Typically, projects expand, and what started out as a small project often becomes a major undertaking. Experienced C++ programmers thoroughly understand the C++ language features. However, as programs expand, different forces come into play; techniques used to create small programs in C++ don't work when tackling larger projects. In his book, Lakos distinguishes between physical and logical design issues.

 First, logical design is about language constructs (classes, functions, and inheritance). It revolves around questions like: Does a class need a copy constructor? Shall an operator be a member function or a free function? Shall a member function be const or non-const, virtual or non-virtual, static or non-static? Logical design does not address issues such as where to place a class definition. Unfortunately, as programs get larger, such physical design issues become more relevant.

 Second, physical design addresses issues of the physical entities of a system (files, directories, and libraries). For instance, making a member function of a class inline forces any client of the class to have seen not only the declaration of the function, but also its definition. What is affected is the degree of physical coupling between the class and its clients.

 Lakos devotes most of his book to physical design, explaining techniques for reducing the degree of interdependency among the physical entities of a C++ system. He discusses the adverse effect of cyclic dependencies, a symptom of which is that no part of the system can be understood independently. Cyclically dependent designs are impossible to be tested modularly, whereas hierarchical designs (for example, noncyclic designs) are easy to comprehend, test, and reuse. He explains the drawbacks of excessive link-time dependencies, which are a result of "fat" interfaces where one class depends on a tremendous amount of code. Link-time dependencies lead to oversized executables. He also discusses the effect of compile-time dependencies, such as situations where changing one innocent-looking header file results in recompilation of the entire system--a costly activity that in very large systems could take days. Reducing the degree of coupling therefore is vital to the success of large-scale development.

 Lakos's analysis and proposed solutions regarding physical design are very detailed (about 450 pages) and distinguish the book from other C++ publications. At times, this discussion may seem overdeveloped, but in multimillion lines-of-code programs these considerations indeed are vital. However, in an ideal world, we would expect the development environment (compiler, linker, run-time system) to take the burden of physical design issues off our shoulders. As C++ programmers, we want to concentrate on logical design. Lakos discusses logical design in a 200-page section of the book, and many of the topics discussed there can be found in other books, as well. Yet this section is recommended reading for C++ programmers who want to gain a deeper understanding of the tradeoffs of C++ language features.

  Large-Scale C++ Software Design is an advanced text that is unique in covering physical design issues relevant for very large systems. It is also valuable for smaller projects because the suggested techniques are good practice and scale well. It's a pity that the book covers only "classic" C++ and does not consider programming with templates and/or exceptions. Let us share with you what Lakos replied when we asked him what he thinks of the impact of these "new" language features on large-scale systems: "That is something my children or grandchildren will be writing books about. Right now, nobody has enough experience with templates and exceptions to understand the full range of consequences." Well, right he is! 

Angelika Langer develops and teaches classes on advanced C++, STL, multithreading, internationalization, and Java. She has served on the ANSI/ISO C++ Committee since 1993. Klaus Kreft is a software architect and consultant with more than a decade of experience in industrial software development. He works for Siemens Business Services in Germany. Langer and Kreft are working on a book about standard C++ iostreams and locales and are columnists for C++ Report .