6.12 — Using declarations and using directives

You’ve probably seen this program in a lot of textbooks and tutorials:

#include <iostream>

using namespace std;

int main()
{
    cout << "Hello world!";

    return 0;
}

Some older IDEs will also auto-populate new C++ projects with a similar program (so you can compile something immediately, rather than starting from a blank file).

If you see this, run. Your textbook, tutorial, or compiler are probably out of date. In this lesson, we’ll explore why.

A short history lesson

Back before C++ had support for namespaces, all of the names that are now in the std namespace were in the global namespace. This caused naming collisions between program identifiers and standard library identifiers. Programs that worked under one version of C++ might have a naming conflict with a newer version of C++.

In 1995, namespaces were standardized, and all of the functionality from the standard library was moved out of the global namespace and into namespace std. This change broke older code that was still using names without std::.

As anyone who has worked on a large codebase knows, any change to a codebase (no matter how trivial) risks breaking the program. Updating every name that was now moved into the std namespace to use the std:: prefix was a massive risk. A solution was requested.

Fast forward to today -- if you’re using the standard library a lot, typing std:: before everything you use from the standard library can become repetitive, and in some cases, can make your code harder to read.

C++ provides some solutions to both of these problems, in the form of using statements.

But first, let’s define two terms.

Qualified and unqualified names

A name can be either qualified or unqualified.

A qualified name is a name that includes an associated scope. Most often, names are qualified with a namespace using the scope resolution operator (::). For example:

std::cout // identifier cout is qualified by namespace std
::foo // identifier foo is qualified by the global namespace

For advanced readers

A name can also be qualified by a class name using the scope resolution operator (::), or by a class object using the member selection operators (. or ->). For example:

class C; // some class

C::s_member; // s_member is qualified by class C
obj.x; // x is qualified by class object obj
ptr->y; // y is qualified by pointer to class object ptr

An unqualified name is a name that does not include a scoping qualifier. For example, cout and x are unqualified names, as they do not include an associated scope.

Using declarations

One way to reduce the repetition of typing std:: over and over is to utilize a using declaration statement. A using declaration allows us to use an unqualified name (with no scope) as an alias for a qualified name.

Here’s our basic Hello world program, using a using declaration on line 5:

#include <iostream>

int main()
{
   using std::cout; // this using declaration tells the compiler that cout should resolve to std::cout
   cout << "Hello world!"; // so no std:: prefix is needed here!

   return 0;
} // the using declaration expires here

The using declaration using std::cout; tells the compiler that we’re going to be using the object cout from the std namespace. So whenever it sees cout, it will assume that we mean std::cout. If there’s a naming conflict between std::cout and some other use of cout, std::cout will be preferred. Therefore on line 6, we can type cout instead of std::cout.

This doesn’t save much effort in this trivial example, but if you are using cout many times inside of a function, a using declaration can make your code more readable. Note that you will need a separate using declaration for each name (e.g. one for std::cout, one for std::cin, etc…).

Although this method is less explicit than using the std:: prefix, it’s generally considered safe and acceptable (when used inside a function).

Using directives

Another way to simplify things is to use a using directive. Slightly simplified, a using directive imports all of the identifiers from a namespace into the scope of the using directive.

For advanced readers

For technical reasons, using directives do not actually import names into the current scope -- instead they import the names into an outer scope (more details about which outer scope is picked can be found here. However, these names are not accessible from the outer scope -- they are only accessible via unqualified (non-prefixed) lookup from the scope of the using directive (or a nested scope).

The practical effect is that (outside of some weird edge cases involving multiple using directives inside nested namespaces), using directives behave as if the names had been imported into the current scope. To keep things simple, we will proceed under the simplification that the names are imported into the current scope.

Here’s our Hello world program again, with a using directive on line 5:

#include <iostream>

int main()
{
   using namespace std; // this using directive tells the compiler to import all names from namespace std into the current namespace without qualification
   cout << "Hello world!"; // so no std:: prefix is needed here
   return 0;
}

The using directive using namespace std; tells the compiler to import all of the names from the std namespace into the current scope (in this case, of function main()). When we then use unqualified identifier cout, it will resolve to the imported std::cout.

Using directives are the solution that was provided for old pre-namespace codebases that used unqualified names for standard library functionality. Rather than having to manually update every unqualified name to a qualified name (which was risky), a single using directive (of using namespace std;) could be placed at the top of the each file, and all of the names that had been moved to the std namespace could still be used unqualified.

Problems with using directives (a.k.a. why you should avoid “using namespace std;”)

In modern C++, using directives generally offer little benefit (saving some typing) compared to the risk. Because using directives import all of the names from a namespace (potentially including lots of names you’ll never use), the possibility for naming collisions to occur increases significantly (especially if you import the std namespace).

For illustrative purposes, let’s take a look at an example where using directives cause ambiguity:

#include <iostream>

namespace a
{
	int x{ 10 };
}

namespace b
{
	int x{ 20 };
}

int main()
{
	using namespace a;
	using namespace b;

	std::cout << x << '\n';

	return 0;
}

In the above example, the compiler is unable to determine whether the x in main refers to a::x or b::x. In this case, it will fail to compile with an “ambiguous symbol” error. We could resolve this by removing one of the using statements, employing a using declaration instead, or qualifying x with an explicit scope qualifier (a:: or b::).

Here’s another more subtle example:

#include <iostream> // imports the declaration of std::cout

int cout() // declares our own "cout" function
{
    return 5;
}

int main()
{
    using namespace std; // makes std::cout accessible as "cout"
    cout << "Hello, world!"; // uh oh!  Which cout do we want here?  The one in the std namespace or the one we defined above?

    return 0;
}

In the above example, the compiler is unable to determine whether our use of cout means std::cout or the cout function we’ve defined, and again will fail to compile with an “ambiguous symbol” error. Although this example is trivial, if we had explicitly prefixed std::cout like this:

    std::cout << "Hello, world!"; // tell the compiler we mean std::cout

or used a using declaration instead of a using directive:

    using std::cout; // tell the compiler that cout means std::cout
    cout << "Hello, world!"; // so this means std::cout

then our program wouldn’t have any issues in the first place. And while you’re probably not likely to write a function named “cout”, there are hundreds, if not thousands, of other names in the std namespace just waiting to collide with your names. “count”, “min”, “max”, “search”, “sort”, just to name a few.

Even if a using directive does not cause naming collisions today, it makes your code more vulnerable to future collisions. For example, if your code includes a using directive for some library that is then updated, all of the new names introduced in the updated library are now candidates for naming collisions with your existing code.

There is a more insidious problem that can occur as well. The updated library may introduce a function that not only has the same name, but is actually a better match for some function call. In such a case, the compiler may decide to prefer the new function instead, and the behavior of your program will change unexpectedly.

Consider the following program:

foolib.h (part of some third-party library):

namespace foo
{
    // pretend there is some useful code that we use here
}

main.cpp:

#include <iostream>
#include <foolib.h> // a third-party library, thus angled brackets used

int someFcn(double)
{
    return 1;
}

int main()
{
    using namespace foo; // Because we're lazy and want to access foo:: qualified names without typing the foo:: prefix
    std::cout << someFcn(0); // The literal 0 should be 0.0, but this is an easy mistake to make

    return 0;
}

This program runs and prints 1.

Now, let’s say we update the foolib library, which includes an updated foolib.h. Our program now looks like this:

foolib.h (part of some third-party library):

namespace foo
{
    // newly introduced function
    int someFcn(int)
    {
        return 2;
    }

    // pretend there is some useful code that we use here
}

main.cpp:

#include <iostream>
#include <foolib.h>

int someFcn(double)
{
    return 1;
}

int main()
{
    using namespace foo; // Because we're lazy and want to access foo:: qualified names without typing the foo:: prefix
    std::cout << someFcn(0); // The literal 0 should be 0.0, but this is an easy mistake to make

    return 0;
}

Our main.cpp file hasn’t changed at all, but this program now runs and prints 2!

When the compiler encounters a function call, it has to determine what function definition it should match the function call with. In selecting a function from a set of potentially matching functions, it will prefer a function that requires no argument conversions over a function that requires argument conversions. Because the literal 0 is an integer, C++ will prefer to match someFcn(0) with the newly introduced someFcn(int) (no conversions) over someFcn(double) (requires a conversion from int to double). That causes an unexpected change to our program results.

This would not have happened if we’d used a using declaration or explicit scope qualifier.

Finally, the lack of explicit scope prefixes make it harder for a reader to tell what functions are part of a library and what’s part of your program. For example, if we use a using directive:

using namespace ns;

int main()
{
    foo(); // is this foo a user-defined function, or part of the ns library?
}

It’s unclear whether the call to foo() is actually a call to ns::foo() or to a foo() that is a user-defined function. Modern IDEs should be able to disambiguate this for you when you hover over a name, but having to hover over each name just to see where it comes from is tedious.

Without the using directive, it’s much clearer:

int main()
{
    ns::foo(); // clearly part of the ns library
    foo(); // likely a user-defined function
}

In this version, the call to ns::foo() is clearly a library call. The call to plain foo() is probably a call to a user-defined function (some libraries, including certain standard library headers, do put names into the global namespace, so it’s not a guarantee).

The scope of using declarations and directives

If a using declaration or using directive is used within a block, the names are applicable to just that block (it follows normal block scoping rules). This is a good thing, as it reduces the chances for naming collisions to occur to just within that block.

If a using declaration or using directive is used in the global namespace, the names are applicable to the entire rest of the file (they have file scope).

Cancelling or replacing a using statement

Once a using statement has been declared, there’s no way to cancel or replace it with a different using statement within the scope in which it was declared.

int main()
{
    using namespace foo;

    // there's no way to cancel the "using namespace foo" here!
    // there's also no way to replace "using namespace foo" with a different using statement

    return 0;
} // using namespace foo ends here

The best you can do is intentionally limit the scope of the using statement from the outset using the block scoping rules.

int main()
{
    {
        using namespace foo;
        // calls to foo:: stuff here
    } // using namespace foo expires
 
    {
        using namespace Goo;
        // calls to Goo:: stuff here
    } // using namespace Goo expires

    return 0;
}

Of course, all of this headache can be avoided by explicitly using the scope resolution operator (::) in the first place.

Best practices for using statements

Avoid using directives (particularly using namespace std;), except in specific circumstances. Using declarations are generally considered safe to use inside blocks. Limit their use in the global namespace of a code file, and never use them in the global namespace of a header file.

Best practice

Prefer explicit namespaces over using statements. Avoid using directives whenever possible. Using declarations are okay to use inside blocks.

Related content

The using keyword is also used to define type aliases, which are unrelated to using statements. We cover type aliases in lesson 8.6 -- Typedefs and type aliases.

guest
Your email address will not be displayed
Avatars from https://gravatar.com/ are connected to your provided email address.
Notify me about replies:  
168 Comments
Newest
Oldest Most Voted
Inline Feedbacks
View all comments