Comparison of C Sharp and Java

This article compares Microsoft's C# programming language with Oracle's (formerly Sun's) Java programming language. While the focus of this article is mainly the programming languages and their features, such a comparison will necessarily also consider some platform features and some library features. For a more detailed comparison of the platforms, please see Comparison of the Java and .NET platforms.

The comparison will naturally focus on areas where the languages differ. In fact the two languages and their platforms are more alike than they are different: Both are (primarily) statically, strongly, and mostly manifestly typed, both are class-based object-oriented, both are designed with semi-interpretation or runtime compilation in mind, both use garbage-collection, and both are "curly brace" languages like C and C++. Common ancestry is also evident in their common terminology and often very similar syntax features.

Language and features

Feature comparison

This section provides a comparison of the languages in terms of features they may or may not offer, or, put differently, properties they may or may not have. The absence of a feature should not automatically be regarded as a disadvantage for the given language; sometimes features may be excluded because the language designers view them as specifically detrimental, and in other cases the designers may have viewed the feature as something that would be nice to have but not worth the added language complexity.

Syntax

Both languages are considered "curly brace" languages in the C/C++ family. Overall the syntaxes of the languages are very similar. The syntax at the statement and expression level is almost identical with obvious inspiration from the C/C++ tradition. At type definition level (classes and interfaces) some minor differences exists. Java is explicit about extending classes and implementing interfaces, while C# infers this from the kind of types a new class/interface derives from.

C# supports more features than Java which to some extent is also evident in the syntax which specifies more keywords and more grammar rules than Java.

Keywords and backward compatibility

As the languages evolved, the language designers for both languages have faced situations where they wanted to extend the languages with new keywords and/or syntax. New keywords in particular may break existing code at source level, i.e. older code may no longer compile if presented to a compiler for a later version of the language. Language designers are keen to avoid such regressions. The designers of the two languages have been following different paths when addressing this problem.

Java language designers have avoided new keywords as much as possible, preferring instead to introduce new syntactic constructs which were not legal before or to reuse existing keywords in new contexts. This way they didn't jeopardize backward compatibility. An example of the former can be found in how the for loop was extended to accept iterable types. An example of the latter can be found in how the extends and (especially) the super keywords were reused for specifying type bounds when generics were introduced in Java 1.5. At one time (Java 1.4) a new keyword assert was introduced which was not reserved as a keyword before. This had the potential to render previously valid code invalid, if for instance the code used assert as an identifier. The designers chose to address this problem with a 4 step solution: 1) Introducing a compiler switch which indicates if Java 1.4 or later should be used, 2) Only marking assert as a keyword when compiling as Java 1.4 and later, 3) Defaulting to 1.3 to avoid rendering previous (non 1.4 aware code) invalid and 4) Issue warnings if the keyword is used in Java 1.3 mode, in order to allow the developers to change the code. Thus, the present Java syntax is backward compatible with all previous versions except for the specific case of assert.

C# language designers have introduced several new keywords since the first version. However, instead of defining these keywords as global keywords, they define them as context sensitive keywords. This means that even when they introduced (among others) the partial and yield keywords in C# 2.0, the use of those words as identifiers is still valid as there is no clash possible between the use as keyword and the use as identifier, given the context. Thus, the present C# syntax is fully backward compatible with source code written for any previous version without specifying the language version to be used.

Type system

The languages use very similar type systems. Both languages are statically typed with class-based object orientation.

Unified type system

In Java the primitive types are special in that they are not object-oriented and they could not have been defined using the language itself. They also do not share a common ancestor with reference types. The Java reference types all derive from a common root type, however. C# has a unified type system in which all types ultimately derive from a common root type. Consequently, all types implement the methods of this root type, and extension methods defined for the object type apply to all types, even primitive int literals and delegates. Note, that unlike Java, this allows C# to support objects with encapsulation which are not reference types.

In Java compound types are synonymous with reference types; methods cannot be defined for a type unless it is also a class reference type. In C# the concepts of encapsulation and methods have been decoupled from the reference requirement so that a type can support methods and encapsulation without being a reference type. Only reference types support virtual methods and specialization, however.

Simple/primitive types

Both languages support a number of built-in types which are copied and passed by value rather than by reference. Java calls these types primitive types, while they are called simple types in C#. The simple/primitive types typically have native support from the underlying processor architecture.

C# has a few more primitive types than Java, because it supports unsigned as well as signed integer types, and a decimal type for decimal floating-point calculations.

Java lacks the unsigned types. In particular, Java lacks a primitive type for an unsigned byte. The Java byte type is signed, while the C# byte is unsigned and sbyte is signed.

Both languages feature a native char (character) datatype as a simple type. Although the Java char type can be used with bitwise operators, this is actually accomplished by promoting the char to an integer before the operation.

C# has a type for high-precision (28 decimal digits) decimal arithmetic for e.g. financial and monetary applications. While Java lacks such a built-in type, the Java library does feature an arbitrary precision decimal type. This is not considered a language type and it does not support the usual arithmetic operators; rather it is a reference type which must be manipulated using the type methods. See more about arbitrary size/precision numbers below.

The C# primitive/simple types implement a number of interfaces and consequently offer a number of methods directly on instances of the types - even on the literals. The C# type names are also merely aliases for Common Language Runtime types. The C# Int64 type is exactly the same type as the long type; the only difference is that the former is the canonical .NET name while the latter is a C# alias for it.

Java does not offer methods directly on the primitive types. Instead methods which operate on the primitive values are offered through companion wrapper classes. A fixed set of such wrapper classes exist each of which wraps one of the fixed set of primitive types. As an example, the Java Long type is a reference type which wraps the primitive long type. They are not the same type, however.

The following table summarizes the corresponding simple/primitive types of the languages:

Advanced numeric types

Both languages offer library-defined arbitrary size integer types.

Only Java offers a data type for arbitrary precision decimal point calculations and only C# offers a type for working with complex numbers.

In both languages the number of operations which can be performed on the advanced numeric types are limited compared to the built-in IEEE 754 floating point types. For instance, none of the arbitrary size types support square root or logarithms.

C# allows library defined types to be integrated with existing types and operators by using custom implicit/explicit conversions and operator overloading as illustrated by the following example:

BigInteger bigNumber =
   new BigInteger("123456789012345678901234567890");
 
BigInteger answer = bigNumber.multiply(new BigInteger("42"));
BigInteger square = bigNumber.multiply(bigNumber);
BigInteger sum = bigNumber.add(bigNumber);

BigInteger bigNumber =
   BigInteger.Parse("123456789012345678901234567890");
 
var answer = bigNumber * 42;
var square = bigNumber * bigNumber;
var sum = bigNumber + bigNumber;

Built-in compound data types

Both languages treat strings as (immutable) objects of reference type. In both languages the type contains a number of methods to manipulate strings, parse, format etc. In both languages regular expressions are considered an external feature and is implemented in separate classes.

Both languages' libraries define classes for working with dates and calendars in different cultures. The Java java.util.Date is a mutable reference type where the C# System.DateTime is a struct value type. C# additionally define a TimeSpan type for working with time periods. Both languages support date/time arithmetic according to different cultures.

Reference types

Both languages use classes and interfaces as the primary means for defining new, object-oriented types. See below for further details.

Value types

C# allows the programmer to create user-defined value types, using the struct keyword. Unlike classes, and like the standard primitives, such value types are passed and assigned by value rather than by reference. They can also be part of an object (either as a field or boxed), or stored in an array, without the memory indirection that normally exists for class types.

Because value types have no notion of a null value and can be used in arrays without initialization, they always come with an implicit default constructor that essentially fills the struct memory space with zeroes. The programmer can only define additional constructors with one or more arguments. Value types do not have virtual method tables, and because of that (and the fixed memory footprint), they are implicitly sealed. However, value types can (and frequently do) implement interfaces. For example, the built-in integer types implement a number of interfaces.

Apart from the built-in primitive types, Java does not include the concept of value types.

Enumerations

In both C# and Java, programmers can use enumerations in a switch statement without conversion to a string or primitive integer type.

Delegates / method references

C# implements object oriented method pointers in the form of delegates. A delegate is a special type which can capture a type-safe reference to a method. This reference can then be stored in a delegate-type variable or passed to a method through a delegate parameter for later invocation. C# delegates support covariance and contravariance, and can hold a reference to any signature-compatible static method, instance method, anonymous method or lambda expression.

Delegates should not be confused with closures and inline functions. The concepts are related because a reference to a closure/inline function must be captured in a delegate reference to be useful at all. But a delegate does not always reference an inline function, it can also reference existing static or instance methods. Delegates form the basis of C# events but should not be confused with those either.

Delegates were deliberately left out of Java because they were considered unnecessary and detrimental to the language, and because of potential performance issues.^[12] Instead, an alternative mechanism must be used. The wrapper pattern, which resembles the delegates of C# in that it barely allows the client to access one or more client-defined subroutines through a known interface, is one way to design such a mechanism.

    // define a common interface for all invokeable objects
    // (will use the wrapper pattern)
    interface Invokeable {
        boolean invoke(String arg);
    }
 
    // a target class
    class Target {
        public boolean targetMethod(String arg) {
            // do something
            return true;
        }
    }
 
    // usage
    void doSomething() {
 
        // construct a target with the target method
        final Target target = new Target();
 
        // wrap the target
        Invokeable ivk = new Invokeable() {
            public boolean invoke(String arg) {
                return target.targetMethod(arg);
            }
        };
 
        // use the target through the wrapper
        boolean result = ivk.invoke("argumentstring");
    }

    // define a delegate
    private delegate bool SomeDelegate(string resource);
 
    // a target class
    class Target
    {
        public bool TargetMethod(string arg)
        {
            // do something
            return true;
        }
    }
 
    // usage
    void doSomething()
    {
        // construct a target with the target method
        var target = new Target();
 
        // capture the delegate for later invocation
        SomeDelegate dlg = target.TargetMethod;
 
        // invoke the delegate
        bool result = dlg("argumentstring");
    }

Lifted (nullable) types

C# allows value/primitive/simple types to be "lifted" to allow the special null value in addition to the type's native values. A type is lifted by adding a ? suffix to the type name. Conversions are implicitly defined to convert between values of the base and the lifted type. The lifted type can be compared against null or it can be tested for HasValue. Also, lifted operators are implicitly and automatically defined based on their non-lifted base, where — with the exception of some boolean operators — a null argument will propagate to the result.

Java does not support type lifting as a concept, but all of the built-in primitive types have corresponding wrapper types, which do support the null value by virtue of being reference types (classes).
According to the Java spec, any attempt to dereference the null reference must result in an exception being thrown at run-time, specifically a NullPointerException. (It would not make sense to dereference it otherwise, because, by definition, it points to no object in memory.) This also applies when attempting to unbox a variable of a wrapper type, which evaluates to null: the program will throw an exception, because actually there is no object to be unboxed - and therefore no boxed value to take part in the subsequent computation.

The following example illustrates the different behavior. In C#, the lifted * operator propagates the null value of the operand; in Java, unboxing the null reference throws an exception.

Integer a = 42;
Integer b = null;
 
// This will generate a runtime NullPointerException,
// because it attempts to unbox the null value.
Integer c = a * b;

int? a = 42;
int? b = null;
 
// c will receive the null value
// because * is lifted and one of the operands are null
int? c = a * b;

Not all C# lifted operators have been defined to propagate null unconditionally if one of the operands is null. Specifically, the boolean operators have been lifted to support ternary logic thus keeping impedance with SQL.

The Java boolean operators do not support ternary logic, nor is it implemented in the base class library.

Late-bound (dynamic) type

C# features a late bound dynamic type which supports no-reflection dynamic invocation, interoperability with dynamic languages as well as ad-hoc binding to (for example) document object models. The dynamic type resolves member access dynamically at runtime as opposed to statically/virtual at compile time. The member lookup mechanism is extensible with traditional reflection as a fall-back mechanism.

There are several use cases for the dynamic type in C#:

• Less verbose use of reflection: By casting an instance to the dynamic type, members such as properties, methods, events etc. can be directly invoked on the instance without using the reflection API directly.
• Interoperability with dynamic languages: The dynamic type comes with a hub-and-spoke support for implementing dynamically typed objects and common runtime infrastructure for efficient member lookup.
• Creating dynamic abstractions on the fly: For instance, a dynamic object could provide simpler access to document object models such as XML or XHTML documents.

Java does not support a late-bound type. The use cases for C# dynamic type have different corresponding constructs in Java:

• For dynamic late-bound by-name invocation of preexisting types, reflection should be used.
• For interoperability with dynamic languages, some form of interoperability API specific to that language will have to be used. The Java Virtual Machine platform does have multiple dynamic languages implemented on top of it, but there is no common standard for how to pass objects between languages. Usually this will involve some form of reflection or reflection-like API. As an example of how to use JavaFX objects from Java, see How to Use JavaFX in Your Swing Application.
• For creating and interacting with objects entirely at runtime, e.g. interaction with a document object model abstraction, a specific abstraction API will have to be used.

Dynamic language interoperability

The C# dynamic enables seamless interoperability with dynamic (late-bound) languages by allowing C# code to manipulate foreign objects using the same syntax as if they were native C# objects. Lacking this capability, Java developers must use a Java based API to access such objects. Consider a Ruby class (defined in a file called Deepthought.rb) which has 2 attributes (a,b) with read/write accessors and a Calculate method which returns the product of the attributes. The following examples illustrate how to instantiate and use such a class from within Java and C#, respectively.

ScriptEngineManager factory = new ScriptEngineManager();
ScriptEngine engine = factory.getEngineByName("jruby");
Invocable invocable = (Invocable) engine;
 
FileReader fr = new FileReader("Deepthought.rb");
engine.eval(fr);

var runtime = ScriptRuntime.CreateFromConfiguration();
dynamic globals = runtime.Globals;
 
runtime.ExecuteFile("Deepthought.rb");

Object calcClass = engine.eval("Deepthought");
Object calc = invocable.invokeMethod(calcClass, "new");
 
invocable.invokeMethod(calc, "a=", 6);
invocable.invokeMethod(calc, "b=", 7);
 
Object answer = invocable.invokeMethod(calc, "Calculate");

var calc = globals.Deepthought.@new();
 
calc.a = 6;
calc.b = 7;
 
var answer = calc.Calculate();

Pointers

C# allows use of pointers and corresponding pointer arithmetic. Pointers and pointer-arithmetic are potentially unsafe in a managed environment as they can be used to bypass the strict rules for object access. C# addresses that concern by requiring that code blocks or methods that use the feature be marked with the unsafe keyword, so that all clients of such code can be aware that the code may be less secure than otherwise. The compiler requires the /unsafe switch to allow compilation of a program that uses such code, and assemblies containing unsafe code may only execute if explicitly granted security permissions. Generally, unsafe code is either used to allow better interoperability with unmanaged APIs or system calls (which are inherently "unsafe"), or for performance reasons. Java does not permit pointers or pointer-arithmetic within the Java runtime environment and native interop is handled externally through JNI or other mechanisms.

Object handling

Both C# and Java are designed from the ground up as object oriented languages using dynamic dispatch, with syntax similar to C++ (C++ in turn derives from C). Neither language is a superset of C or C++, however.

Both languages mainly use garbage collection as a means of reclaiming memory resources, rather than explicit deallocation of memory. In both cases, if an object holds resources of different kinds other than memory, such as file handles, graphical resources, etc., then it will have to be notified explicitly when the application no longer uses it. Both C# and Java offer interfaces for such deterministic disposal and both C# and Java (since Java 7) feature automatic resource management statements which will automatically invoke the disposal/close methods on those interfaces.

Both languages include thread synchronization mechanisms as part of their language syntax.

References

In both languages references are a central concept. All instances of classes are by reference.

While not directly evident in the language syntax per se, both languages support the concept of weak references. An instance which is only referenced by weak references is eligible for garbage collection just as if there were no references at all. In both languages this feature is exposed through the associated libraries, even though it is really a core runtime feature.

In addition to weak references, Java has soft references. Soft references are much like weak references, but the JVM will not deallocate softly-referenced objects until the memory is actually needed.

Object initialization

In both C# and Java, an object's fields can be initialized either by variable initializers (expressions that can be assigned to variables where they are defined) or by constructors (special subroutines that are executed when an object is being created). In addition, Java contains instance initializers, which are anonymous blocks of code with no arguments that are run after the explicit (or implicit) call to a superclass's constructor but before the constructor is executed.

C# initializes object fields in the following order when creating an object:

1. Derived static fields
2. Derived static constructor
3. Derived instance fields
4. Base static fields
5. Base static constructor
6. Base instance fields
7. Base instance constructor
8. Derived instance constructor

Some of the above fields may not be applicable (e.g. if an object does not have static fields). Derived fields are those that are defined in the object's direct class, while base field is a term for the fields that are defined in one of the object's superclasses. Note that an object representation in memory contains all fields defined in its class or any of its superclasses, even if some fields in superclasses are defined as private.

It is guaranteed that any field initializers take effect before any constructors are called, since both the instance constructor of the object's class and its superclasses are called after field initializers are called. There is, however, a potential trap in object initialization when a virtual method is called from a base constructor. The overridden method in a subclass may reference a field that is defined in the subclass, but this field may not have been initialized because the constructor of the subclass that contains field initialization is called after the constructor of its base class.

In Java, the order of initialization is as follows:

1. Invocation of another constructor (either of the object's class or of the object's superclass)
2. Instance variable initializers and instance initializers (in the order they appear in the source code)
3. The constructor body

Like in C#, a new object is created by calling a specific constructor. Within a constructor, the first statement may be an invocation of another constructor. If this is omitted, the call to the argumentless constructor of the superclass is added implicitly by the compiler. Otherwise, either another overloaded constructor of the object's class can be called explicitly, or a superclass constructor can be called. In the former case, the called constructor will again call another constructor (either of the object's class or its subclass) and the chain sooner or later ends up at the call to one of the constructors of the superclass.

After another constructor is called (which causes direct invocation of the superclass constructor, and so forth, down to the Object class), instance variables defined in the object's class are initialized. Even if there are no variable initializers explicitly defined for some variables, these variables are initialized to default values. Note that instance variables defined in superclasses are already initialized by this point, because they were initialized by a superclass constructor when it was called (either by the constructor's code or by variable initializers performed before the constructor's code or implicitly to default values). In Java, variable initializers are executed according to their textual order in the source file.

Finally, the constructor body is executed. This ensures proper order of initialization, i.e. the fields of a base class finish initialization before initialization of the fields of an object class begins.

There are two main potential traps in Java's object initialization. First, variable initializers are expressions that can contain method calls. Since methods can reference any variable defined in the class, the method called in a variable initializer can reference a variable that is defined below the variable being initialized. Since initialization order corresponds to textual order of variable definitions, such a variable would not be initialized to the value prescribed by its initializer and would contain the default value. Another potential trap is when a method that is overridden in the derived class is called in the base class constructor, which can lead to behavior the programmer would not expect when an object of the derived class is created. According to the initialization order, the body of the base class constructor is executed before variable initializers are evaluated and before the body of the derived class constructor is executed. The overridden method called from the base class constructor can, however, reference variables defined in the derived class, but these are not yet initialized to the values specified by their initializers or set in the derived class constructor. The latter issue applies to C# as well, but in a less critical form since in C# methods are not overridable by default.

Arrays and collections

Arrays and collections are concepts featured by both languages. The syntax used to declare and access arrays is identical, except that C# has added syntax for declaring and manipulating multidimensional arrays.

Multidimensional arrays can in some cases increase performance because of increased locality (as there is a single pointer dereference, instead of one for every dimension of the array as is the case for jagged arrays). However, since all array element access in a multidimensional array requires multiplication/shift between the two or more dimensions, this is an advantage only in very random access scenarios.

Another potential advantage is that the entire multidimensional array can be allocated with a single application of operator new, while jagged arrays require loops and allocations for every dimension. Note, though, that Java provides a syntactic construct for allocating a multidimensional jagged array with regular lengths; the loops and multiple allocations are then performed by the virtual machine and need not be explicit at the source level.

Both languages feature an extensive set of collection types which includes various ordered and unordered types of lists, maps/dictionaries, sets, etc.

Type members

Operator overloading and implicit and explicit conversions

Operator overloading and user-defined casts are separate features which both aim to allow new types to become first-class citizens in the type system. By using these features in C#, types such as Complex and decimal have been integrated so that the usual operators like addition and multiplication work with the new types.

Java does not include operator overloading nor custom conversions in order to prevent abuse of the feature, and to keep the language simple.^[13]

Indexers

C# also includes indexers which can be considered a special case of operator overloading (like the C++ operator[] ), or parameterized get/set properties. An indexer is a property named this[] which uses one or more parameters (indexes); the indices can be objects of any type:

myList[4] = 5;
string name = xmlNode.Attributes["name"];
orders = customerMap[theCustomer];

Java does not include indexers. The common Java pattern involves writing explicit getters and setters where a C# programmer would use an indexer.

Partial classes

C# allows a class definition to be split across several source files using a feature called partial classes. Each part must be marked with the keyword partial. All the parts must be presented to the compiler as part of a single compilation. Parts can reference members from other parts. Parts can implement interfaces and one part can define a base class. The feature is useful in code generation scenarios (such as UI design) where a code generator can supply one part and the developer another part to be compiled together. The developer can thus edit their part without the risk of a code generator overwriting that code at some later time. Unlike the class extension mechanism, a partial class allows "circular" dependencies among its parts as they are guaranteed to be resolved at compile time. Java has no corresponding concept.

Inner and local classes

Both languages allow inner classes, where a class is defined lexically inside another class. However, in each language these inner classes have rather different semantics.

In Java, unless the inner class is declared static, a reference to an instance of an inner class carries a reference to the outer class with it. As a result, code in the inner class has access to both the static and non-static members of the outer class. To create an instance of a non-static inner class, one has to name the instance of the embracing outer class.^[14] This is done via a new new-operator introduced in JDK 1.3: outerClassInstance.new Outer.InnerClass(). This can be done in any class that has a reference to an instance of the outer class.

In C#, an inner class is conceptually the same as a normal class. In a sense, the outer class only acts as a namespace. Thus, code in the inner class cannot access non-static members of the outer class unless it does so through an explicit reference to an instance of the outer class. Programmers can declare the inner class private to allow only the outer class to have any access to it.

Java provides another feature called local classes or anonymous classes, which can be defined within a method body. These are generally used to implement an interface with only one or two methods, which are typically event handlers. However, they can also be used to override virtual methods of a superclass. The methods in those local classes have access to the outer method's local variables declared final. C# satisfies the use-cases for these by providing anonymous delegates; see event handling for more about this.

C# also provides a feature called anonymous types/classes, but it is rather different from Java's concept with the same name. It allows the programmer to instantiate a class by providing only a set of names for the properties the class should have, and an expression to initialize each. The types of the properties are inferred from the types of those expressions. These implicitly-declared classes are derived directly from object.

Explicit member implementation

C# also has "Explicit Member Implementation" which allows a class to specifically implement methods of an interface, separate to its own class methods, or to provide different implementations for two methods with the same name and signature inherited from two base interfaces.

Events

C# multicast-delegates are called events. Events provide support for event-driven programming and is an implementation of the observer pattern. To support this there is a specific syntax to define events in classes, and operators to register, unregister or combine event handlers.

See here for information about how events are implemented in Java.

Statements and expressions

Boxing and unboxing

Both languages allow automatic boxing and unboxing, i.e. they allow for implicit casting between any primitive types and the corresponding reference types.
In C#, the primitive types are subtypes of the Object type. In Java this is not true; any given primitive type and the corresponding wrapper type have no specific relationship with each other, except for autoboxing and unboxing, which act as syntactic sugar for interchanging between them. This was done intentionally, to maintain backward compatibility with prior versions of Java, in which no automatic casting was allowed, and the programmer worked with two separate sets of types: the primitive types, and the wrapper (reference) type hierarchy.^[15]

This difference has the following consequences. First of all, in C#, primitive types can define methods, such as an override of Object's ToString() method. In Java, this is task is accomplished by the primitive wrapper classes.
Secondly, in Java an extra cast is needed whenever one tries to directly dereference a primitive value, as it will not be boxed automatically. The expression ((Integer)42).toString() will convert an integer literal to string in Java while 42.ToString() performs the same operation in C#. This is because the latter one is actually an instance call on the primitive value 42, while the former one is an instance call on an object of type java.lang.Integer.

Finally, another difference is that Java makes heavy use of boxed types in generics (see below).

Generics

In the field of generics the two languages show a superficial syntactical similarity, but they have deep underlying differences.

Type erasure versus reified generics

Generics in Java are a language-only construction; they are implemented only in the compiler. The generated classfiles include generic signatures only in the form of metadata (allowing the compiler to compile new classes against them). The runtime has no knowledge of the generic type system; generics are not part of the JVM. Instead, generics classes and methods are transformed during compilation through a process known as type erasure. During this process the compiler replaces all generic types with their raw version and inserts casts/checks appropriately in client code where the type and its methods are used. The resulting byte code will contain no references to any generic types or parameters (See also Generics in Java).
The language specification intentionally prohibits certain uses of generics; this is necessary to allow for implementing generics through type erasure, and to allow for migration compatibility.^[16]

C# builds on support for generics from the virtual execution system itself, i.e. it is not just a language feature. The language is merely a front-end for cross-language generics support in the CLR. During compilation generics are verified for correctness, but code generation for actually implementing the generics are deferred to class-load time. Client code (code invoking generic methods/properties) are fully compiled and can safely assume generics to be type-safe. This is called reification. At runtime, when a unique set of type parameters for a generic class/method/delegate is encountered for the first time, the class loader/verifier will synthesize a concrete class descriptor and generate method implementations. During the generation of method implementations all reference types will be considered a single type, as reference types can safely share the same implementations. Note, this is merely for the purpose of the implementing code. Different sets of reference types will still have unique type descriptors; their method tables will merely point to the same code.

The following list illustrates some differences between Java and C# when managing generics. It is not exhaustive:^[17]

Object tenPairs =
    new Pair<Integer, String>[10]; // error

object tenPairs =
    new Pair<int, string>[10]; // ok

public class Lookup<TKey,TValue> {
    public TValue[] getEmptyValues(TKey key) {
        return new TValue[0]; // error
    }
}

public class Lookup<TKey, TValue>
{
    public TValue[] GetEmptyValues(TKey key)
    {
        return new TValue[0]; // ok
    }
}

C# allows generics directly for primitive types. Java, instead, allows the use of boxed types as type parameters (e.g., List<Integer> instead of List<int>). This comes at a cost since all such values need to be boxed/unboxed when used, and they all need to be heap-allocated. However, a generic type can be specialized with an array type of a primitive type in Java, for example List<int[]> is allowed.^[22]

Migration compatibility

Java's type erasure design was motivated by a design requirement to achieve migration compatibility - not to be confused with backward compatibility. In particular, the original requirement was "... there should be a clean, demonstrable migration path for the Collections APIs that were introduced in the Java 2 platform".^[23] This was designed so that any new generic collections should be passable to methods which expected one of the pre-existing collection classes.^[24]

C# generics were introduced into the language while preserving full backward compatibility, but did not preserve full migration compatibility: Old code (pre C# 2.0) runs unchanged on the new generics-aware runtime without recompilation. As for migration compatibility, new generic collection classes and interfaces were developed which supplemented the non-generic .NET 1.x collections rather than replacing them. In addition to generic collection interfaces, the new generic collection classes implement the non-generic collection interfaces where possible. This prevents the use of new generic collections with pre-existing (non-generic aware) methods, if those methods are coded to use the collection classes.

Covariance and contravariance

Covariance and contravariance is supported by both languages. Java has use-site variance which allows a single generic class to declare members using both co- and contravariance. C# has define-site variance for generic interfaces and delegates. Variance is not supported directly on classes but is supported through their implementation of variant interfaces. C# also has use-site covariance support for methods and delegates.

Notation and special features

Special feature keywords

switch(color)
{
    case Color.Blue:
         Console.WriteLine("Color is blue"); break;
    case Color.DarkBlue:
         Console.WriteLine("Color is dark");
         goto case Color.Blue;
    // ...
}

public int readItem() throws java.io.IOException
{
    // ...
}

// Create a small file "test.txt", write a string,
// ... and close it (even if an exception occurs)
using (StreamWriter file = new StreamWriter("test.txt"))
{
    file.Write("test");
}

Numeric applications

To adequately support applications in the field of mathematical and financial computation, several language features exist.^[25] In this category, Java provides the strictfp keyword, which enables strict floating-point calculations for a region of code. This will ensure that calculations return exactly the same result on all platforms. C# provides no equivalent, but does provide the built-in decimal type, for accurate decimal floating-point calculations. This forgoes the problems that exist with binary floating-point representations (float, double). Such binary representations are not suited to accurately represent decimal numbers and hence introduce rounding errors. For financial applications, an accurate decimal type is essential.

The BigDecimal class also provides such characteristics for Java. BigDecimal and BigInteger are types provided with Java that allow arbitrary-precision representation of numbers. As of 2010^[update] the current stable release of the .NET framework (4.0) includes classes for manipulating arbitrary-precision integers and complex numbers (with operators overloaded for easy use so that BigInteger objects can be used just like any other primitive data type), but still the .NET Framework lacks classes to deal with arbitrary-precision floating point numbers (see software for arbitrary-precision arithmetic).

In Java there is no way to provide the same level of integration for library-defined types such as BigDecimal or complex numbers as there is for the primitive types. For this purpose, C# provides the following:

• Operator overloading and indexers providing convenient syntax (see below).
• Implicit and explicit conversions; allow conversions such as exist for the built-in int type that can implicitly convert to long.
• Valuetypes and generics based on valuetypes; in Java every custom type must be allocated on the heap, which is detrimental for performance of both custom types and collections.

In addition to this, C# can help mathematical applications with the checked and unchecked operators that allow the enabling or disabling of run-time checking for arithmetic overflow for a region of code.

Methods

Methods in C# are non-virtual by default, and have to be declared virtual explicitly if desired. In Java, all non-static non-private methods are virtual. Virtuality guarantees that the most recent override for the method will always be called, but incurs a certain runtime cost on invocation as these invocations cannot be normally inlined, and require an indirect call via the virtual method table. However, some JVM implementations, including the Sun reference implementation, implement inlining of the most commonly called virtual methods.

Java methods are virtual by default (although they can be "sealed" by using the final modifier to disallow overriding). There is no way to let derived classes define a new, unrelated method with the same name.

This means that by default in Java, and only when explicitly enabled in C#, new methods may be defined in a derived class with the same name and signature as those in its base class. When the method is called on a superclass reference of such an object, the "deepest" overridden implementation of the base class' method will be called according to the specific subclass of the object being referenced.

In some cases, when a subclass introduces a method with the same name and signature as a method already present in the base class, problems can occur. In Java, this will mean that the method in the derived class will implicitly override the method in the base class, even though that may not be the intent of the designers of either class.

To mitigate this, C# requires that if a method should be overridden, the override keyword must be specified. Otherwise, the method will "hide" the inherited method. A compiler warning to this effect is issued, which can be silenced by specifying the new keyword. This avoids the problem which can arise from a base class being extended with a protected/public method whose signature is already in use by a derived class.

Generator methods

Any C# method declared as returning IEnumerable, IEnumerator and/or the generic versions of these interfaces can be implemented using yield syntax. This is a form of limited, compiler-generated continuations and can drastically reduce the code required to traverse or generate sequences, although that code is just generated by the compiler instead. The feature can also be used to implement infinite sequences, e.g. the sequence of Fibonacci numbers.

Java does not have an equivalent feature. Instead generators are typically defined by providing a specialized implementation of a well-known collection or iterable interface, which will compute each element on demand. For such a generator to be used in a for each statement, it must implement interface java.lang.Iterable.

Below these two approaches are shown for the two languages. Both use the designated iterator/enumerator interface of the language. Note how the C# version use multiple yield return statements to alternate between accumulating in one or the other local variable. The Java version would have to use an extra boolean flag to achieve the same, or use a temporary variable as in this case.

public class Fibonacci implements Iterable<Integer> {
 
    public Iterator<Integer> iterator() {
        return new Iterator<Integer>() {
 
            private int a = 0;
            private int b = 1;
 
            public boolean hasNext() {
                return true; // infinite sequence
            }
 
            public Integer next() {
                int tmp = a;
                a = b;
                b = a + tmp;
                return tmp;
            }
 
            public void remove() {
                throw new UnsupportedOperationException
                    ("Not supported on the Fibonacci sequence.");
            }
        };
    }
}

public static IEnumerable<int> Fibonacci()
{
    int a = 0;
    int b = 1;
    while (true) // infinite sequence
    {
        yield return a;
        a += b;
        yield return b;
        b += a;
    }
}

Explicit interface implementation

In either language if a method (or property in C#) is specified with the same name and signature in multiple interfaces, the members will clash when a class is designed which implements those interfaces. An implementation will by default implement a common method for all of the interfaces. If separate implementations are required (because the methods really do serve separate purposes, or because return values differ between the interfaces) C#'s explicit interface implementation will solve the problem, though allowing different results for the same method, depending on the current cast of the object. In Java there is no way to solve this problem other than cleaning the code and refactoring one or more of the interfaces to avoid name clashes.

Closures

A closure is an inline function which captures variables from its lexical scope.

C# supports closures as anonymous methods or lambda expressions with full-featured closure semantics.^[26][27]

In Java, anonymous inner classes remains the preferred way to emulate closures. Aside from being a more verbose construction, this approach also has some drawbacks compared to real closures, such as limited access to variables from the enclosing scopes (only final members can be referenced).

When a reference to a method can be passed around for later execution, a problem arises about what to do when the method has references to variables/parameters in its lexical scope. C# closures can fully capture any variable/parameter from its lexical scope. In Java's anonymous inner classes only references to final members of the lexical scope are allowed, thus requiring the developer to artificially introduce extra levels of indirections and boxing primitive types if he wants to reference and update those from the inner class.

While Java does not currently feature closures, it has been announced that some form of closures or lambdas will be included in JDK 8 which at latest update (10 October 2010) is scheduled for release "late 2012".^[28]

Language integrated query (LINQ)

C#s Language INtegrated Query (LINQ) is not really a single feature; rather it is a number of features designed to work together to allow for in-language querying capabilities. LINQ has emerged as one of the most distinguishing features between Java and C#.

LINQ consists of the following features:

• Extension methods allow existing interfaces or classes to be extended with new methods. Implementations can be shared or an interface can have a dedicated implementation.
• Lambdas allow for expression of criteria in a functional fashion.
• Expression trees allow a specific implementation to capture a lambda as a concrete syntax tree rather than an executable block. This can be utilized by implementations to represent criteria in a different language, e.g. in the form of an SQL where clause as is the case with e.g. Linq, LINQ to SQL.
• Anonymous types and type inference supports capturing and working with the result type of a query. A query may both join and project over query sources which may lead to a result type which cannot be named.
• Query expressions to support a syntax familiar to SQL users.
• Nullable (lifted) types to allow for a better match with query providers which support nullable types, like e.g. SQL.

Lambdas and expression trees

C# features a special type of in-line closures called lambdas. These are anonymous methods: they have a signature and a body, but no name. They are mainly used to specify local function-valued arguments in calls to other methods, a technique mainly associated with functional programming. On top of that, lambda functions can double as a way to define special data structures called expression trees. Whether they are seen as an executable function or as a data structure depends on compiler type inference and what type of variable or parameter they are assigned or cast to. Lambdas and expression trees play key roles in LINQ. Java does not feature lambdas or expression trees; its primary mechanism for inline scope capture and method definition is the anonymous inner class syntax.

Partial methods

Related to partial classes C# allows partial methods to be specified within partial classes. A partial method is an intentional declaration of a method with a number of restrictions on the signature. These restrictions ensure that if a definition is not actually provided by any class part, then the method and every call to it can be safely erased. This feature allows code to provide a large number of interception points (like the template method GoF design pattern) without paying any runtime overhead if these extension points are not being used by another class part at compile time. Java has no corresponding concept.

Extension methods

Using a special this designator on the first parameter of a method, C# allows the method to act as if it were a member method of the type of the first parameter. This extension of the foreign class is purely syntactical. The extension method needs to be static and defined within a purely static class. It must obey any restriction on such external static methods and thus cannot break object encapsulation. The "extension" is only active within scopes where the namespace of the static host class has been imported. Java does not have an equivalent feature (although one is being discussed for Java 8).

Conditional compilation

Unlike Java, C# implements conditional compilation using preprocessor directives. It also provides a Conditional attribute to define methods that are only called when a given compilation constant is defined. This way, assertions can be provided as a framework feature with the method Debug.Assert(), which is only evaluated when the DEBUG constant is defined. Since version 1.4, Java provides a language feature for assertions, which are turned off at runtime by default but can be enabled using the -enableassertions or -ea switch when invoking the JVM.

Namespaces and source files

In C#, namespaces are similar to those in C++. Unlike package names in Java, a namespace is not in any way tied to the location of the source file. While it is not strictly necessary for a Java source file location to mirror its package directory structure, it is the conventional organization.

Both languages allow importing of classes (e.g., import java.util.* in Java), allowing a class to be referenced using only its name. Sometimes classes with the same name exist in multiple namespaces or packages. Such classes can be referenced by using fully qualified names, or by importing only selected classes with different names. To do this, Java allows importing a single class (e.g., import java.util.List). C# allows importing classes under a new local name using the following syntax: using Console = System.Console. It also allows importing specializations of classes in the form of using IntList = System.Collections.Generic.List<int>.

Java has a static import syntax that allows using the short name of some or all of the static methods/fields in a class (e.g., allowing foo(bar) where foo() can be statically imported from another class). C# has a static class syntax (not to be confused with static inner classes in Java), which restricts a class to only contain static methods. C# 3.0 introduces extension methods to allow users to statically add a method to a type (e.g., allowing foo.bar() where bar() can be an imported extension method working on the type of foo).

The Sun Microsystems Java compiler requires that a source file name must match the only public class inside it, while C# allows multiple public classes in the same file, and puts no restrictions on the file name. C# 2.0 and later allows splitting a class definition into several files by using the partial keyword in the source code. In Java, a public class will always be in its own source file. In C#, source code files and logical units separation are not tightly related.

Exception handling

Java supports checked exceptions (in addition to unchecked exceptions). C# only supports unchecked exceptions. Checked exceptions force the programmer to either declare the exception thrown in a method, or to catch the thrown exception using a try-catch clause.

Checked exceptions can encourage good programming practice, ensuring that all errors are dealt with. However Anders Hejlsberg, chief C# language architect, argues that they were to some extent an experiment in Java and that they have not been shown to be worthwhile except in small example programs.^[29][30]

One criticism is that checked exceptions encourage programmers to use an empty catch block (catch (Exception e) {}), which silently eats exceptions, rather than letting the exceptions propagate to a higher-level exception-handling routine. In some cases, however, exception chaining can be applied instead, by re-throwing the exception in a wrapper exception. For example, if an object is changed to access a database instead of a file, an SQLException could be caught and re-thrown as an IOException, since the caller may not need to know the inner workings of the object.

However, not all programmers agree with this stance, with James Gosling and others maintaining that checked exceptions are a good idea and it's people misusing them that cause the issues. Silently catching exceptions is possible, yes, but you have to explicitly say what you want to do with the exception as oppose to unchecked exceptions which let you do nothing by default. You can ignore it, but you have to explicitly write code to ignore it.^[31][32]

There are also differences between the two languages in treating the try-finally statement. The finally block is always executed, even if the try block contains control-passing statements like throw or return. In Java, this may result in unexpected behavior if the try block is left by a return statement with some value, and then the finally block that is executed afterward is also left by a return statement with a different value. C# resolves this problem by prohibiting any control-passing statements like return or break in the finally block.

A common reason for using try-finally blocks is to guard resource managing code, thus guaranteeing the release of precious resources in the finally block. C# features the using statement as a syntactic shorthand for this common scenario, in which the Dispose() method of the object of the using is always called.

A rather subtle difference is the moment a stack trace is created when an exception is being thrown. In Java, the stack trace is created in the moment the exception is created.

class Foo {
    Exception e = new Exception();
    int foo() throws Exception {
        throw e;
    }
}

The exception in the statement above will always contain the constructor's stack-trace - no matter how often foo is called. In C# on the other hand, the stack-trace is created the moment "throw" is executed.

class Foo {
    Exception e = new Exception();
    int foo() {
        try {
            throw e;
        } catch (Exception e) {
            throw;
        }
    }
}

In the code above, the exception will contain the stack-trace of the first throw-line. When catching an exception, there are two options in case the exception should be rethrown: throw will just rethrow the original exception with the original stack, while throw e would have created a new stack trace.

Finally blocks

Java allows flow of control to leave the finally block of a try statement, regardless of the way it was entered. This can cause another control flow statement (such as return) to be terminated mid-execution. For example:

int foo() {
    try {
        return 0;
    } finally {
        return 1;
    }
}

In the above code, the return statement within try block causes control to leave it, and therefore finally block is executed before the actual return happens. However, finally block itself performs a return as well; thus, the original return that caused it to be entered is not actually executed, and the above method returns 1 rather than 0.

C# does not allow any statements which allow control flow to leave the finally block prematurely, except for throw. In particular, return is not allowed at all, goto is not allowed if the target label is outside the finally block, and continue and break are not allowed if the nearest enclosing loop is outside the finally block.

Lower level code

The Java Native Interface (JNI) feature allows Java programs to call non-Java code. However, JNI does require the code being called to follow several conventions and imposes restrictions on types and names used. This means that an extra adaption layer between legacy code and Java is often needed. This adaption code must be coded in a non-Java language, often C or C++. Java Native Access (JNA) allows easier calling of native code that only requires writing Java code, but comes at a performance cost.

In addition, third party libraries provide for Java-COM bridging, e.g. JACOB (free), and J-Integra for COM (proprietary).

.NET Platform Invoke (P/Invoke) offers the same capability by allowing calls from C# to what Microsoft refers to as unmanaged code. Through metadata attributes the programmer can control exactly how the parameters and results are marshalled, thus avoiding the need for extra adaption code. P/Invoke allows almost complete access to procedural APIs (such as Win32 or POSIX), but limited access to C++ class libraries.

In addition, .NET Framework also provides a .NET-COM bridge, allowing access to COM components as if they were native .NET objects.

C# also allows the programmer to disable the normal type-checking and other safety features of the CLR, which then enables the use of pointer variables. When using this feature, the programmer must mark the code using the unsafe keyword. JNI, P/Invoke, and "unsafe" code are equally risky features, exposing possible security holes and application instability. An advantage of unsafe, managed code over P/Invoke or JNI is that it allows the programmer to continue to work in the familiar C# environment to accomplish some tasks that otherwise would require calling out to unmanaged code. An assembly (program or library) using unsafe code must be compiled with a special switch and will be marked as such. This enables runtime environments to take special precautions before executing potentially harmful code.

Runtime environments

Java (the programming language) is designed to execute on the Java platform via the Java Runtime Environment (JRE). The Java platform includes the Java Virtual Machine (JVM) as well as a common set of libraries. The JRE was originally designed to support interpreted execution with final compilation as an option. Most JRE environments execute fully or at least partially compiled programs, possibly with adaptive optimization. The Java compiler produces Java bytecode. Upon execution the bytecode is loaded by the Java runtime and either interpreted directly or compiled to machine instructions and then executed.

C# is designed to execute on the Common Language Runtime (CLR). The CLR is designed to execute fully compiled code. The C# compiler produces Common Intermediate Language instructions. Upon execution the runtime loads this code and compiles to machine instructions on the target architecture.

Language history and evolution

Java

Java is older than C# and has built up a large and highly active user base, becoming the lingua franca in many modern branches of computer science, particularly areas which involve networking. Java dominates programming courses at high school and college level in the United States, and there are currently more Java books than C# books.^[33] Java's maturity and popularity have ensured more third party Java APIs and libraries (many of them open source) than C#.

An occasionally voiced criticism ^[34] of the Java language is that it evolves slowly, lacking some features which make fashionable programming patterns and methodologies easier. Some critics suggest that the designers of C# may pander too quickly to current trends in programming, thus lacking focus and simplicity. Java's designers seem to have taken a more conservative stand on adding major new features to their language syntax than other current languages, perhaps not wanting to tie the language too tightly with trends which may prove to be dead ends.

These trends have been broken with the Java 5.0 release, which introduced several new major language features: a foreach construct, autoboxing, methods with variable number of parameters (varargs), enumerated types, generic types, and annotations. With the exception of Generics, C# included all these features from its beginning, some under different names.^[35] Proposals and specifications for the new features had been worked on in the Java community for a considerable time before they were introduced. Indeed, some had been in gestation since before C#'s initial release (e.g., work on Generics formally began in May 1999^[36]), such was the Java community's conservatism at that time.

Problem-specific language additions to Java have been considered and, for now at least, rejected. This approach, along with a number of other new languages and technologies that address themselves specifically toward current programming trends, has sparked a renewed debate within the Java camp about the future direction of the Java language and whether its 'conservative evolution' is right.

As of 2008 debate continued over the inclusion of closures^[37] and properties^[38] into the language syntax for Java 7.

C#

By contrast, C# is a relatively new language. Microsoft has studied existing languages such as Java and Object Pascal, and has changed some aspects of the language and runtime environment in response to perceived failures and difficulties with its predecessors. C# accommodates constructs more commonly found in languages such as C++, Delphi (the design of which was Anders Hejlsberg's principal job when he was at Borland), and, in recent C# versions, borrows from dynamic scripting languages such as Ruby and Python.

C# 3.0 adds SQL-like language integrated queries suited for querying data from collections, databases or XML documents, building upon general-purpose language features, including lambda expressions and extension methods, to allow queries to be expressed and optimized for user types.

Before creating C#, Microsoft implemented a modified Java environment, called J++, adding new features in a manner which was in direct contravention to the standards and conventions that ensure the platform neutrality which lies at the heart of Java. This violated the license agreement Microsoft had signed, requiring that standards and specifications be strictly adhered to in return for using the Java name and brand logos. Sun Microsystems sued, and in settling the suit, Microsoft agreed to discontinue J++. (The settlement allowed other existing Java-using Microsoft products to continue such use for seven years.^[39]) With the release of the .NET framework (and C#), the project was revived in the form of J#.

See also

• Comparison of C# and VB.NET
• Comparison of Java and C++
• Java programming language
• Comparison of the Java and .Net platforms

References

External links

• Moving to C# and the .NET Framework at MSDN
• C# and Java: Comparing Programming Languages at MSDN
• A Comparison of C# and Java
• Java vs. C# - Code for Code Comparison
• Nine Language Performance Round-up
• Java and C-Sharp Compared
• MSDN: The C# Programming Language for Java Developers
• Standard ECMA-334 C# Language specification
• Java Language Specification (Sun)