Skip to content

Latest commit

 

History

History
240 lines (210 loc) · 9.45 KB

0012-abstract-classes.md

File metadata and controls

240 lines (210 loc) · 9.45 KB
Raw

Abstract classes

Introduction

Introduce classes that provide incomplete implementation and cannot be instantiated directly. To instantiate an abstract class one has to extend it with supplementing implementation first.

Motivation

Requests for this feature keep coming up from time to time. And while Haxe doesn't have such a feature even standard library uses this concept.

E.g. here is the doc for haxe.io.Input:

An Input is an abstract reader. See other classes in the haxe.io package for several possible implementations.

It could lead to inconsistencies like some abstract methods throw "not implemented" exceptions:

/**
	Read and return one byte.
**/
public function readByte():Int {
	#if cpp
	throw "Not implemented";
	#else
	return throw "Not implemented";
	#end
}

while others just do nothing:

/**
	Close the input source.

	Behaviour while reading after calling this method is unspecified.
**/
public function close():Void {}

And it's not obvious if an empty close method is a part of a valid implementation or if it's an abstract method, which must be overridden.

As a consequence I can't tell if Stdout implementation for Neko is missing close implementation by mistake. Because Stdin implementation in that same file has close implemented.

With an upcoming asys feature standard library is likely to get a few more abstract classes.

Five of Haxe targets have native abstract classes. Writing correct externs for such classes doesn't seem viable without macros.

Haxe developers use throw "Not implemented" and similar constructs to emulate abstract classes quite a lot. That means some of these runtime errors survive to production.

With a proper abstract classes all of that could be fixed and unified.

Detailed design

Class becomes an abstract class if it has abstract access modifier:

abstract class SomeAbstr {

It does not matter if a class has private modifier before or after abstract:

private abstract class SomeAbstr {
abstract private class AnotherAbstr {

Abstract method is a method that is declared without an implementation and has abstract access modifier:

abstract function implementMe():Void;

Abstract method is not allowed to have an implementation

abstract function implementMe2():Void { // Error: abstract method cannot have an implementation
	trace('hello');
}

abstract access modifier position among other access modifiers of a method does not matter:

abstract public function implementMe1():Void; //ok
public abstract function implementMe2():Void; //also ok

Static methods cannot be abstract:

abstract class SomeAbstr {
	static abstract function implementMe():Void; // Error: static functions cannot be abstract
}

Abstract classes may contain static fields just like normal classes:

abstract class SomeAbstr {
	static public var field:String = 'hello';
	static public function method():Void {}
}

Abstract classes may or may not contain abstract methods.

abstract class SomeAbstr {
	abstract function implementMe():Void;
}
abstract class AnotherAbstr { } // also ok

Abstract classes cannot be instantiated:

abstract class Main {
	static function main() {
		var m = new Main(); // Error: cannot instantiate abstract class Main
	}

	public function new() {}
}

Abstract classes may contain an implementation, but still cannot be instantiated:

abstract class Greeter {
	public final name:String;
	public function new(name:String) {
		this.name = name;
	}
	public function greet() trace('Hello, $name!');
}

class Main {
	static public function main() {
		new Greeter('world'); //Error: cannot instantiate abstract class Greeter
	}
}

If a class contains abstract methods, then the class itself must be declared abstract.

class Some {
	abstract function implementMe():Void; // Error: non-abstract class Some cannot contain abstract functions
}
abstract class Some {
	abstract function implementMe():Void; // ok
}

If a class extends an abstract class, but does not implement all abstract methods, then the class itself must be declared abstract:

abstract class Abstr {
	abstract function implementMe():Void;
}
class Concrete extends Abstr {} // Error : Concrete is missing an implementation for function implementMe
abstract class AnotherAbstr extends Abstr {} // ok

Abstract class is not required to implement all the methods of interfaces that class implements. Instead missing interface implementations are implicitly treated as abstract methods:

interface IFace {
	function implementMe():Void;
}
abstract class Abstr implements IFace {} // ok
class Concrete extends Abstr {} // Error : Concrete is missing an implementation for function implementMe

abstract access modifier cannot be used with final or inline access modifiers:

final abstract class Abstr { // Error: abstract class cannot be final
	final abstract function implementMe():Void; // Error: abstract function cannot be final
	inline abstract function doSomething():Void {} // Error: inline function cannot be abstract
	inline abstract function doSomething():Void; // Error: abstract function cannot be inlined
}

It's not allowed to inline abstract functions at call site:

abstract class Abstr {
	abstract function implementMe():Void;
}
//<...>
var a:Abstr = getAbstr();
inline a.implementMe(); // Error: abstract function cannot be inlined

Interfaces and their fields cannot have abstract access modifier:

abstract interface IFace { // Error: interfaces don't need abstract modifier
	abstract function implementMe():Void; // Error: interface fields don't need abstract modifier
}

If a class provides an implementation for an abstract method it does not need override keyword as no actual implementation is overridden.

abstract class Abstr {
	abstract function implementMe():Void;
}
class Concrete extends Abstr {
	function implementMe():Void { //no "override"
		trace('implemented');
	}
}

Abstract classes may define abstract getters and/or setters:

abstract class Abstr {
	public var prop1(get,set):Int;
	function get_prop1() return Std.random(10);
	abstract function set_prop1(value:Int):Int;

	public var prop2(get,set):Int;
	abstract function get_prop2():Int;
	abstract function set_prop2(value:Int):Int;
}

Abstract classes don't imply any special rules for constructors.

AST changes

  1. Add isAbstract flag to type definitions:
  2. Add AAbstract constructor to haxe.macro.Access enum
  3. Add var isAbstract:Bool to haxe.macro.BaseType
  4. Add var isAbstract:Bool to haxe.macro.ClassField
  5. Make corresponding changes on the ocaml side.

Impact on existing code

Proper implementation as described in AST changes would introduce breaking changes for macro development. Although, proposed changes seem to be minor and should not affect most of Haxe projects.

Alternatively we can start with a meta @:abstract instead of a proper access modifier. Or parse an access modifier into a meta to avoid any breaking changes for macros while still having a proper access modifier in Haxe code.

Drawbacks

Using abstract keyword for this feature may be confusing especially in a documentation since this keyword is already used for Abstract types

But using any other keyword would be even more confusing since "abstract classes" is a common term for almost any OOP programming language.

This probably can be solved to some extent by always using "abstract" accompanied with the correct word (either "abstract type" or "abstract class") in docs or speeches to form a strong distinction between these two features in the "community mind".

Alternatives

It is possible to implement abstract classes with macros (most known implementation by @AndyLi).

But a macro implementation has obvious disadvantages:

  • Could not be used while developing other macros
  • Affects compilation time
  • Unwanted in the standard library
  • Not supported by the compiler to generate target-native abstract classes for better interoperability with the target platform.