Volume Number: 14 (1998)
Issue Number: 7
Column Tag: From The Factory Floor

The New C++ Standard: Namespaces

by by Andreas Hommel, Howard Hinnant, and Dave Mark, ©1998 by Metrowerks, Inc., all rights reserved.

By the time you read this, the Final Draft International Standard for C++ should be finalized, approved, and made available for purchase (see last month's interview with Ron Liechty for specifics on where to go to get your copy). In this month's column, our old friend Andreas Hommel, along with new friend Howard Hinnant will tackle an important part of the new standard: C++ namespaces.

Andreas Hommel is currently the C/C++ front-end and 68K back-end architect at Metrowerks. After finishing his Master's degree in Computer Science, Andreas did some contract programming in the desktop publishing area and also published several games on the Macintosh and Amiga. He has been with Metrowerks for five years.

Andreas lives in a small country village about 20 kilometers north of Hamburg, with his wife, two Australian Shepherd dogs, two Arabian horses and one Quarter horse. When he is not coding, riding horses or walking his dogs, Andreas likes running, traveling, playing a good video game, and driving really fast on the Autobahn. He also likes cooking and fine red wines (California cabernets, in particular).

Howard Hinnant is a software engineer on the MSL team at Metrowerks, and is responsible for the C++ and EC++ libraries. Howard is a refugee from the aerospace industry where FORTRAN still rules. He has extensive experience in scientific computing including C++ implementations of linear algebra, finite difference and finite element solvers.

Dave: What are C++ namespaces?

Andreas: Namespaces are one of the newer ANSI C++ features. They allow a programmer to define new named or unnamed declarative regions. The original C++ (and ANSI C) only had one global namespace, but now it is possible to have many user defined namespaces.

For example:

namespace metro {
 int foo();
 int bar;
}

defines a namespace 'metro' and declares the namespace member function 'foo' and defines a namespace member variable 'bar'. You can define anything inside a namespace that can be defined in the global C++ namespace, even other nested namespaces. All these namespace members can then be used using qualified-ids.

For example:

int i = metro::foo() + metro::bar;

would call metro's member 'foo()'. So this is very similar to a C++ class definition. In fact we could have created something very similar using static class members:

class metro_class {
public:
 static int foo();
 static int bar;
};

int j = metro_class::foo() + metro_class::bar;

However, there are differences between a class and a namespace. You cannot create any instances of a namespace (ie a namespace variable or a pointer to a namespace) and all data and function namespace members behave like static class members (non-static namespace members wouldn't make any sense when you cannot have a namespace instance). So namespaces have more restrictions than classes. However, they have the advantage that you can split the definition of a namespace over several parts of one or more translation units (ie source and header files). So you can add new definitions later on.

For example

namespace metro {
 int baz();
}

adds the function 'baz' to our 'metro' namespace and this could be done in the same or any other source or header file.

Dave: So you have to use qualified-ids to access namespace members. Are there any mechanisms in C++ to simplify this?

Andreas: Yes, First, you don't have to use qualified-ids if you are accessing namespace members within the same or nested namespace. So you can do something like this:

namespace metro {
  int k = bar;  // no qualification necessary, uses metro::bar
}

or even:

int metro::baz()  // define metro::baz outside of namespace
{
  return foo();  // no qualification necessary, uses metro::foo
}

But, there are also language extensions that simplify the use of namespace members outside of the namespace. One is a 'using-declaration' that can be useful if you are using a particular namespace member over and over again. For example:

using metro::bar;  // using declaration
int m = bar;  // no qualification necessary, uses metro::bar

introduces metro's member bar to the current namespace which enables you to use metro::bar without any qualification.

The other major extension is called a 'using-directive' which will introduce all names defined in a namespace. For example:

using namespace metro;  // using-directive
int n = foo() + baz();  // no qualification necessary, uses 
                        // metro::foo and metro::baz

Dave: What is an unnamed namespace?

Andreas: An unnamed namespace like:

namespace { int o; }

behaves as if was replaced by:

namespace <unique> { }
using namespace <unique>;
namespace <unique> { int o; }

where <unique> is replaced with a translation unit specific identifier that is different from all other identifiers in a program. So it will be possible to access 'o' without any qualifications in the same translation unit:

void f() { o++; } // uses <unique>::o

but, not from any other translation unit. So this is very similar to:

static int o;
void f() { o++; }

The use of the 'static' keyword in namespace scope is actually deprecated in the current C++ draft.

Dave: Why would you want to use namespaces?

Andreas: Namespaces are very useful for libraries because they can be used to avoid name collisions. If you have a program that has a function 'foo' and you want to use a third party library that has a different function with the same name you would have to change your on program or the library to be able to use this library. If this library would have used its own namespace (e.g., 'metro' from the example above) you wouldn't have this problem. A good example for this is the std:: namespace that is used to implement the standard C++ library.

Dave: What did it take to get the libraries under namespace std?

Howard: Putting MSL (Metrowerks Standard Libraries) under namespace std took a lot more work than we had originally envisioned. There were many issues which needed sorting out, prioritizing, and dealing with within the time allowed. Without the aid of the entire MSL Team (headed by Vicki Scott) we would never have pulled it off so quickly. But it was more than just team work within the MSL Team that counted. Tight and rapid response between the MSL Team and the Compiler Team was crucial. There were many very subtle effects and interactions between the compiler's implementation of namespaces, and the library's use of namespaces. Andreas is great at getting to the heart of a problem, and providing a solution in an amazingly short amount of time.

Dave: What effect will namespaces have on legacy code?

Howard: Standard namespaces have quite a bit of backward compatibility built into them to ease the porting of namespace-ignorant code. We have implemented all of this compatibility, and where we thought was important, added even more backward compatibility.

Standard header names have become very particular, and very important. In previous releases, <iostream> and <iostream.h> were interchangeable. So were <cstdio> and <stdio.h>. This is no longer the case. As a general rule of thumb, namespace-ignorant code should use headers that end with the .h extension. If you use the extension-less headers, then your code should be "namespace std aware".

For example, the following HelloWorld will compile and run correctly:

#include <iostream.h>

int main()
{
 cout << "Hi\n";
}

But this will fail with a compiler error:

#include <iostream>

int main()
{
 cout << "Hi\n";
}

You can make the above HelloWorld work in one of two ways: You can provide a using declaration:

#include <iostream>

int main()
{
 using namespace std;
 cout << "Hi\n";
}

Or you can provide the full name of objects in the standard library:

#include <iostream>

int main()
{
 std::cout << "Hi\n";
}

The purpose of namespace std is to keep library names from conflicting with your own. With a name like cout, a conflict doesn't seem likely. But, consider the name vector or stack. Many users might want to use such names. Such users may even be unaware of their existence in the library. Now that these names are under a namespace, such conflicts are less likely.

Dave: Does namespace affect the C library?

Howard: The "C" library is also under namespace std when used from a C++ program. That is, printf's new name is std::printf. Remember, this is only when using the C++ compiler. C programs will continue to use just plain printf. Also note that if a C++ program includes <stdio.h> instead of <cstdio>, then printf is promoted to the global namespace. So again, just plain printf can be used.