems with ADL (although ADL now appears to follow using directives), and out-of-line specialisation of a library class / function etc. by the user wouldn't work if done outside of the true namespace (whose name the user wouldn't and shouldn't know, i.e. the user would have to use B::abi_v2:: rather than just B:: for the specialisation to resolve).

//library code
namespace B { //library name the user knows
    namespace A { //ABI version the user doesn't know about
        template<class T> class myclass{int a;};
    }
    using namespace A; //pre inline-namespace versioning trick
} 

// user code
namespace B { //user thinks the library uses this namespace
    template<> class myclass<int> {};
}

This will show a static analysis warning first declaration of class template specialization of 'myclass' outside namespace 'A' is a C++11 extension [-Wc++11-extensions]. But if you make namespace A inline, then the compiler correctly resolves the specialisation. Although, with the C++11 extensions, the issue goes away.

Out-of-line definitions don't resolve when using using; they have to be declared in a nested/non-nested extension namespace block (which means the user needs to know the ABI version again, if for whatever reason they were permitted to provide their own implementation of a function).

#include <iostream>
namespace A {
    namespace B{
        int a;
        int func(int a);
        template<class T> class myclass{int a;};
        class C;
        extern int d;
    } 
    using namespace B;
} 
int A::d = 3; //No member named 'd' in namespace A
class A::C {int a;}; //no class named 'C' in namespace 'A' 
template<> class A::myclass<int> {}; // works; specialisation is not an out-of-line definition of a declaration
int A::func(int a){return a;}; //out-of-line definition of 'func' does not match any declaration in namespace 'A'
namespace A { int func(int a){return a;};} //works
int main() {
    A::a =1; // works; not an out-of-line definition
}

The issue goes away when making B inline.

The other functionality inline namespaces have is allowing the library writer to provide a transparent update to the library 1) without forcing the user to refactor code with the new namespace name and 2) preventing lack of verbosity and 3) providing abstraction of API-irrelevant details, whilst 4) giving the same beneficial linker diagnostics and behaviour that using a non-inline namespace would provide. Let's say you're using a library:

namespace library {
    inline namespace abi_v1 {
        class foo {
        } 
    }
}

It allows the user to call library::foo without needing to know or include the ABI version in the documentation, which looks cleaner. Using library::abiverison129389123::foo would look dirty.

When an update is made to foo, i.e. adding a new member to the class, it will not affect existing programs at the API level because they wont already be using the member AND the change in the inline namespace name will not change anything at the API level because library::foo will still work.

namespace library {
    inline namespace abi_v2 {
        class foo {
            //new member
        } 
    }
}

However, for programs that link with it, because the inline namespace name is mangled into symbol names like a regular namespace, the change will not be transparent to the linker. Therefore, if the application is not recompiled but is linked with a new version of the library, it will present a symbol abi_v1 not being found error, rather than it actually linking and then causing a mysterious logic error at runtime due to ABI incompatibility. Adding a new member will cause ABI compatibility due to the change in type definition, even if it doesn't affect the program at compile time (API level).

In this scenario:

namespace library {
    namespace abi_v1 {
        class foo {
        } 
    }

    inline namespace abi_v2 {
        class foo {
            //new member
        } 
    }
}

Like using 2 non-inline namespaces, it allows for a new version of the library to be linked without needing to recompile the application, because abi_v1 will be mangled in one of the global symbols and it will use the correct (old) type definition. Recompiling the application would however cause the references to resolve to library::abi_v2.

Using using namespace is less functional than using inline (in that out of line definitions don't resolve) but provides the same 4 advantages as above. But the real question is, why continue to use a workaround when there is now a dedicated keyword to do it. It's better practice, less verbose (have to change 1 line of code instead of 2) and makes the intention clear.

Inline namespaces can also be used to provide fine(r)-grained access to features/names within namespaces.

This is used in std::literals. The literals namespaces in std are all inline namespaces, so that:

• if you use using namespace std; somewhere, you also get access to all user defined literals in the std.
• but if you just need a set of udl in you local code, you can also do using namespace std::literals::string_literals; and you will just get the udl symbols defined in that namespace.

This seems a useful technique for symbols that you would want to access unqualified (udl, operators, etc.) where you can just bundle them up in an inline namespace so that you can do a specific using on just that (sub-) namespace instead of the namespace of the whole library.