Proxy Pattern

Synopsis

Context

• in a different process
• in a disk archive
• compressed
• not computed yet

Forces

• Turning an object into a Proxy, or vice versa, should be transparent.

Solution

Make a Proxy object with the same interface as the subject and which holds a reference to the subject. This is usually not a conventional reference, but may be a

• host/port/object triple
• file name and offset
• piece of code for producing the subject

This solution looks like a Decorator and may be implemented by identical means but it is not accompanied by the same policy. A Decorator adds capabilities to an object; a Proxy is a bridge to an inaccessible object. Since a Proxy is not an addition to the object it stands for, there is no need for delegation as in the Decorator pattern. (However, there is still a use for the hot swap technique.)

Consequences

• The normal inaccessibility of the subject is transparent.

• Since the Proxy has the same interface as its subject, they may be freely interchanged.

• Subjects generally do not access their own Proxies. First, the Proxy is usually just as inaccessible to the subject as the subject is to the Proxy. Second, the Proxy is not a real object but just a "mirror image" of the subject.

Implementation

• A Proxy can cache stable information about the subject to postpone accessing it. Caching can also be done with a separate cache Decorator. Caching should be done with caution since the subject generally does not access the Proxy and so cannot invalidate the cache.

• The hot swap was introduced as a technique for transparently introducing and removing a Decorator. It is also useful for transparently introducing and removing a Proxy.

See Design Patterns for sample code.

Known Uses

• Proxies are commonly used to make objects appear local in distributed systems. For example, the client stub which initiates an RPC is a Proxy for the server function. NEXTSTEP extends this to objects via the NXProxy class. CORBA uses Proxies so that database objects can be treated as native language objects.

• Lazy evaluation in Scheme is based on Proxies which encode a pending computation. Once the value is needed, the computation is executed and the Proxy is replaced by the output. This technique is used to implement streams.

• Copy-on-write is a special kind of lazy evaluation used in Matlab. Where a matrix would normally be copied, only a Proxy is created. Reads to the Proxy get forwarded to the original matrix. Writes force the Proxy to replace itself with a real copy. This technique automatically cuts down on spurious copies and lets Matlab code be simpler without hampering performance.