The Artima Developer Community
Interviews | Discuss | Print | Email | First Page | Previous | Next
Sponsored Link

Abstraction and Efficiency
A Conversation with Bjarne Stroustrup, Part III
by Bill Venners
February 16, 2004

Page 1 of 3  >>


Bjarne Stroustrup talks with Bill Venners about raising the level of abstraction, why programming is understanding, how "oops happens," and the difference between premature and prudent optimization.

Bjarne Stroustrup is the designer and original implementer of C++. He is the author of numerous papers and several books, including The C++ Programming Language (Addison-Wesley, 1985-2000) and The Design and Evolution of C++ (Addison-Wesley, 1994). He took an active role in the creation of the ANSI/ISO standard for C++ and continues to work on the maintenance and revision of that standard. He is currently the College of Engineering Chair in Computer Science Professor at Texas A&M University.

On September 22, 2003, Bill Venners met with Bjarne Stroustrup at the JAOO conference in Aarhus, Denmark. In this interview, which is being published in multiple installments on, Stroustrup gives insights into C++ best practice.

Raising the Level of Abstraction

Bill Venners: I originally learned C++ from Borland's "World of C++" video. At the beginning of that video, you have a brief cameo appearance in which you state that what you were trying to do in C++ is raise the level of abstraction for programming.

Bjarne Stroustrup: That's right.

Bill Venners: What does raising the level of abstraction mean, and why is a high level of abstraction good?

Bjarne Stroustrup: A high level of abstraction is good, not just in C++, but in general. We want to deal with problems at the level we are thinking about those problems. When we do that, we have no gap between the way we understand problems and the way we implement their solutions. We can understand the next guy's code. We don't have to be the compiler.

Abstraction is a mechanism by which we understand things. Expressing a solution in terms of math, for instance, means we really did understand the problem. We didn't just hack a bunch of loops to try out special cases. There is always the temptation to provide just the solution to a particular problem. However, unless we try to generalize and see the problem as an example of a general class of problems, we may miss important parts of the solution to our particular problems and fail to find concepts and general solutions that could help us in the future. If somebody has a theory, such as a theory for matrix manipulation, you can just work at the level of those concepts and your code will become shorter, clearer, and more likely to be correct. There's less code to write, and it's easier to maintain.

I believe raising the level of abstraction is fundamental in all practical intellectual endeavors. I don't consider that a controversial statement, but people sometimes consider it controversial because they think code at a higher level abstraction is necessarily less efficient. For example, I got an email two days ago from somebody who had heard me give a talk in which I had been arguing for using a matrix library with proper linear algebra support. He said, "How much does using the matrix library cost more than using arrays directly, because I'm not sure I can afford it." To his great surprise, my answer was, "If you want the efficiency I pointed to, you cannot use the arrays directly."

The only code faster than the fastest code is no code. By abstracting to matrix manipulation operations, you give the compiler enough type information to enable it to eliminate many operations. If you were writing the code at the level of arrays, you would not eliminate those operations unless you were smarter than just about everybody. So you'd not only have to write ten times as much code if you used arrays instead of the matrix library, but you'd also have to accept a program that runs more slowly. By operating at the level where we can understand things, sometimes you can also operate at the level where we can analyze the code�we being compilers in the second case�and get better code.

My two favorite examples of this phenomenon are matrix times vector operations, where C++ on a good day can beat Fortran, and simple sorts, where C++ on a good day can beat C. The reason in both cases is you've expressed the program so directly, so cleanly, that the type system can help in generating better code�and you can't do that unless you have a suitable level of abstraction. You get this beautiful case where your code gets clearer, shorter, and faster. It doesn't happen all the time, of course, but it is so beautiful when it does.

Page 1 of 3  >>

Interviews | Discuss | Print | Email | First Page | Previous | Next

Sponsored Links

Copyright © 1996-2018 Artima, Inc. All Rights Reserved. - Privacy Policy - Terms of Use