An object copy is an action in computing where a data object has its attributes copied to another object of the same data type. An object is a composite data type in object-oriented programming languages. The copying of data is one of the most common procedures that occurs in computer programs. An object may be copied to reuse all or part of its data in a new context.
Contents |
The design goal of most objects is to give the semblance of being made out of one monolithic block even though most are not. As objects are made up of several different parts, copying becomes non trivial. Several strategies exist to attack this problem.
Consider two objects, A and B, which each refer to two memory blocks xi and yi (i = 1, 2,...). Think of A and B as strings and of xi and yi (i = 1, 2,...) as the characters they contain. There are different strategies for copying A into B.
One method of copying an object is the shallow copy. In this process, B is attached to the same memory block as A. This is otherwise known as address copy.
This results in a situation in which same data is shared between A and B, thus modifying the one will alter the other. The original memory block of B is now no longer referred to from anywhere. If the language does not have automatic garbage collection the original memory block of B has probably been leaked.
The advantage of shallow copies is that their execution speed is fast and does not depend on the size of the data.
Bitwise copies of objects which are not made up of a monolithic block are shallow copies.
An alternative is a deep copy. Here the data are actually copied over. The result is different from the result a shallow copy gives. The advantage is that A and B do not depend on each other but at the cost of a slower more expensive copy.
A lazy copy is a combination of both strategies above. When initially copying an object, a (fast) shallow copy is used. A counter is also used to track how many objects share the data. When the program wants to modify an object, it can determine if the data is shared (by examining the counter) and can do a deep copy if necessary.
Lazy copy looks to the outside just as a deep copy but takes advantage of the speed of a shallow copy whenever possible. The downside are rather high but constant base costs because of the counter. Also, in certain situations, circular references can also cause problems.
Lazy copy is related to copy-on-write.
Nearly all object-oriented programming languages provide some way to copy objects. As the majority of languages do not provide most objects themselves, the programmer has to define how an object should be copied, just as he or she has to define if two objects are identical or even comparable in the first place. Many languages provide some default behavior.
How copying is solved varies from language to language and what concept of an object it has. The following presents examples for two of the most widely used object-oriented languages, C++ and Java, which should cover nearly every way that an object-oriented language can attack this problem.
C++ is designed so that user-defined types should be able to function exactly like native types such as char and int. One way in which C++ achieves this is by allowing any class to define a copy constructor and an assignment operator. C++ provides default versions of these which perform a memberwise copy, but any class is free to disable copying or define its own copy behaviour.
Unlike in C++, objects in Java are always accessed indirectly through references. Objects are never created implicitly but instead are always passed or assigned by a reference variable. (Methods in Java are always pass by value, however, it is the value of the reference variable that is being passed.) The Java Virtual Machine manages garbage collection so that objects are cleaned up after they are no longer reachable. There is no automatic way to copy any given object in Java.
Copying is usually performed by a clone() method of a class. This method usually, in turn, calls the clone() method of its parent class to obtain a copy, and then does any custom copying procedures. Eventually this gets to the clone() method of Object (the uppermost class), which creates a new instance of the same class as the object and copies all the fields to the new instance (a "shallow copy"). If this method is used, the class must implement the Cloneable marker interface, or else it will throw a CloneNotSupportedException. After obtaining a copy from the parent class, a class's own clone() method may then provide custom cloning capability, like deep copying (i.e. duplicate some of the structures referred to by the object) or giving the new instance a new unique ID.
One disadvantage is that the return type of clone() is Object, and needs to be explicitly cast back into the appropriate type (technically a custom clone() method could return another type of object; but that is generally inadvisable). One advantage of using clone() is that since it is an overridable method, we can call clone() on any object, and it will use the clone() method of its actual class, without the calling code needing to know what that class is (which would be necessary with a copy constructor).
Another disadvantage is that one often cannot access the clone() method on an abstract type. Most interfaces and abstract classes in Java do not specify a public clone() method. As a result, often the only way to use the clone() method is if the actual class of an object is known, which is contrary to the abstraction principle of using the most generic type possible. For example, if one has a List reference in Java, one cannot invoke clone() on that reference because List specifies no public clone() method. Actual implementations of List like ArrayList and LinkedList all generally have clone() methods themselves, but it is inconvenient and bad abstraction to carry around the actual class type of an object.
Another way to copy objects in Java is to serialize them through the Serializable interface. This is typically used for persistence and wire protocol purposes, but it does create copies of objects and, unlike clone, a deep copy that gracefully handles cycled graphs of objects is readily available with minimal effort from the programmer.
Both of these methods suffer from a notable problem: the constructor is not used for objects copied with clone or serialization. This can lead to bugs with improperly initialized data, prevents the use of final member fields, and makes maintenance challenging. Some utilities attempt to overcome these issues by using reflection to deep copy objects, such as generic deep copy[1] and the deep-cloning library.[2]
Runtime objects in Eiffel are accessible either indirectly through references or as expanded objects whose fields are embedded within the objects that use them. That is, fields of an object are stored either externally or internally.
The Eiffel class ANY contains features for shallow and deep copying and cloning of objects. All Eiffel classes inherit from ANY, so these features are available within all classes, and are applicable both to reference and expanded objects.
The copy feature effects a shallow, field-by-field copy from one object to another. In this case no new object is created. If y were copied to x, then the same objects referenced by y before the application of copy, will also be referenced by x after the copy feature completes.
To effect the creation of a new object which is a shallow duplicate of y, the feature twin is used. In this case, a single new object is created with its fields identical to those of the source.
The feature twin relies on the feature copy, which can be redefined in descendants of ANY, if necessary. The result of twin is of the anchored type like Current.
Deep copying and creating deep twins can be done using the features deep_copy and deep_twin, again inherited from class ANY. These features have the potential to create many new objects, because they duplicate all the objects in an entire object structure. Because new duplicate objects are created instead of simply copying references to existing objects, deep operations will become a source of performance issues more readily than shallow operations.
In C Sharp (C#), rather than using the interface ICloneable, a generic extension method can be used to create a deep copy using reflection. This has two advantages: First, it provides the flexibility to copy every object without having to specify each property and variable to be copied manually. Second, because the type is generic, the compiler ensures that the destination object and the source object have the same type.
In Objective-C, the methods copy and mutableCopy are inherited by all objects and intended for performing copies; the latter is for creating a mutable type of the original object. These methods in turn call the copyWithZone and mutableCopyWithZone methods, respectively, to perform the copying. An object must implement the corresponding copyWithZone method to be copyable. To perform the copying, the copying function may call the NSCopyObject() function on the object to get a byte-for-byte shallow copy.
In Objective Caml, the library function Oo.copy performs shallow copying of an object.
In PHP, the clone keyword performs shallow copying of an object.[1] Programmers may define a special method __clone() in an object to provide custom copying implementation.
In Python, the library's copy module provides shallow copy and deep copy of objects through the copy() and deepcopy() functions, respectively. Programmers may define special methods __copy__() and __deepcopy__() in an object to provide custom copying implementation.
In Ruby, all objects inherit two methods for performing shallow copies, clone and dup. The two methods differ in that clone copies an object's tainted state, frozen state, and any singleton methods it may have, whereas dup copies only its tainted state. Deep copies may be achieved by dumping and loading an object's byte stream or YAML serialization.[2]