Morgan Stanley | Columbia University | Texas A&M University

home | C++ | FAQ | technical FAQ | C++11 FAQ | publications | WG21 papers | TC++PL | Tour++ | Programming | D&E | bio | interviews | videos | applications | glossary | compilers

Interview of Bjarne Stroustrup by "Developpeur Reference"

Interviewer Pierre Tran (tran@idg.fr ).

1. The ANSI/ISO standardization of C++ was a long process which ended only in 1998. Don't you regret that it wasn't done more quickly ?

Do you think that it had slowed down the penetration of C++ (for example in education, many continue to teach C language and claiming that C++ is not yet standardized, its an infuriating situation!)?

Do you think that it had slowed down the evolution of C++ (better evolution of standard libraries, taking care about distributed computing...)?

Bjarne Stroustrup: Of course it would have been nice if the C++ standard had been completed sooner, but the C++ standards process didn't actually take much longer than other standards. Formal standardization takes longer than most people think because the needs of so many people and organizations need to be met.

Many people taught - and some continue to teach - C++ as either a very low level language with a focus on features shared with C, or as a language for expressing class hierarchies. Both approaches fail to emphasize C++'s greatest strengths. Worse: such approaches often spend so much time on parts of C++ that are not very supportive of programmers that they fail to tech facilities and techniques critical to effective use of C++. The standard library containers and algorithms and the use of exceptions in resource management are examples of key topics that are often neglected, or wrongly considered advanced.

I think that such failures of teaching has done more harm to C++ use than the lengthy ISO standardization process. Clearly, it would have been much, much better if we had had the standard library ready for use in 1985, but we didn't and then we simply didn't know how to design and implement something that general, elegant, and efficient.

It is indeed very sad to hear people claiming that C++ isn't standardized. The ISO standard was ratified in 1998 and no significant new features have been added to the language or the standard library since 1996. In computing, that's a long time.

And as for the speed of evolution. That's limited by our understanding of problems and - even more so - by our ability to teach good use of the language. I don't think the language could or should evolve much faster than it does. Experimental languages can evolve rapidly because they have few users, mainstream languages - such as C++ - cannot move faster that their user community and must emphasize stability.

2. For the next versions of C++, you seem to prefer to expand the standard libraries to an evolution of the syntax of the language. Can you explain you position ?

Which libraries would you like to see standardize : multi-threading, distributed computing? Do you want to go as far as Java and provide a standard library for GUI?

Bjarne Stroustrup: It is my proposal that the work on the next revision of the ISO C++ standard focus on the standard library. The language is already powerful enough to express most of what we need to express and stability is of major importance. Consequently, I propose that we be conservative and cautious with language extensions, focusing on generalizations and of things that help teaching rather than on major new features. On the other hand, we should be aggressive and opportunistic when it comes to standard library extensions.

I think that libraries for making C++ better as a tool for systems development will be key: resource management, threading, physical distribution, TCP/IP, etc. are obvious candidates. We'll almost certainly also see useful facilities such as hash_maps and pattern matching. I fear that GUI is too complex and too controversial topic for the standards committee. The committee consists of volunteers and we don't have the resources to build a major GUI library. Also, a standards committee can't compete with commercial (and non-commercial) vendors. It has to try to serve the complete community.

3. Linux an Open Source communities did not really welcome for C++, most of developments are still done in C. Some claim a performance advantage for the C language (as for the Linux kernel), some reinvent the wheel reimplementing an object layer in C. What do you think?

Do you consider the success of KDE (written in C++) and the speed of his development, against Gnome (written in C) which stagnates, as a demonstration of the superiority of C++? have you other examples?

Bjarne Stroustrup: I think that reinventing the wheel is silly and that use of C too often shows ignorance of what C++ can do for systems builders. It is a sign of the immaturity of our field that people resists adopting more advanced tools and prefer to spend their energies reinventing things using primitive tools rather than taking the time to learn more powerful ones.

4. In mind of many programmers, especially Java programmers, C++ remains an object oriented version of C language. What could you say to convince them that C++ is more than that?

Bjarne Stroustrup: It is hard to convince people who do not want to be convinced.

	#include<string>
	#include<vector>
	#include<iostream>
	#include<algorithm>
	using namespace std;

	int main()
	{
		vector<string> v;
		string s;
		while (cin>>s) v.push_back(s); // read a file of words
		sort(v.begin(),v.end());		// sort the words
		ostream_iterator<string> os(cout,"\n");
		unique_copy(v.begin(),v.end(),os);	// output unique words
	}

Write that in C and compare. Be sure not to introduce buffer overflows or memory leaks.

5. In response to a question about C#, you said that it's a proprietary language on a a proprietary platform. But C# was standardized by ECMA and goes to ISO standardization and you can find some implementation on other platform (Mono on Linux). If tomorrow C# pass the ISO certification, do you change your mind ?

Bjarne Stroustrup: Probably not, and I consider that unlikely. I'd look carefully at the standardization process to see if there had been real input from many interested parties and that the evolution of the language wasn't in the hands of a single company. There is more to proper standardization than producing a piece of paper.

6. C# as Java take the way of single inheritance unlike C++ which implements multiple inheritance. Do you think that multiple inheritance is always the best solution ?

Bjarne Stroustrup: Multiple inheritance is not always the best solution, but sometimes the best solution to a problem involves multiple inheritance. Please note that even Java includes a limited form of multiple inheritance: inheritance of interfaces. I'm very comfortable with multiple inheritance and I know many example that - in my opinion - cannot be done elegantly without multiple inheritance. My book The C++ Programming Language contains more than a dozen such examples. One key technique is to separately inherit an interface (typically an abstract class) and a partial implementation.

You can always re-write an example using multiple inheritance into on the uses single inheritance only (by using forwarding functions). However, the result is often an example that is longer, reflect the design less directly, and is harder to maintain. Note that you can also rewrite every example using single inheritance to an example using no inheritance using the same technique and with the same negative impact on code clarity. A language that does not support multiple inheritance is simply less expressive than one that supports multiple inheritance and thereby forces the programmer to occasionally complicate code.

7. These days people are talking much more about frameworks rather languages, for example J2EE for Java where language is strongly bound to platform, or .NET where Microsoft advances the framework to the detriment of languages. What do you think about this approach? Do C++ need a framework for its evolution?

Bjarne Stroustrup: People talk a lot about frameworks, but history is littered with frameworks that didn't live up to their expectations. I have seen successful frameworks, but they were generally limited in scope. I'm skeptical of "universal" frameworks, and even more so when such frameworks are products of a platform vendor competing with similar frameworks from other vendors. As a user, I prefer to maintain my independence from vendors as far as possible.

I'd like to seen libraries providing cleaner and more general access to frameworks - as opposed to languages intimately tied to a single framework.

8. In an interview with C/C++ Journal, you said that one of your goals for C++ is to increase the level of abstraction. Was it always your objective ?

Can you really implement a strong level of abstraction without impacting the performance (Java made a lot of concessions for example)?

Which mechanisms would you like to see integrated in C++ (garbage collector... )? Which mechanisms would you like to see improved?

Bjarne Stroustrup: It was always my aim to make it possible to express the intent of the programmer more clearly and directly in code, so yes, increasing the level of abstraction was always my goals for C++.

Yes, you can significantly increase the level of abstraction without sacrificing efficiency. The key is a flexible and extensible static (compile time) type system. The STL part of the standard library is a good example. For example, a vector is defined so that it can take elements of any type (both built-in and user-defined) without overheads. To compare with Java, note that a C++ vector holds values of a user-defined type, rather than references to objects of user defined types. This saves both in memory (you don't have to have memory for references or memory management information for individual elements on the heap) and in access costs (no cost of indirection through a reference and no run-time type checking of elements extracted from a vector). The lack of a need for run-time type checking (casting) also lead to significantly cleaner and shorter code. Another example is the standard algorithms (such as sort), which can be applied to any suitable container and parameterized with a comparison criteria, yet are significantly faster than their C counterparts (such as qsort()).

I'd like to see the C++ standards committee explicitly acknowledge that garbage collection is a valid implementation technique for C++, but I don't want to make the C++ semantics dependent on a garbage collector. There are application areas (such as device drivers and some other kernel code) for which garbage collection isn't suitable, and if you want garbage collection in C++ you can use one of the existing garbage collectors. The work very well for the kinds of applications for which automatic garbage collection is a reasonable technique.

9. You describe C++ as a multiple paradigms language, mainly object oriented and generic. Which are the next paradigms could be find their place in a next version of C++ ?

Bjarne Stroustrup: I think that "paradigm" is an overused word, and I prefer the less pretentious "programming style". Also, one should not forget the use of simple free-standing classes (i.e. that are not part of a hierarchy). They are crucial for the flexibility and efficiency of many modern C++ techniques and are examples of classical data abstraction.

I don't think that C++ will support a new paradigm anytime soon, but maybe I'm too conservative in what I call a paradigm. I hope that the next C++ standard will support distributed programming, but that it will do so primarily through the standard library.

10. Have you some plan to integrate Aspect Oriented programming?

Bjarne Stroustrup: I don't. I'm still not sure exactly how "Aspect-Oriented Programming" fits with my guiding principles of representing independent concepts independently in code and combining them freely (and only) when needed. Nor am I sure how general and generally useful the notion of aspect-oriented programming is. Please remember that adding new features to a large and widely-used programming language is something you do only after considering alternatives.

Please note that many frequently asked questions about C++ can be found on my home pages together with answers. There, you can also find many articles and some useful C++ links.

Morgan Stanley | Columbia University | Texas A&M University

home | C++ | FAQ | technical FAQ | C++11 FAQ | publications | WG21 papers | TC++PL | Tour++ | Programming | D&E | bio | interviews | videos | applications | glossary | compilers