LINUX GAZETTE
[ Prev ][ Table of Contents ][ Front Page ][ Talkback ][ FAQ ][ Next ]

"Linux Gazette...making Linux just a little more fun!"


Creating Reusable Software Libraries

By Rob Tougher


1. Introduction
2. Making It Easy To Use
2.1 Keeping It Simple
2.2 Being Consistent
2.3 Making It Intuitive
3. Testing Thoroughly
4. Providing Detailed Error Information
5. Conclusion

1. Introduction

Software libraries provide functionality to application developers. They consist of reusable code that developers can utilize in their projects. Software libraries targeted for Linux are usually available in both binary and source code form.

A well-written software library:

This article describes the above principles of library creation, and gives examples in C++.

Is This Article For You?

Create software libraries only when you have to. Ask yourself these questions before proceeding:

  • Will anyone (including you) need functionality X in future applications?
  • If so, does a library implementing functionality X already exist?

If no one will need the functionality you are developing, or a software library implementing it already exists, don't create a new library.

2. Making It Easy To Use

The first step in creating a software library is designing its interface. Interfaces written in procedural languages, like C, contain functions. Interfaces written in object-oriented languages, like C++ and Python, can contain both functions and classes.

Remember this motto when designing your interface:

As a library designer, I am constantly faced with finding the right balance between functionality and ease of use. The above motto helps me resist adding too much functionality into my designs.

Stick with the following guidelines, and you'll be fine.

2.1 Keeping It Simple

The more complex a library, the harder it is to use.

I recently encountered a C++ library that consisted of one class. This class contained 150 methods. 150 methods! The designer was most likely a C veteran using C++ - the class acted like a C module. Because this class was so complex, it was very difficult to learn.

Avoid complexity in your designs, and your interfaces will be cleaner and easier to understand.

2.2 Being Consistent

Users learn consistent interfaces more easily. After learning the rules once, they feel confident in applying those rules across all classes and methods, even if they haven't used those classes and methods before.

One example I am guilty of involves public accessors for private variables. I sometimes do the following:

class point
{
public:
  int get_x() { return m_x; }
  int set_x ( int x ) { m_x = x; }

  int y() { return m_y; }

private:
  int m_x, m_y;
};

Do you see the problem here? For the m_x member, the public accessor is "get_x()", but for the m_y member, the public accessor is "y()". This inconsistency generates more work for the users - they have to look up the definition of each accessor before using it.

Here's another example of an inconsistent interface:

class DataBase
{
public:

  recordset get_recordset ( const std::string sql );
  void RunSQLQuery ( std::string query, std::string connection );

  std::string connectionString() { return m_connection_string; }

  long m_sError;

private:

  std::string m_connection_string;
};

Can you spot its problems? I can think of at least these items:

Here is a revised version that solves these problems:

class database
{
public:

  recordset get_recordset ( const std::string sql );
  void run_sql_query ( std::string sql );

  std::string connection_string() { return m_connection_string; }
  long error() { return m_error; }

private:

  std::string m_connection_string;
  long m_error;
};

Keep your interfaces as consistent as possible - your users will find them much easier to learn.

2.3 Making It Intuitive

Design an interface how you would expect it to work from a user's point of view - don't design it with the internal implementation in mind.

I find that the easiest way to design an intuitive interface is to write code that will use the library before actually writing the library. This forces me to think about the library from the user's standpoint.

Let's look at an example. I was recently considering writing an encryption library based on OpenSSL. Before thinking about the library design, I wrote some code snippets:

crypto::message msg ( "My data" );
crypto::key k ( "my key" );

// blowfish algorithm
msg.encrypt ( k, crypto::blowfish );
msg.decrypt ( k, crypto::blowfish ):

// rijndael algorithm
msg.encrypt ( k, crypto::rijndael );
msg.decrypt ( k, crypto::rijndael ):

This code helped me think about how I should design the interface - it put me in the user's shoes. If I decide to implement this library, my design will flow from these initial ideas.

3. Testing Thoroughly

A software library should work flawlessly. Well not flawlessly, but as close to flawless as possible. Users of a library need to know that the library is performing its tasks correctly.

I test my software libraries using automated scripts. For each library, I create a corresponding application that exercises all features of the library.

For example, say I decided to develop the encryption library I introduced in the previous section. My test application would look like the following:

#include "crypto.hpp"

int main ( int argc, int argv[] )
{
  //
  // 1. Encrypt, decrypt, and check
  //    message data.
  //
  crypto::message msg ( "Hello there" );
  crypto::key k ( "my key" );

  msg.encrypt ( k, crypto::blowfish );
  msg.decrypt ( k, crypto::blowfish );

  if ( msg.data() != "Hello there" )
    {
      // Error!
    }

  //
  // 2. Encrypt with one algorithm,
  //    decrypt with another, and check
  //    message data.
  //

  // etc....
}

I would occasionally run this application to make sure that my software library did not have any major errors.

4. Providing Detailed Error Information

Users need to know when a software library cannot perform its tasks correctly.

Software libraries written in C++ use exceptions to pass information to its users. Consider the following example:

#include <string>
#include <iostream>


class car
{
public:
  void accelerate() { throw error ( "Could not accelerate" ); }
};


class error
{
public:
  Error ( std::string text ) : m_text ( text ) {}
  std::string text() { return m_text; }
private:
  std::string m_text;
};


int main ( int argc, int argv[] )
{
  car my_car;

  try
    {
      my_car.accelerate();
    }
  catch ( error& e )
    {
      std::cout << e.text() << "\n";
    }
}

The car class uses the throw keyword to alert the caller to an erroneous situation. The caller catches this exception with the try and catch keywords, and deals with the problem.

5. Conclusion

In this article I explained the important principles of well-written software libraries. Hopefully I've explained everything clearly enough so that you can incorporate these principles into your own libraries.

Rob Tougher

Rob is a C++ software engineer in the New York City area.


Copyright © 2002, Rob Tougher.
Copying license http://www.linuxgazette.com/copying.html
Published in Issue 81 of Linux Gazette, August 2002

[ Prev ][ Table of Contents ][ Front Page ][ Talkback ][ FAQ ][ Next ]