Read an Excerpt
API Design for C++
By Martin Reddy
MORGAN KAUFMANN
Copyright © 2011 Elsevier, Inc.
All right reserved.
ISBN: 978-0-12-385004-1
Chapter One
Introduction
1.1 WHAT ARE APPLICATION PROGRAMMING INTERFACES?
An Application Programming Interface (API) provides an abstraction for a problem and specifies how clients should interact with software components that implement a solution to that problem. The components themselves are typically distributed as a software library, allowing them to be used in multiple applications. In essence, APIs define reusable building blocks that allow modular pieces of functionality to be incorporated into end-user applications.
An API can be written for yourself, for other engineers in your organization, or for the development community at large. It can be as small as a single function or involve hundreds of classes, methods, free functions, data types, enumerations, and constants. Its implementation can be proprietary or open source. The important underlying concept is that an API is a well-defined interface that provides a specific service to other pieces of software.
A modern application is typically built on top of many APIs, where some of these can also depend on further APIs. This is illustrated in Figure 1.1, which shows an example application that depends directly on the API for three libraries (1–3), where two of those APIs depend on the API for a further two libraries (4 and 5). For instance, an image viewing application may use an API for loading GIF images, and that API may itself be built upon a lower-level API for compressing and decompressing data.
API development is ubiquitous in modern software development. Its purpose is to provide a logical interface to the functionality of a component while also hiding any implementation details. For example, our API for loading GIF images may simply provide a LoadImage() method that accepts a filename and returns a 2D array of pixels. All of the file format and data compression details are hidden behind this simple interface. This concept is also illustrated in Figure 1.1, where client code only accesses an API via its public interface, shown as the dark section at the top of each box.
1.1.1 Contracts and Contractors
As an analogy, consider the task of building your own home. If you were to build a house entirely on your own, you would need to possess a thorough understanding of architecture, plumbing, electronics, carpentry, masonry, and many other trades. You would also need to perform every task yourself and keep track of the minutest of details for every aspect of the project, such as whether you have enough wood for your floorboards or whether you have the right fasteners to fit the screws that you have. Finally, because you are the only person working on the project, you can only perform a single task at any point in time and hence the total time to complete the project could be very large.
An alternative strategy is to hire professional contractors to perform key tasks for you (Figure 1.2). You could hire an architect to design the plans for the house, a carpenter for all of your woodwork needs, a plumber to install the water pipes and sewage system for your house, and an electrician to set up the power systems. Taking this approach, you negotiate a contract with each of your contractors—telling them what work you want done and agreeing upon a price—they then perform that work for you. If you're lucky, maybe you even have a good friend who is a contractor and he offers you his services for free. With this strategy, you are freed from the need to know everything about all aspects of building a house and instead you can take a higher-level supervisory role to select the best contractors for your purpose and ensure that the work of each individual contractor is assembled together to produce the vision of your ideal home.
The analogy to APIs is probably obvious: the house that you are building equates to a software program that you want to write, and the contractors provide APIs that abstract each of the tasks you need to perform and hide the implementation details of the work involved. Your task then resolves to selecting the appropriate APIs for your application and integrating them into your software. The analogy of having skilled friends that provide contracting services for free is meant to represent the use of freely available open source libraries in contrast to commercial libraries that require a licensing fee to use in your software. The analogy could even be extended by having some of the contractors employing subcontractors, which corresponds to certain APIs depending on other APIs to perform their task.
The contractor analogy is a common one in object-oriented programming. Early practitioners in the field talked about an object defining a binding contract for the services or behavior that it provides. An object then implements those services when asked to by a client program, potentially by subcontracting some of the work out to other objects behind the scenes (Meyer, 1987; Snyder, 1986).
1.1.2 APIs in C++
Strictly speaking, an API is simply a description of how to interact with a component. That is, it provides an abstraction and a functional specification for a component. In fact, many software engineers prefer to expand the acronym API as Abstract Programming Interface instead of Application Programming Interface.
In C++, this is embodied as one or more header (.h) files plus supporting documentation files. An implementation for a given API is often represented as a library file that can be linked into end-user applications. This can be either a static library, such as a .lib file on Windows or .a on Mac OS X and Linux, or a dynamic library such as a .dll file on Windows, .dylib on Mac, or .so on Linux.
A C++ API will therefore generally include the following elements:
1. Headers: A collection of .h header files that define the interface and allow client code to be compiled against that interface. Open source APIs also include the source code (.cpp files) for the API implementation.
2. Libraries: One or more static or dynamic library files that provide an implementation for the API. Clients can link their code against these library files in order to add the functionality to their applications.
3. Documentation: Overview information that describes how to use the API, often including automatically generated documentation for all of the classes and functions in the API.
As an example of a well-known API, Microsoft's Windows API (often referred to as the Win32 API) is a collection of C functions, data types, and constants that enable programmers to write applications that run on the Windows platform. This includes functions for file handling, process and thread management, creating graphical user interfaces, talking to networks, and so on.
The Win32 API is an example of plain C API rather than a C++ API. While you can use a C API directly from a C++ program, a good example of a specific C++ API is the Standard Template Library (STL). The STL contains a set of container classes, iterators for navigating over the elements in those containers, and various algorithms that act on those containers (Josuttis, 1999). For instance, the collection of algorithms includes high-level operations such as std::search(), std::reverse(), std::sort(), and std::set_intersection(). The STL therefore presents a logical interface to the task of manipulating collections of elements, without exposing any of the internal details for how each algorithm is implemented.
TIP
An API is a logical interface to a software component that hides the internal details required to implement it.
1.2 WHAT'S DIFFERENT ABOUT API DESIGN?
Interfaces are the most important code that a developer writes. That's because problems in an interface are far more costly to fix than problems in the associated implementation code. As a result, the process of developing shared APIs demands more attention than standard application or Graphical User Interface (GUI) development. Of course, both should involve best design practices; however, in the case of API development, these are absolutely critical to its success. Specifically, some of the key differentiating factors of API development include the following.
• An API is an interface designed for developers, in much the same way that a GUI is an interface designed for end users. In fact, it's been said that an API is a user interface for programmers (Arnold, 2005). As such, your API could be used by thousands of developers around the world, and it will undoubtedly be used in ways that you never intended (Tulach, 2008). You must anticipate this in your design. A well-designed API can be your organization's biggest asset. Conversely, a poor API can create a support nightmare and even turn your users toward your competitors (Bloch, 2005), just as a buggy or difficult-to-use GUI may force an end user to switch to a different application.
• Multiple applications can share the same API. Figure 1.1 showed that a single application can be composed of multiple APIs. However, any one of those APIs could also be reused in several other applications. This means that while problems in the code for any given application will only affect that one application, errors in an API can affect all of the applications that depend on that functionality.
• You must strive for backward compatibility whenever you change an API. If you make an incompatible change to your interface, your clients' code may fail to compile, or worse their code could compile but behave differently or crash intermittently. Imagine the confusion and chaos that would arise if the signature of the printf() function in the standard C library was different for different compilers or platforms. The simple "Hello World" program may not look so simple any more:
#include <stdio.h> #ifdef _WIN32 #include <windows.h> #endif #ifdef __cplusplus #include <iostream> #endif
int main(int, char *argv) { #if __STRICT_ANSI__ printf("Hello World\n"); #elif defined(_WIN32) PrintWithFormat("Hello World\n"); #elif defined(__PRINTF_DEPRECATED__) fprintf(stdout, "Hello World\n"); #elif defined(__PRINTF_VECTOR__) const char *lines[2] 1/4 {"Hello World", NULL}; printf(lines); #elif defined(__cplusplus) std::cout << "Hello World" << std::endl; #else #error No terminal output API found #endif return 0; }
This may seem like a contrived example, but it's actually not that extreme. Take a look at the standard header files that come with your compiler and you will find declarations that are just as convoluted and inscrutable, or perhaps worse.
• Due to the backward compatibility requirement, it is critical to have a change control process in place. During the normal development process, many developers may fix bugs or add new features to an API. Some of these developers may be junior engineers who do not fully understand all of the aspects of good API design. As a result, it is important to hold an API review before releasing a new version of the API. This involves one or more senior engineers checking that all changes to the interface are acceptable, have been made for a valid reason, and are implemented in the best way to maintain backward compatibility. Many open source APIs also enforce a change request process to gain approval for a change before it is added to the source code.
• APIs tend to live for a long time. There can be a large upfront cost to produce a good API because of the extra overhead of planning, design, versioning, and review that is necessary. However, if done well, the long-term cost can be substantially mitigated because you have the ability to make radical changes and improvements to your software without disrupting your clients. That is, your development velocity can be greater due to the increased flexibility that the API affords you.
• The need for good documentation is paramount when writing an API, particularly if you do not provide the source code for your implementation. Users can look at your header files to glean how to use it, but this does not define the behavior of the API, such as acceptable input values or error conditions. Well-written, consistent, and extensive documentation is therefore an imperative for any good API.
(Continues...)
Excerpted from API Design for C++ by Martin Reddy Copyright © 2011 by Elsevier, Inc.. Excerpted by permission of MORGAN KAUFMANN. All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
