Skip to content

Design patterns

Design Patterns#

Table of Contents

What#

  • a model solution to common design problems
    • which we encouter often, occurs over-and-over
  • describes the problem, and a genereal approach to solve them

How we discovered?#

Why#

  • to ensure our work is
    • consistent
    • reliable
    • understandable
  • we dont want to re-invent the wheel, just check if for current problem there is design solution

When#

Classification#

  • Creational (helps create objects in oops world)
    • Builder
    • Factory
    • Abstract factory
    • Singleton
    • Prototype
    • Object Pool
  • Structural (helps design relationship between the objects)
    • Adaptor
    • Repository
    • Facade
    • Composite
  • Behavioral (helps design object interaction and responsibilites; control the operation of some objects)
    • Strategy / Policy
    • Observer / Dependant / PubSub
    • Command / Action / Transaction
  • Concurrency
  • No Pattern (python only?)
  • Mono-State Pattern (python only?)
  • Visitor Pattern

Builder Pattern#

What#

  • helps separate the construction of a complex object
    • e.g. a custom/assembled computer
  • encapsulates the construction of the object
    • ie. follows this: "encapsulate what varies" - hence single responsibility principle
  • allows multistep construction process (main property)

Why#

  • separates the "how" from "what"
  • assembly separates from components
  • encapsulates what varies - the parts
  • permits diff. representations
  • can leverage Director concept, e.g. to manage order of building attributes
  • builder adds parts to the products

How#

When#

Factory Pattern#

What#

  • define interface for creating an object
  • lets subclasses decide which object to create
    • it achieves it through factory() method
  • makes subclasses responsible for instantiation / defer instantiation to subclasses
  • also known as Virtual Constructor

Why#

  • when we want to build/instantiate and return at runtime based on some params

How#

When#

Thoughts#

  • i think dependency-inversion could not be solved

Abstract Factory#

What#

  • whereas Factory creates from one product class, AbstractFactory can produce from family of product class
  • enforces dependencies between concrete class
  • defer instantiation to subclasses
  • also knows as Kit pattern

Why#

How#

When#

Thoughts#

  • i think dependency-inversion could not be solved

Singleton Pattern#

What#

  • ensures to create a single object/instance of a class
    • .e.g. Web/DB connection pool, device access, buffer pools
    • sometimes based on implementation - objects (its memory loc) could be diff. but there state would remain same (see monostate e.g.)
  • responsible for creating that instance
  • provides a global point of access
  • can also facilitate "lazy instantiation", if instantiation is costly

Why#

  • read What
  • disallowing creation on new instance
  • reduces the global namespace, as it self it global
  • subsclassible for extended purposes
  • variable number of instances - as all of them are same only
    • e.g. base class + meta class variants
  • more flexible than a static class, since we can have multiple instances
    • (in python, static class is one with no instances)
  • monostate variant can share all/full state

How#

When#

Cons#

  • violates single-responsibilty principle
    • how? it does it own task + creates and stores itw own instance as well
  • non-standard class access / design (who knows there is .instance() static method)
  • harder to test ?? why??
  • carry a global state
  • hard to sub-class ?? why - just mandatoryily override the .instance() ?? else don't follow .instance() design
  • singletons are considered harmful by experts
    • and are called antipattern

Prototype#

What#

Why#

How#

When#

Cons#

Object Pool#

What#

Why#

How#

When#

Cons#

  • structural (helps design relationship between the objects)
    • adaptor
      • what
      • why
      • how
      • when
    • facade
      • what
      • why
      • how
      • when
    • composite
      • what
      • why
      • how
      • when

  • behavioral (helps design object interaction and responsibilites; control the operation of some objects)

    • strategy / policy
      • what
      • why
      • how
      • when
    • observer / dependant / pubsub
      • what
        • defines one-to-many relationship between a set of objects
          • so that when the state of one changes, all its dependents are notified
        • e.g. Lets say there is a News class to which many Reader can subscribe to
          • here News is a Subject class and Reader is an Observer class
          • many Observers can subscribe to the Subject
          • the Subject should have facility to attach, detach, and notify the Observers
        • we can implement a Push model using this pattern
      • why
        • separates the concerns subject, object, and main program
        • acheives SOLID
          • single-responsibilty
            • individual classes for each
          • interface-segregation
            • by creating abstract classes for Subject and Object
          • open-closed
            • Subject and Observer are open for extension but closed for modification
          • dependency-inversion
            • Subject and Observer are programmed towards abstraction and not implementation
      • note
        • its a good idea to handle automatic detach of observers once they finish
          • this will avoid memory leak and help automatic garbage collector
            • by removing dangling references
    • command / action / transaction
      • usage
        • CLI, GUI
      • why
        • encapsulate behaviors
        • seperate command logic from client
        • easiliy provide/provision additional capabilities
          • validate
          • undo
      • how
        • encapsulate a request as an object
          • and facilitates to parameterize an object for diff. requests
        • supports queues and logs
          • queues for:
            • updating a DB
            • executing a list of commands
        • can bake support for undoable operations and macros
          • macros (sequences of commands)

  • concurrency

    • what
    • why
    • how
    • when
  • no pattern (python only?)
    • what
    • why
      • in python helps avoid check for none
    • how
    • when
  • mono-state pattern (python only?)
    • what
      • every instance of the class should have the same state (say if state is a dict, then all objects will have same dict)
    • why
      • one ans. is to achieve singleton
    • how
      • create a class dict state; update that dict with instance inside __new__; add state to instance.__dict__
      • this way, first, second, .. last every instance will have same state
    • when
      • singleton
      • many other use cases
  • visitor pattern