Object-Oriented Software E ngineering Practical Software Development using UML and Java Design Patterns – Part 3 Sources: Chapter 6: Using Design Patterns, and Chapter 30 (parts) Designing the Logical Architecture with Patterns Craig Larman’s Text: Applying UML and Patterns, Chaps: 16, 22, & 23.
25
Embed
Object-Oriented Software E ngineering Practical Software Development using UML and Java Design Patterns – Part 3 Sources: Chapter 6: Using Design Patterns,
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Object-Oriented Software E ngineeringPractical Software Development using UML and Java
Design Patterns – Part 3Sources:
Chapter 6: Using Design Patterns, and
Chapter 30 (parts) Designing the Logical Architecture with Patterns
• For packages that represent subsystems, the most common pattern of access is Façade, a GoF design pattern.
• Have a public façade object defining the services for the subsystem• Clients collaborate with the façade, not internal subsystem components.
• The façade should not normally expose many low-level operations. Rather, it exposes a small number of high-level operations – the coarse-grained services.
• A façade does not normally do its own work. — Is consolidator/mediator to underlying subsystem objects, which do the work.— Façade is a wrapper - single point of access into the ‘rules engine’ of subsystem. — Other packages do not see implementation of subsystem - hidden behind façade.
• If the subsystem is implemented by collaborating with another package or subsystem, it may be that this other package is fine-grained operations. But again, only the high-level operations are shown.
Create special class called a <<Façade>>Façade simplifies the use of the package.Simply contains a simplified set of public methods.
Create a class containing the most often used / needed facilities in the package. Clients desiring services of this complex package can avail themselves of the Façade class, where it is likely what services they want are simpler to acquire. (sometime present…)
Since we are talking about a complex package, other clients that need some of the complexities of the package can access these services as they normally would – via the public interfaces of these other classes – if this is indeed a package and not a subsystem.
Airline reservation systems have many classes and methods.
Some design elements, like subsystems, may need to interact with the airline system but don’t want to be exposed to any changes that might be made to it (via dependencies).
So, we can define and Airline class to be a <<Façade>> that contains methods (access) to the most important query and booking operations in other classes in the airline system.
Thus any dependency arising from this relation is only on the Façade class.
Have a private class variable possibly called, ‘theInstance.’ This stores the instance itself.
Then have a public class method (static method) possibly called, ‘getInstance.’
First time method is called, it creates a single instance and stores it in theInstance. Subsequent calls simply return theInstance.
The public class method getInstance() makes this instance globally accessible. Note, effectively, the Singleton instance is effectively a global variable. Minimize these.
—When you have a two-way association created between two classes, the code for the classes becomes inseparable.
—If you want to reuse one class, then you also have to reuse the other. There is a dependency.
• Problem:
—How do you reduce the interconnection between classes, especially between classes that belong to different modules or subsystems? Can easily see how this increases coupling!
• Forces:
—You want to maximize the flexibility of the system to the greatest extent possible
So, what do we do? We create an abstract class we will call <<Observable>> that maintains a collection of <<Observer>> instances.
<<Observable>> class is very simple; it merely has a mechanism to add and remove observers (method addObserver) as well as a method, notifyObservers, that can send an update message to each <<Observer>> in the collection.
Now, any application class can declare itself to be a subclass of the <<Observable>> class.
<<Observer>> is an interface, defining only an abstract update method.
Any class can thus be made to observe an <<Observable>> by declaring that it implements the interface, and by asking to be a member of the observer list of the <<Observable>>.
The <<Observer>> can then expect a call to its update method whenever the <<Observable>> changes.
Using this pattern, the <<Observable>> neither has to know the nature of the number of classes that will be interested in receiving an update messages nor what they will do with this information.
WeatherViewer
* ******
Observers are notified when a new prediction is readyForecaster
Java has an Observer interface and an Observable class. This is a specificimplementation of this pattern.
Consider: a ‘forecast’ requires a lot ofcomputations. Once done, it ‘notifies’ all interested instances.
Forecaster is thus an observable object.
One observer object (Weather Viewer) might be an interface object responsible for displaying weather forecast; it needs to know when something happens; (another might be dependent on weather information to plan a schedule).
So, whenever forecaster changes, a message is sent to WeatherViewer!
Observer pattern is widely used to structure software cleanly into relatively independent modules. It is the basis of MVC architecture.
WeatherViewer
Observers are notified when a new prediction is readyForecaster
—You realize that another class has a method which provides the required service
—Inheritance is not appropriate (e.g. because the ‘isa’ rule does not apply
• Problem:
—How can you most effectively make use of a method that already exists in the other class?
• Forces:
—You want to minimize development cost by reusing methods; also reduce coupling between classes. Too, methods should be near the data, and the existing class with the method might be the more appropriate place for the existing method. So, …
• Overuse generalization and inherit the method that is to be reused instead of creating a single method in the «Delegator» that does nothing other than call a method in the «Delegate>>
• Handling by generalization involves much unneeded baggage created only to get access to this single method. Delegator is certainly an improvement.
We don’t want to directly incorporate the reused class into our inheritance hierarchy.Better: Use an <<Adapter>> class that is connected via association to the reused class. (Adaptee)
The polymorphic methods of <<Adapter>> delegate to methods of <<Adaptee>>, and delegatemethod in <<Adaptee>> may / may not have same name as delegating polymorphic method. Anything else in <<Adapter>> will be unaware that it is indirectly using the facilities of an instance of <<Adaptee>>. Importantly: the Adaptee is NOT added into the inheritance hierarchy of <<Superclass>>
Our existing hierarchy above… Here (below) is class whose method we wish to use.This class is already a part of its own inheritance hierarchy
We want to create a hierarchy of three-dimensional shapes, and we’d like to use an existing implementation of calcVolume in TimsTorus. (TimsTorus is already in its own hierarchy)We cannot modify the code in TimsTorus because it is being used by others, and thus cannot make it a subclass of ThreeDShape. (multiple inheritance?) So, we develop Torus (an <<Adapter>>) such that its instances have a link to the instance of TimsTorus and these instances can delegate calcVolume (and other operations) to TimsTorus.
Adapters are sometimes called Wrappers. Java wrapper classes Integer, Float, Double, etc are adapters for the Java primitive types.
—Often, it is time-consuming and complicated to create instances of a class (heavyweight classes).
—There is a time delay and a complex mechanism involved in creating the object in memory. Large classes must be loaded from a database to support usage.
• Problem: —How to reduce the need to create instances of a heavyweight
class? —Other objects may need to refer to or use instances of heavyweight
classes.—Many domain classes are heavyweight classes; also common
for the collection classes used to implement associations (like ArrayList or Vector) to be heavyweight classes too.
—So, how can we reduce the need to create instances of heavyweight classes and/or to load large numbers of them from a database or server when not all of them may be needed?
—We want all the objects in a domain model (like the heavyweight classes) to be available for programs to use when they execute a system’s various responsibilities.
—It is also important for many objects to persist from run to run of the same program and it is frequently very impractical for all the objects to be loaded into memory when a program starts.
—It would be much better to do the programming as if all objects were loaded into memory.
- Some programmers concerned with loading and storing the objects
- Some programmers concerned with implementing the responsibilities of the domain model.
Create a simpler version of the heavyweight class, which we call a Proxy, which has the same interface as the heavyweight class so programmers can declare variables without caring about whether the Proxy or its Heavyweight version will be put in the variable.
The Proxy object is really only a placeholder, and the operations in the Proxy delegate the operation to the HeavyWeight – if and when needed…. But application progresses as if heavyweights present!
Advantages: If/when needed, the Proxy can obtain the real heavyweight object. Proxy only needs to obtain the heavyweight one time and make it available
in memory to all others who use the proxy. (Can have many
occurrences of the proxy.) Some proxies may have implementations of a limited number of operations
that can be performed without the effort of loading the Heavyweight object.
Proxy: much simplier version of heavyweight class; contains same interface as heavyweight.
Note that both the Proxy and the Heavyweight implement the Interface.
The list elements will be loaded into local memory only when needed.
ListProxy PersistentList
«interface»Student
PersistentStudentStudentProxy
Here we have a variable that is to contain a List. This variable would, however, actually contain a ListProxy, since it would be expensive to load an entire list of objects into memory, and the list might not actually be needed. However, as soon as an operation accesses the list, the ListProxy might at that point create an instance of PersistentList. (Both ListProxy and PersistentList implement ListIF)
On the other hand, the ListProxy might be able to answer certain queries, such as the number of elements in the list (list size given via a class variable), without going to the effort of loading the PersistentList.
Imagine that the PersistentList is actually a list of students. These objects might also be proxies – in this case, instances of StudentProxy. Again, instances of PersistentStudent would only be loaded when necessary.