GeezerVison

The World Wide Web, restored to 800 pixels of glory.

GeezerVison is showing thinkinginpython.com — Pattern Catalog · Thinking in Python

Contents

Chapter 39

Pattern Catalog

You will encounter most patterns first as a name someone uses in a design review, not as code. This catalog gives each name a one-line intent, so you can place it and look it up.

This chapter gathers the original Design Patterns (GoF) set, along with other patterns that the literature documents widely. It draws from Pattern-Oriented Software Architecture (POSA, Buschmann et al.), Patterns of Enterprise Application Architecture (Fowler), Enterprise Integration Patterns (Hohpe and Woolf), and the common distributed and cloud patterns that emerged later.

Listing a pattern here does not recommend it. Many overlap, some solve the same problem in different ways, and several exist only to work around limits of a particular language. The body of this book argues that those workarounds often dissolve in Python.

One overlapping pair is State and State Machine. State changes an object’s behavior when its internal state changes. The client chooses each new state. State Machine builds on State and drives an object through a known set of states in response to inputs. The client supplies the inputs. The machine chooses each next state.

The tables follow each source’s own grouping. One of those groupings is GoF’s Creational/Structural/Behavioral split, which the Design Patterns chapter accepts for Creational and questions for Structural and Behavioral. A few idioms below belong to no single source. They sit in three groups by what they share: language and implementation idioms (most of them tied to C++ or Java’s limits), functional idioms, and dependency supply (patterns that supply a collaborator from outside). Everything else sits in Other Patterns and Idioms, a catch-all section. The source and group tables list their rows alphabetically. GoF lists the classic patterns in the same order.

When this book covers a pattern, its name links to that coverage. A name without a link leaves that pattern to the literature.

Finding a Pattern by Problem

The tables below group by source, so use this section’s table when you know the problem but not the name.

If the problem is	Look at
Creating objects without naming their classes	Abstract Factory, Builder, Factory Method, Prototype, Registry, Plugin
Controlling access to another object	Proxy
Adding behavior to an object without changing its class	Decorator
Converting one interface into another a client expects	Adapter
Simplifying access to a subsystem	Façade
Proxying a service’s calls from a helper process	Ambassador, Sidecar
Swapping an algorithm at runtime	Strategy
Varying some steps of an algorithm while its outline stays the same	Template Method
Encapsulating a request as an object	Command
Passing a request along a chain until something handles it	Chain of Responsibility
Changing behavior when an object’s internal state changes	State
Driving an object through a known set of states	State Machine
Adding an operation without changing the classes it visits	Visitor
Resolving behavior from the runtime types of two objects	Double Dispatch
Stepping through a collection’s elements in order	Iterator
Structuring recursive or tree-shaped data	Composite, Interpreter, Visitor
Keeping the number of objects down	Flyweight, Multiton, Object Pool, Singleton
Saving and restoring state	Memento, Event Sourcing
Reacting to change	Observer, Publish-Subscribe Channel, Model-View-Controller
Coordinating concurrent work	Thread Pool, Producer-Consumer, Future/Promise, Structured Concurrency, Active Object, Reactor
Keeping a program running when a dependency fails	Circuit Breaker, Retry, Bulkhead, Timeout, Dead Letter Channel
Returning several named values from one call	Messenger, Data Transfer Object
Moving data across a boundary	Data Transfer Object, Message Translator, Gateway, Data Mapper
Persisting domain objects to a database	Active Record, Data Mapper, Repository, Table Module, Lazy Load, Unit of Work, Identity Map
Organizing application logic by request or use case	Transaction Script, Domain Model, Service Layer, Front Controller
Modeling a value, amount, or special case instead of null	Value Object, Money, Special Case, Null Object
Routing or transforming a message	Content-Based Router, Message Router, Splitter, Aggregator
Connecting an application to a messaging system	Message, Message Channel, Message Endpoint, Point-to-Point Channel
Supplying a collaborator from outside, or looking one up in a registry	Dependency Injection, Service Locator, Strategy

Creational (GoF)

Pattern	Intent
Abstract Factory	Create families of related objects without naming concrete classes.
Builder	Build a complex object in steps, keeping the assembly separate from the finished object.
Factory Method	Defer instantiation to a method so subclasses choose the concrete type.
Prototype	Create new objects by cloning an existing instance.
Singleton	Ensure a class has one instance with a single point of access.

Structural (GoF)

Pattern	Intent
Adapter	Convert one interface into another a client expects.
Bridge	Separate an abstraction from its implementation so both vary independently.
Composite	Treat individual objects and compositions of them uniformly through a tree.
Decorator	Attach responsibilities to an object dynamically by wrapping it.
Façade	Provide one simplified interface to a subsystem.
Flyweight	Share fine-grained objects to support large numbers of them efficiently.
Proxy	Provide a surrogate that controls access to another object.

Behavioral (GoF)

Pattern	Intent
Chain of Responsibility	Pass a request along a chain until a handler processes it.
Command	Encapsulate a request as an object, enabling queues, logging, and undo.
Interpreter	Represent a grammar and evaluate sentences written in it.
Iterator	Access the elements of a collection in order without exposing its structure.
Mediator	Route communication between objects through one place to reduce coupling.
Memento	Capture and restore an object’s state without breaking encapsulation.
Observer	Notify dependents automatically when an object changes state.
State	Change an object’s behavior when its internal state changes.
Strategy	Make a family of algorithms interchangeable at runtime.
Template Method	Define an algorithm’s skeleton, letting subclasses fill in steps.
Visitor	Add operations to an object structure without changing its classes.

Concurrency (POSA and others)

Pattern	Intent
Active Object	Decouple a method call from its execution by giving the object its own thread.
Balking	Refuse an action when the object is in the wrong state.
Double-Checked Locking	Skip the lock once a lazily initialized shared resource exists.
Future/Promise	Represent a result that becomes available later.
Guarded Suspension	Block a call until a precondition becomes true.
Half-Sync/Half-Async	Separate synchronous and asynchronous work, joined by a queue.
Leader/Followers	Let a pool of threads take turns receiving and handling events.
Monitor Object	Serialize access so one method at a time runs on an object.
Proactor	Dispatch the completion of asynchronous operations to handlers.
Producer-Consumer	Decouple work creation from processing through a shared queue.
Reactor	Dispatch incoming requests to handlers synchronously as they arrive.
Read-Write Lock	Allow concurrent readers but exclusive writers.
Structured Concurrency	Tie concurrent tasks to a scope that exits only after every task finishes, cancelling the rest when one fails.
Thread Pool	Reuse a set number of worker threads across many tasks.
Thread-Specific Storage	Give each thread its own copy of a value. `ContextVar` scopes the value to the context instead.

The name Reactor says little about what the pattern does. The pattern is a dispatcher. One loop waits on many event sources and calls the handler registered for each event. The original description of Reactor lists Dispatcher as another name for the pattern. This catalog keeps the name Reactor because the literature and the libraries use it. Both asyncio’s SelectorEventLoop and Twisted’s reactor implement Reactor.

Proactor is the counterpart of Reactor for completed operations. A Reactor calls a handler when a source is ready, and the handler then performs the read or write. A Proactor starts the read or write and calls the handler once the operation has finished. On Windows, asyncio’s default event loop is a Proactor, ProactorEventLoop, built on the operating system’s I/O completion ports.

Architectural (POSA)

Pattern	Intent
Blackboard	Let independent components cooperate through a shared data store.
Broker	Coordinate requests and replies between distributed components.
Layers	Stack responsibilities so each layer uses only the one beneath it.
Microkernel	Keep a minimal core and add capability through plug-ins.
Model-View-Controller (MVC)	Separate data, presentation, and input handling.
Pipes and Filters	Process a stream through a chain of independent transforms.
Presentation-Abstraction-Control (PAC)	Build interactive systems from cooperating agents, each split three ways.
Reflection	Let a program inspect and adjust its structure at runtime.

Enterprise Application (Fowler)

Pattern	Intent
Active Record	Wrap a table row in an object that loads and saves itself.
Data Mapper	Move data between objects and the database, so neither names the other.
Data Transfer Object (DTO)	Carry data between processes in one batched object.
Domain Model	Model business logic as a graph of objects.
Front Controller	Route all requests through a single handler.
Gateway	Wrap access to an external system in a simple interface.
Identity Map	Load each object once per session.
Lazy Load	Defer loading a persisted object until something needs it.
Money	Represent monetary amounts together with their currency.
Plugin	Select an implementation by naming its class in configuration rather than in code.
Registry	Keep one well-known object where the rest of the program looks up services or data.
Repository	Stand between the domain and the data store, presenting stored objects as a queryable collection.
Service Layer	Define an application boundary as a set of operations.
Special Case	Supply a subclass for a particular case, such as a missing or unknown customer, instead of a conditional at every use.
Table Module	Let one class handle all rows of a table.
Transaction Script	Organize logic as one procedure per request.
Unit of Work	Track changes in a transaction and commit them together.
Value Object	Model a small value as an immutable object compared by value, not identity.

Integration and Messaging (Hohpe and Woolf)

Pattern	Intent
Aggregator	Combine related messages into one.
Content-Based Router	Route by inspecting the message content.
Dead Letter Channel	Hold messages the messaging system cannot deliver.
Message	Package data to send over a channel.
Message Channel	Connect senders and receivers through a logical pipe.
Message Endpoint	Connect an application to the messaging system.
Message Router	Send a message to a destination chosen at runtime.
Message Translator	Convert a message from one format to another.
Point-to-Point Channel	Deliver a message to exactly one receiver.
Publish-Subscribe Channel	Broadcast a message to every subscriber to its topic.
Splitter	Break one message into several.

Distributed and Cloud

Pattern	Intent
Ambassador	Route a service’s outbound calls through a helper.
API Gateway	Offer one entry point in front of many services.
Bulkhead	Isolate resources so one failure leaves the rest of the system running.
Circuit Breaker	Stop calling a failing service until it recovers.
Command Query Responsibility Segregation (CQRS)	Separate the read model from the write model.
Event Sourcing	Store state as a log of events instead of current values.
Retry	Re-attempt a failed operation, often with backoff.
Saga	Run a long transaction as a series of compensable steps.
Service Discovery	Locate service instances dynamically.
Sidecar	Attach helper functionality to a service as a separate process.
Strangler Fig	Replace a legacy system incrementally by routing around it.
Timeout	Bound how long to wait for a response.

Language and Implementation Idioms

Pattern	Intent
Curiously Recurring Template Pattern (CRTP)	Parameterize a base class by the subclass that inherits from it.
Marker Interface	Tag a class with an empty interface to signal a capability.
Mixin	Add reusable behavior through multiple inheritance.
Pointer to Implementation (Pimpl)	Store a class’s implementation in a separate object reached through a pointer, so the implementation can change while clients keep their compiled code.
Resource Acquisition Is Initialization (RAII)	Acquire a resource in a constructor and release it in the matching destructor.

Functional Idioms

Pattern	Intent
Function Composition	Build a function by passing one function’s output to the next.
Memoization	Cache a function’s results keyed by its arguments.
Monad	Sequence computations inside a context such as optionality, error, or async.
Partial Application	Preset some of a function’s arguments and get a function expecting the rest.

Dependency Supply

Pattern	Intent
Dependency Injection	Supply an object’s collaborators from outside it.
Inversion of Control	Let a framework call your code rather than the reverse. Dependency Injection applies Inversion of Control to an object’s collaborators. The usual alternative to Dependency Injection is Service Locator. Service Locator keeps control in your code, which asks a registry for each collaborator.
Service Locator	Look up dependencies through a central registry.

Other Patterns and Idioms

Pattern	Intent
Borg (Monostate)	Let every instance share one set of state instead of sharing one instance.
Double Dispatch	Resolve behavior from the runtime types of two objects, through two calls.
Fluent Interface	Return the receiver from each method so calls chain into one expression.
Function Object	Decouple the choice of function to call from the place that calls it.
Lazy Initialization	Create a value on first use.
Messenger	Bundle values into one object with named attributes. Messenger is the in-process form of Data Transfer Object.
Multiton	Manage a pool of singletons, one per key.
Null Object	Use an object with neutral behavior in place of null.
Object Pool	Reuse expensive objects from a managed pool.
Specification	Encapsulate a rule as a predicate that combines with others.
State Machine	Drive an object through a known set of states in response to inputs.
Surrogate	Put a stand-in in front of the object that does the work. Proxy and State are both forms of Surrogate.
Type Object	Represent a “kind of” thing as data rather than a subclass.

Patterns Python Absorbed

Several entries above are in the catalog because the literature documents them, not because you need to write them. For each of them, Python provides a language feature or a standard-library tool in place of the classes the pattern describes.

Pattern	What Python gives you instead
Iterator	The iteration protocol, called for you by `for`
Singleton	A module, imported once and cached
Factory Method	A dictionary of classes, since a class is an object
Prototype	`copy.deepcopy()` and `copy.replace()`
Strategy	A function passed as an argument
Command	A function saved to run later
Chain of Responsibility	A list of functions, tried in order
Visitor	`functools.singledispatch`, or a `match` over a union of types
Flyweight	Interned strings and cached small integers

Each link leads to the section that shows the Python form of its pattern.

Python supplies each pattern’s structure, but the pattern’s intent remains. You still choose which algorithm to pass, which action to save for later, or which values to share.

You are visitor 01402520 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.