Permalink
Switch branches/tags
Nothing to show
Find file Copy path
517 lines (366 sloc) 15.7 KB

Golang concepts from an OOP point of view

Introduction

Objectives of this document

  • Ease golang learning by associating golang-specific concepts with previously known concepts in the OOP field.

  • Promote golang usage by easing language understanding for people coming from a heavy OOP background

Why github?

This is a discovery process, I'm writing this document to help myself understand golang and maybe help others. This document is published in github. Pull requests are welcomed. There are a lot of things to improve, but please, please do not start a pull request with "Technically..."

Golang Concepts

Golang introduces words with a new golang-specific meaning, such as struct and interface. This is not bad, but sometimes it is nice to have a "translation" available to be able to understand golang-concepts by relating them to previously known concepts.

This is important in order to understand concepts of a new language. If you can translate a golang-word to previously known concepts, the learning is by far easier.

Cheat Sheet

Golang Classic OOP
struct class with fields, only non-virtual methods
interface class without fields, only virtual methods
embedding multiple inheritance AND composition
receiver implicit this parameter

Golang-struct is a class (non-virtual)

A golang-struct is a class with fields where all the methods are non-virtual. e.g.:

type Rectangle struct {
	Name          string
	Width, Height float64
}

func (r Rectangle) Area() float64 {
	return r.Width * r.Height
}

This can be read as (pseudo code):

class Rectangle
  field Name: string
  field Width: float64
  field Height: float64
  method Area() //non-virtual
     return this.Width * this.Height

Constructor

  • There is a zero-value defined for each core-type, so if you do not provide values at instantiation, all fields will have the zero-value

  • No specific in-class constructors. There is a generic constructor for any class instance, receiving a generic literal in a JSON-like format and using reflection to create instances of any class.

pseudo-code for the generic constructor:

function construct ( class,  literal)

  helper function assignFields ( object, literal)  //recursive
    if type-of object is "object"
        if literal has field-names
            for each field in object.fields
                if literal has-field field.name
                    assignFields(field.value,  literal.fields[field.name].value) //recurse
        else
        //literal without names, assign by position
            for n=0 to object.fields.length
                assignFields(object.fields[n].value,  literal.fields[n].value) //recurse
    else
        //atom type
        set object.value = literal.value


// generic constructor main body
var classInstance = new class
assignFields(classInstance, literal)
return classInstance

Constructors example:

package main

import . "fmt"

type Rectangle struct {
	Name          string
	Width, Height float64
}

func main() {
	var a Rectangle
	var b = Rectangle{"I'm b.", 10, 20}
	var c = Rectangle{Height: 12, Width: 14}

	Println(a)
	Println(b)
	Println(c)
}

Output:

{ 0 0}
{I'm b. 10 20}
{ 14 12}

Golang "embedding" is akin to multiple inheritance with non-virtual methods

By embedding a struct into another you have a mechanism similar to multiple inheritance with non-virtual members.

Let's call "base" the struct embedded and "derived" the struct doing the embedding.

After embedding, the base fields and methods are directly available in the derived struct. Internally a hidden field is created, named as the base-struct-name.

Note: The official documents call it "an anonymous field", but this add to confusion, since internally there is an embedded field named as the embedded struct.

Base fields and methods are directly available as if they were declared in the derived struct, but base fields and methods can be "shadowed"

Shadowing means defining another field or method with the same name (and signature) of a base field or method.

Once shadowed, the only way to access the base member is to use the hidden field named as the base-struct-name.

type base struct {
	a string
	b int
}

type derived struct {
	base // embedding
	d    int
	a    float32 //-SHADOWS
}

func main() {
	var x derived

	fmt.Printf("%T\n", x.a) //=> x.a, float32 (derived.a shadows base.a)

	fmt.Printf("%T\n", x.base.a) //=> x.base.a, string (accessing shadowed member)
}

All base members can be accessed via the hidden field named as the base-struct-name.

It is important to note that all inherited methods are called on the hidden-field-struct. It means that a base method cannot see or know about derived methods or fields. Everything is non-virtual.

When working with structs and embedding, everything is STATICALLY LINKED. All references are resolved at compile time.

Multiple embedding

type NamedObj struct {
	Name string
}

type Shape struct {
	NamedObj  //inheritance
	color     int32
	isRegular bool
}

type Point struct {
	x, y float64
}

type Rectangle struct {
	NamedObj            //multiple inheritance
	Shape               //^^
	center        Point //standard composition
	Width, Height float64
}

func main() {
	var aRect = Rectangle{
		NamedObj{"name1"},
		Shape{NamedObj{"name2"}, 0, true},
		Point{0, 0},
		20, 2.5
	}

	fmt.Println(aRect.Name)
	fmt.Println(aRect.Shape)
	fmt.Println(aRect.Shape.Name)
}

This can be read: (pseudo code)

class NamedObj
   field Name: string

class Shape
   inherits NamedObj
   field color: int32
   field isRegular: bool

class Rectangle
   inherits NamedObj
   inherits Shape
   field center: Point
   field Width: float64
   field Height: float64

Example:

In var aRect Rectangle:

  • aRect.Name and aRect.NamedObj.Name refer to the same field

  • aRect.color and aRect.Shape.color refer to the same field

  • aRect.name and aRect.NamedObj.name refer to the same field, but aRect.NamedObj.name and aRect.Shape.NamedObj.name are different fields

  • aRect.NamedObj and aRect.Shape.NamedObj are the same type, but refer to different objects

Method Shadowing

Since all golang-struct methods are non-virtual, you cannot override methods (you need interfaces for that)

If you have a method show() for example in class/struct NamedObj and also define a method show() in class/struct Rectangle, Rectangle/show() will SHADOW the parent's class NamedObj/Show()

As with base class fields, you can use the inherited class-name-as-field to access the base implementation via dot-notation, e.g.:

type base struct {
	a string
	b int
}

//method xyz
func (this base) xyz() {
	fmt.Println("xyz, a is:", this.a)
}

//method display
func (this base) display() {
	fmt.Println("base, a is:", this.a)
}

type derived struct {
	base // embedding
	d    int
	a    float32 //-SHADOWED
}

//method display -SHADOWED
func (this derived) display() {
	fmt.Println("derived a is:", this.a)
}

func main() {
	var a derived = derived{base{"base-a", 10}, 20, 2.5}

	a.display() // calls Derived/display(a)
	// => "derived, a is: 2.5"

	a.base.display() // calls Base/display(a.base), the base implementation
	// => "base, a is: base-a"

	a.xyz() // "xyz" was not shadowed, calls Base/xyz(a.base)
	// => "xyz, a is: base-a"
}

Multiple inheritance and The Diamond Problem

Golang solves the diamond problem by not allowing diamonds.

Since inheritance (embedded fields) include inherited field names in the inheriting class (struct), all embedded class-field-names should not collide. You must rename fields if there is a name collision. This rule avoids the diamond problem, by not allowing it.

Note: Golang allows you to create a "Diamond" inheritance diagram, and only will complain when you try to access a parent's class field ambiguously.

Golang methods and "receivers" (this)

A golang struct-method is the same as a class non-virtual method but:

  • It is defined outside of the class(struct) body
  • Since it is outside the class, it has an extra section before the method name to define the "receiver" (this).
  • The extra section defines this as an explicit parameter (The this/self parameter is implicit in most OOP languages).
  • Since there is such a special section to define this (receiver), you can also select a name for this/self. Idiomatic golang is to use a short var name with the class initials. e.g.:

Example

//class NamedObj
type NamedObj struct {
	Name string
}

//method show
func (n NamedObj) show() {
	Println(n.Name) // "n" is "this"
}

//class Rectangle
type Rectangle struct {
	NamedObj      //inheritance
	Width, Height float64
}

//override method show
func (r Rectangle) show() {
	Println("Rectangle ", r.Name) // "r" is "this"
}

Pseudo-code:

      class NamedObj
         Name: string

         method show
           print this.Name

      class Rectangle
         inherits NamedObj
         field Width: float64
         field Height: float64

         method show //override
           print "Rectangle", this.Name

Using it:

func main() {
	var a = NamedObj{"Joe"}
	var b = Rectangle{NamedObj{"Richard"}, 10, 20}

	a.show("Hello")
	b.show("Hello")
}

Output:

Hello I'm Joe
Hello I'm Richard
- I'm a Rectangle named Richard

Structs vs Interfaces

A golang-Interface is a class with no fields and ONLY VIRTUAL methods.

The interface in Golang is designed to complement structs. This is a very important "symbiotic" relationship in golang. Interface fits perfectly with structs.

You have in Golang:

Structs: classes, with fields, ALL NON-VIRTUAL methods

Interfaces: classes, with NO fields, ALL VIRTUAL methods

By restricting structs to non-virtual methods, and restricting interfaces to all-virtual methods and no fields. Both elements can be perfectly combined by embedding to create fast polymorphism and multiple inheritance without the problems associated to multiple inheritance in classical OOP

Interfaces

A golang-Interface is a class, with NO fields, and ALL VIRTUAL methods

Given this definition, you can use an interface to:

  • Declare a var or parameter of type interface.
  • implement an interface, by declaring all the interface virtual methods in a concrete class (a struct)
  • inherit(embed) a golang-interface into another golang-interface

Declare a var/parameter with type interface

By picturing an Interface as a class with no fields and only virtual abstract methods, you can understand the advantages and limitations of Interfaces

By declaring a var/parameter with type interface you define the set of valid methods for the var/parameter. This allows you to use a form of polymorphism via method dispatch

When you declare a var/parameter with type interface:

  • The var/parameter has no fields
  • The var/parameter has a defined set of methods
  • When you call a method on the var/parameter, a concrete method is called via method dispatch from a jmp-table. (polymorphism via method dispatch)
  • When the interface is used as a parameter type:
    • You can call the function with any class implementing the interface
    • The function works for every class implementing the interface (the function is polymorphic)

Note on golang implementation: The ITables used for method dispatch are constructed dynamically as needed and cached. Each class(struct) has one ITable for each Interface the class(struct) implements, so, if all classes(structs) implement all interfaces, there's a ITable for each class(struct)-Interface combination. See: Go Data Structures: Interfaces

Examples from How to use interfaces in Go , with commented pseudo-code

package main

import (
	"fmt"
)

/*
class Animal
   virtual abstract Speak() string
*/
type Animal interface {
	Speak() string
}

/*
class Dog
  method Speak() string //non-virtual
     return "Woof!"
*/
type Dog struct {
}

func (d Dog) Speak() string {
	return "Woof!"
}

/*
class Cat
  method Speak() string //non-virtual
     return "Meow!"
*/
type Cat struct {
}

func (c Cat) Speak() string {
	return "Meow!"
}

/*
class Llama
  method Speak() string //non-virtual
     return "LaLLamaQueLLama!"
*/
type Llama struct {
}

func (l Llama) Speak() string {
	return "LaLLamaQueLLama!"
}

/*
func main
  var animals = [ Dog{}, Cat{}, Llama{} ]
  for each animal in animals
     print animal.Speak() // method dispatch via jmp-table
*/

func main() {
	animals := []Animal{Dog{}, Cat{}, Llama{}}
	for _, animal := range animals {
		fmt.Println(animal.Speak()) // method dispatch via jmp-table
	}
}

The empty Interface

By picturing an Interface as a class with no fields and only virtual abstract methods, you can understand what the golang Empty Interface: "Interface{}" is.

Interface{} in golang is a class with no fields and no methods

But, since by definition all classes(structs) implement Interface{} it means that a var x Interface{} can hold any value

What can you do with a var x Interface{}? Well, initially, nothing, because you don't know the type of the concrete value stored inside the var x Interface{}

To actually use the value inside a var x Interface{} you must use a Type Switch, a type assertion, or reflection

There is no automatic type-conversion from Interface{}

Using struct embedding

When you use only struct embedding to create a multi-root hierarchy, via multiple inheritance, you must remember that all struct methods are non-virtual.

That's why a struct hierarchy is always faster. When no interfaces are involved, the compiler knows exactly what concrete function to call, so all calls are direct calls resolved at compile time. Also the functions can be inlined.

The problem is with second-level methods. You cannot alter the execution of a base-class second-level method, because all struct-methods are non-virtual. You will inherit fields and methods, but methods are always executed on the base-class context.

When working with structs and embedding, everything is STATICALLY LINKED. All references are resolved at compile time. There are no virtual methods in structs

Using interfaces

When you use only interface embedding to create a multi-root hierarchy, via multiple inheritance, you must remember that all interface methods are virtual.

That's why a interface hierarchy is always slower than structs. When interfaces are involved, the compiler does not know at compile time, what concrete function to call. It depends on the concrete content of the interface var/parameter, so all calls are resolved at run-time via method dispatch. The mechanism is fast, but not as fast as compile-time resolved concrete calls. Also, with an interface-method call, since the concrete function to call is unknown until run-time, the call cannot be inlined.

The advantage of Interfaces is that: with second-level methods. You can alter the execution of the base-interface second-level method, because all interface-methods are virtual. Since all calls are made via method-dispatch, methods are executed on the context of the actual instance.

To be continued...

Drafts: Type Switch Internals