Multithreading

Introduction

This chapter describes how to write multithreaded programs in C++ using the Boost Threads library written by William Kempf. Boost is a set of open source, peer-reviewed, portable, high-performance libraries ranging from simple data structures to a complex parsing framework. The Boost Threads library is a framework for multithreading. For more information on Boost, see www.boost.org.

Standard C++ contains no native support for multithreading, so it is not possible to write portable multithreaded code the same way you would write portable code that uses other standard library classes like string, vector, list, and so on. The Boost Threads library goes a long way toward making a standard, portable multithreading library though, and it is designed to minimize many common multithreading headaches.

Unlike the standard library or third-party libraries, however, using a multithreading library is not as easy as unzipping it into a directory, adding your #includes, and coding away. For all but trivial multithreaded applications, you must design carefully using proven patterns and known tactics to avoid bugs that are otherwise virtually guaranteed to happen. In a typical, single-threaded application, it is easy to find common programming errors: off-by-one loops, dereferencing a null or deleted pointer, loss of precision on floating-point conversions, and so on. Multithreaded programs are different. Not only is it tedious to keep track of what several threads are doing in your debugger, but multithreaded programs are nondeterministic, meaning that bugs may only show up under rare or complicated circumstances.

It is for this reason that this chapter should not be your introduction to multithreaded programming. If you have already done some programming with threads, but not with C++ or the Boost Threads library, this chapter will get you on your way. But describing the fundamentals of multithreaded programming is beyond the scope of this book. If you have never done any multithreaded programming before, then you may want to read an introductory book on multithreading, though such titles are scant because most programmers don't use threads (though they probably ought to).

Much of the Boost documentation and some of the following recipes discuss the classes using the concept/model idea. A concept is an abstract description of something, usually a class, and its behavior, without any assumptions about its implementation. Typically, this description includes construction and destruction behavior, and each of the methods, including their preconditions, parameters, and postconditions. For example, the concept of a Mutex is something that can be locked and unlocked by one thread at a time. A model is a concrete manifestation of a concept, such as the mutex class in the Boost Threads library. A refinement on a concept is a specialization of it, such as a ReadWriteMutex, which is a Mutex with some additional behavior.

Finally, threads are doing one of three things: working, waiting for something, or ready to go but not waiting for anything or doing any work. These states are called run, wait, and ready. These are the terms I will use in the following recipes.