You have no doubt heard a lot about the fact that Java is designed to be a safe language. But what do we mean by safe? Safe from what or whom? The security features that attract the most attention for Java are those features that make possible new types of dynamically portable software. Java provides several layers of protection from dangerously flawed code, as well as more mischievous things like viruses and trojan horses. In the next section, we'll take a look at how the Java virtual machine architecture assesses the safety of code before it's run, and how the Java class loader builds a wall around untrusted classes. These features provide the foundation for high-level security policies that allow or disallow various kinds of activities on an application-by-application basis.
In this section though, we'll look at some general features of the Java programming language. What is perhaps more important, although often overlooked in the security din, is the safety that Java provides by addressing common design and programming problems. Java is intended to be as safe as possible from the simple mistakes we make ourselves, as well as those we inherit from contractors and third-party software vendors. The goal with Java has been to keep the language simple, provide tools that have demonstrated their usefulness, and let users build more complicated facilities on top of the language when needed.
Java is parsimonious in its features; simplicity rules. Compared to C, Java uses few automatic type coercions, and the ones that remain are simple and well-defined. Unlike C++, Java doesn't allow programmer-defined operator overloading. The string concatenation operator + is the only system-defined, overloaded operator in Java. All methods in Java are like C++ virtual methods, so overridden methods are dynamically selected at run-time.
Java doesn't have a preprocessor, so it doesn't have macros, #define statements, or conditional source compilation. These constructs exist in other languages primarily to support platform dependencies, so in that sense they should not be needed in Java. Another common use for conditional compilation is for debugging purposes. Debugging code can be included directly in your Java source code by making it conditional on a constant (i.e., static and final) variable. The Java compiler is smart enough to remove this code when it determines that it won't be called.
Java provides a well-defined package structure for organizing class files. The package system allows the compiler to handle most of the functionality of make. The compiler also works with compiled Java classes because all type information is preserved; there is no need for header files. All of this means that Java code requires little context to read. Indeed, you may sometimes find it faster to look at the Java source code than to refer to class documentation.
Java replaces some features that have been troublesome in other languages. For example, Java supports only a single inheritance class hierarchy, but allows multiple inheritance of interfaces. An interface, like an abstract class in C++, specifies the behavior of an object without defining its implementation, a powerful mechanism borrowed from Objective C. It allows a class to implement the behavior of the interface, without needing to be a subclass of the interface. Interfaces in Java eliminate the need for multiple inheritance of classes, without causing the problems associated with multiple inheritance.
As you'll see in Chapter 4, The Java Language, Java is a simple, yet elegant, programming language.
One attribute of a language is the kind of type checking it uses. When we categorize a language as static or dynamic we are referring to the amount of information about variable types that is known at compile time versus what is determined while the application is running.
In a strictly statically typed language like C or C++, data types are etched in stone when the source code is compiled. The compiler benefits from having enough information to enforce usage rules, so that it can catch many kinds of errors before the code is executed. The code doesn't require run-time type checking, which is advantageous because it can be compiled to be small and fast. But statically typed languages are inflexible. They don't support high-level constructs like lists and collections as naturally as languages with dynamic type checking, and they make it impossible for an application to safely import new data types while it's running.
In contrast, a dynamic language like Smalltalk or Lisp has a run-time system that manages the types of objects and performs necessary type checking while an application is executing. These kinds of languages allow for more complex behavior, and are in many respects more powerful. However, they are also generally slower, less safe, and harder to debug.
The differences in languages have been likened to the differences between kinds of automobiles. Statically typed languages like C++ are analogous to a sports car--reasonably safe and fast--but useful only if you're driving on a nicely paved road. Highly dynamic languages like Smalltalk are more like an offroad vehicle: they afford you more freedom, but can be somewhat unwieldy. It can be fun (and sometimes faster) to go roaring through the back woods, but you might also get stuck in a ditch or mauled by bears.
 The credit for the car analogy goes to Marshall P. Cline, author of the C++ FAQ.
Another attribute of a language deals with when it binds method calls to their definitions. In an early-binding language like C or C++, the definitions of methods are normally bound at compile-time, unless the programmer specifies otherwise. Smalltalk, on the other hand, is a late-binding language because it locates the definitions of methods dynamically at run-time. Early-binding is important for performance reasons; an application can run without the overhead incurred by searching method tables at run-time. But late-binding is more flexible. It's also necessary in an object-oriented language, where a subclass can override methods in its superclass, and only the run-time system can determine which method to run.
Java provides some of the benefits of both C++ and Smalltalk; it's a statically typed, late-binding language. Every object in Java has a well-defined type that is known at compile time. This means the Java compiler can do the same kind of static type checking and usage analysis as C++. As a result, you can't assign an object to the wrong type of reference or call nonexistent methods within it. The Java compiler goes even further and prevents you from messing up and trying to use uninitialized variables.
However, Java is fully run-time typed as well. The Java run-time system keeps track of all objects and makes it possible to determine their types and relationships during execution. This means you can inspect an object at run-time to determine what it is. Unlike C or C++, casts from one type of object to another are checked by the run-time system, and it's even possible to use completely new kinds of dynamically loaded objects with a level of type safety.
Since Java is a late-binding language, all methods are like virtual methods in C++. This makes it possible for a subclass to override methods in its superclass. But Java also allows you to gain the performance benefits of early-binding by declaring (with the final modifier) that certain methods can't be overridden by subclassing, removing the need for run-time lookup.
Java carries all data-type and method-signature information with it from its source code to its compiled byte-code form. This means that Java classes can be developed incrementally. Your own Java classes can also be used safely with classes from other sources your compiler has never seen. In other words, you can write new code that references binary class files, without losing the type safety you gain from having the source code. The Java run-time system can load new classes while an application is running, thus providing the capabilities of a dynamic linker.
A common irritation with C++ is the "fragile base class" problem. In C++, the implementation of a base class can be effectively frozen by the fact that it has many derived classes; changing the base class may require recompilation of the derived classes. This is an especially difficult problem for developers of class libraries. Java avoids this problem by dynamically locating fields within classes. As long as a class maintains a valid form of its original structure, it can evolve without breaking other classes that are derived from it or that make use of it.
Some of the most important differences between Java and C or C++ involve how Java manages memory. Java eliminates ad hoc pointers and adds garbage collection and true arrays to the language. These features eliminate many otherwise insurmountable problems with safety, portability, and optimization.
Garbage collection alone should save countless programmers from the single largest source of programming errors in C or C++: explicit memory allocation and deallocation. In addition to maintaining objects in memory, the Java run-time system keeps track of all references to those objects. When an object is no longer in use, Java automatically removes it from memory. You can simply throw away references to objects you no longer use and have confidence that the system will clean them up at an appropriate time. Sun's current implementation of Java uses a conservative, mark and sweep garbage collector that runs intermittently in the background, which means that most garbage collecting takes place between mouse clicks and keyboard hits. Once you get used to garbage collection, you won't go back. Being able to write air-tight C code that juggles memory without dropping any on the floor is an important skill, but once you become addicted to Java you can "realloc" some of those brain cells to new tasks.
You may hear people say that Java doesn't have pointers. Strictly speaking, this statement is true, but it's also misleading. What Java provides are references--a safe kind of pointer--and Java is rife with them. A reference is a strongly typed handle for an object. All objects in Java, with the exception of primitive numeric types, are accessed through references. If necessary, you can use references to build all the normal kinds of data structures you're accustomed to building with pointers, like linked lists, trees, and so forth. The only difference is that with references you have to do so in a type-safe way.
Another important difference between a reference and a pointer is that you can't do pointer arithmetic with references. A reference is an atomic thing; you can't manipulate the value of a reference except by assigning it to an object. References are passed by value, and you can't reference an object through more than a single level of indirection. The protection of references is one of the most fundamental aspects of Java security. It means that Java code has to play by the rules; it can't peek into places it shouldn't.
Unlike C or C++ pointers, Java references can point only to class types. There are no pointers to methods. People often complain about this missing feature, but you will find that most tasks that call for pointers to methods, such as callbacks, can be accomplished using interfaces and anonymous adapter classes instead. 
 Java 1.1 does have a Method class, which lets you have a reference to a method. You can use a Method object to construct a callback, but it's tricky.
Finally, arrays in Java are true, first-class objects. They can be dynamically allocated and assigned like other objects. Arrays know their own size and type, and although you can't directly define array classes or subclass them, they do have a well-defined inheritance relationship based on the relationship of their base types. Having true arrays in the language alleviates much of the need for pointer arithmetic like that in C or C++.
Java's roots are in networked devices and embedded systems. For these applications, it's important to have robust and intelligent error management. Java has a powerful exception-handling mechanism, somewhat like that in newer implementations of C++. Exceptions provide a more natural and elegant way to handle errors. Exceptions allow you to separate error-handling code from normal code, which makes for cleaner, more readable applications.
When an exception occurs, it causes the flow of program execution to be transferred to a predesignated catcher block of code. The exception carries with it an object that contains information about the situation that caused the exception. The Java compiler requires that a method either declare the exceptions it can generate or catch and deal with them itself. This promotes error information to the same level of importance as argument and return typing. As a Java programmer, you know precisely what exceptional conditions you must deal with, and you have help from the compiler in writing correct software that doesn't leave them unhandled.
Applications today require a high degree of parallelism. Even a very single-minded application can have a complex user interface. As machines get faster, users become more sensitive to waiting for unrelated tasks that seize control of their time. Threads provide efficient multiprocessing and distribution of tasks. Java makes threads easy to use because support for them is built into the language.
Concurrency is nice, but there's more to programming with threads than just performing multiple tasks simultaneously. In many cases, threads need to be synchronized, which can be tricky without explicit language support. Java supports synchronization based on the monitor and condition model developed by C.A.R. Hoare--a sort of lock and key system for accessing resources. A keyword, synchronized, designates methods for safe, serialized access within an object. Only one synchronized method within the object may run at a given time. There are also simple, primitive methods for explicit waiting and signaling between threads interested in the same object.
Learning to program with threads is an important part of learning to program in Java. See Chapter 6, Threads for a discussion of this topic. For complete coverage of threads, see Java Threads, by Scott Oaks and Henry Wong (O'Reilly & Associates).
At the lowest level, Java programs consist of classes. Classes are intended to be small, modular components. They can be separated physically on different systems, retrieved dynamically, and even cached in various distribution schemes. Over classes, Java provides packages, a layer of structure that groups classes into functional units. Packages provide a naming convention for organizing classes and a second level of organizational control over the visibility of variables and methods in Java applications.
Within a package, a class is either publicly visible or protected from outside access. Packages form another type of scope that is closer to the application level. This lends itself to building reusable components that work together in a system. Packages also help in designing a scalable application that can grow without becoming a bird's nest of tightly coupled code dependency.