obsidian/Main

Fork 0

Files

Claudio Fritsche e440f128a5 vault backup: 2025-01-17 14:08:14

2025-01-17 14:08:14 +01:00

13 KiB

Raw Blame History

title, created_date, updated_date, aliases, tags, type, course_name, author, source, status

title

created_date

updated_date

aliases

Design Patterns by Construx

🏷️Tags : #01-2025 #book

Summary

[!summary] Summary 3 Sentences only!

What are the main ideas?

If I implemented one idea from this book right now, which one would it be?

How would I describe the course to someone else?

Ideas and Thoughts

[!info]+ Inspiring Questions

Did you think about other concepts from other resources?

Do the concepts fit to your past, to your memories?

Can you relive them and reflect them from a different angle?

Introduction

Getting Started

[!definition] UML Notation Contents

[!Definition] What is a Design Pattern A standard solution to a recurring problem.

We should not reinvent the wheel. Patterns are much more reusable than code.

A bridge is a design pattern because it is a standard solution to a recurring problem: we need to get things across a not passable obstacle. For every problem you have design criteria to choose the best type of bridge (e.g. span, clearance, loads, cost, etc.). In design patterns design criteria are called Goal forces and constraint forces. As a beginner: focus on the problem, not the solutions!

UML Notations:

Note

Class, Attribute and Operation

Operations are viewed from extern Methods are viewed from intern

Minimize Overall Complexity

Measures of complexity

Cyclomatic complexity: the number of decisions that are being made --> local complexity
Depth of decision making: embedding of decision within decision and decision --> local complexity
number of parameters: --> global complexity
fan out: number of functions that are called by the function --> global complexity

Local complexity is a measure on how complicated each function is, whereas the global complexity is the complexity in between functions. Remember to get an appropriate balance.

Use Abstractions

Syntax: Is all about structure Semantics: Is all about meaning

Colorless green dreams sleep furiously: correct syntax, completely wrong semantics --> defective or buggy code, because it is semantically meaningless

The compiler is a master of syntax but cannot know anything about semantics. The programmer needs to focus much more on semantics, because the complier will take care of the syntax.

[!NOTE] Abstraction the principle of ignoring those aspects of a subject that are not relevant to the current purpose in order to concentrate solely on those that are.

It is a tool to reduce and manage complexity.

what is the role of one function / subroutine? The more abstraction you have, the less performant the code becomes. Abstraction is not learnable. Good abstractions lead to clean code, which is maintainable and leads to less defects.

Encapsulate Design Decisions

Abstraction: permission to ignore Encapsulation: I prevent you from knowing Meaning they are not the same thing.

Goal:

hide design decision (data representation, algorithms, etc.)
Modules should be black boxes: programm to an interface, not through an interface

Idea: Design by Contract Requires vs Guarantees A function promises to give certain outputs depending on the inputs This is important for documentation of code if the code is used by different people. Docstrings for example. We only worry about semantic things. Explain exactly what the input values should be and what the return values will be. Explain exactly what errors and exceptions. Include Constraints on state (function may open file), include performance limits (use no more than 200 bytes per instance, etc). Defensive contract: requires less and guarantees more (check input for example). Contracts are great, because you can treat the function as a blackbox. When you modify the contract you need to inform all the users of the function about the updates of the contract

Desing by Contract is the means to encapsulation Use: Purpose, Requires and Guarantees

Maximize Cohesion, Minimize Coupling

Cohesion: Indivisibility of a given part: Does this function do exactly one thing? It should also not split functionality into two separate functions Coupling: Dependence between parts: Do not connect things that don't need to be connected. If you must connect things, connect them such that they can be disconnected as easily as possible.

Goal: Highly cohesive, very loosely coupled

Generally software solves problems by Decomposition, solution, recomposition. There are an infinite amount of different solutions, which have different quality. Cohesion and Coupling characterise the quality of the solution.

Always try to Decouple code:

Avoid	Do
Global variables (form of coupling)	Design by Contract
Friend (C++)	Dependency inversion: if you depend on things, depend on the most generic thing (type of car, not model)
Self-modifying code	Publish-subscribe (observer pattern)
Reflection	Law of Demeter

Law of Demeter

Principle of least knowledge. a.b().c() breaks the law --> violates coupling a.b() doesn't break the law

Design for Invariants, Design for Change

Product Families: attack different target markets with different value propositions with the same base platform: (pickup truck and suv example. For ford it's the same car (slight difference in build but 95% the same)). The core of the software design is made for the invariants, meaning all the things that do not change and are the same throughout the software. An example might be the base driver code and the base I2C/SPI interface, which all the drivers use. Every driver has different registers that need to be set with different values, but the core structure can be designed to be invariant. Design for Change means the same thing: to separate common from variable functionality: use templates, inheritance, conditional compilation, frameworks. Hide variation behind abstract interfaces: Outside sees a routine that moves data from A to B and doesn't care how it is being done. It can be done by ftp, scp, tcp/ip, bluetooth, etc. Design patterns that make abstract interfaces are adapter, bridge, strategy, factory method, abstract factory, template method, iterator, decorator, proxy, etc. The client side code does not need to be changed when the bridge implementation changes. Use delayed-binding strategies: C does early binding, python does very late binding. Examples: named constants, configuration/preference files, dependency injection/ function pointers, inversion of control, data-driven design (set of data to configure software, e.g xml files or so?), self-configuration (android: different hardware configurations. It pings the hardware upon boot and asks about specifications of the hardware.). A problem is that usually software is developed sequentially: first for customer A, then customer B, etc. Better would be to find out about more about the landscape before starting the development.

UML: Association and Multiplicity

Association: represents links between objects Multiplicity: constraints on number of links !

Multiplicity: 1 to many. If you are a customer must you have at least 1 order? The *-Notation means that it can also be 0. There are four cases which are noted differently:

Lower Bound	Upper Bound	UML Notation
0	1	0..1
0	Many	*
1	1	1
1	Many	1..*
Note that it can also be specific numbers: e.g. 3..7

UML Notation: Composition and Aggregation

Composition: exclusive membership: ♦ Aggregation: non-exclusive membership: ♢ When using composition and aggregation, be specific about the multiplicity. Diamonds are trendy and are often misused: you can specify multiplicity to represent the exact same thing as composition and aggregation.

UML: Sequence Diagram

Describes interactions between objects (dynamics). Primary Concepts: Objects, lifeline, activation context, message Naming: objectName:class, objectName, :class

! ---- is the lifeline time goes from top to bottom opt: optional [too cold] is a condition

UML: Inheritance and Abstraction

An arrow pointing to the base class is done by Inheritance !

An abstract class is a base class, which itself cannot create any objects, have any instances. It just creates an interface. If shape was abstract you could only create rectangles and circles, but no other shapes. Abstract methods within abstract classes need to be implemented

Design Principle: Liskov Substitutability

T (superclass) <-- S (subclass) Goal: objects of class T may be replaced with objects of class S without altering any desirable properties of that program (e.g. correctness)

preconditions in subclass S as strict or less strict than in superclass T
postconditions in subclass S as strict or less strict than in superclass T
No new exceptions should be introduced in subclass S --> A subclass S must require no more and promise no less than its superclass T

Example: the contract requires must be the same or less strict (preconditions) and for the guarantees the subclass must guarantee the same or more than the superclass T (postconditions)

Be cautious about accidental polymorphisms: a completely different class that uses the same names. Usually the requires and guarantees are very different. This means syntax is the same, but semantics are different. Syntactically substitutionable but not semantically substitubable.

Design Principle: Favor Association over Inheritance

Inheritance cannot change the behaviour at run-time: a rectangle shape cannot morph into a circle shape.

Design Patterns

Patterns are not mutually exclusive and can be used on top of each other.

Adapter Pattern

No prerequisits
Usually problems have same semantic, but a different syntax.
Reasons: too much client code to change, code not accessible, many clients that use the same code
Adapters are an intermediate code block that adapts the syntax between the two parts.
- meaning you have one more layer, thus slightly less performance.
!
- use an adapter base class (PCLinuxGraphicsAdapter)
There is no guarantee that it is only the syntax that changes, maybe you need to adapt the semantics as well.
Object vs class adaption.
- In multiple inheritance classes like C++ you inherit from both Target and Adaptee
- !
Relationship to fundamental design principles
- design to invariants/desing for change
- encapsulation
- liskov substitutability
- high cohesion/loose coupling

[!Important] Key Points

minimizes code changes

portability across multiple providers

two versions: object vs class adapter

Interesting application: In testing you might want to test hidden / private functions. So you can create a inherited object lets call it test-object, where those functions are exposed. So the test-object is basically an adapter.

Façade (also called Wrapper)

Build a wrapper around a complex service to simplify the interface.

Depending on the language you might not be able to preventing the client to go around the facade.
An adapter can be a wrapper. It all depends on your intent: If your intent is syntax adapting it is an adapter. If your intent is hiding complexity it is a wrapper. It can be both at the same time if your intent is to do both at the same time

Design principles

Encapsulation
Abstraction (hide details that are not needed)
High cohesion / loose coupling (secondary, decouple the service provider)

Bridge

Prerequisite: Adapter Pattern

Strategy

Composite

Observer

Template Method

Factory Method

Abstract Factory

Singleton

Iterator

Proxy

Decorator

Command

State

Data Access Object

Exercises

Design Patterns Construx 
-file: "Design Patterns by Construx.md"

13 KiB Raw Blame History