When to use mixins and when to use interfaces in D

问题:

I'm very familiar with the concepts of interfaces and abstract classes, but not super familiar with the concepts of mixins.

Right now, in Dart, every class A defines an implicit interface, which can be implemented by another class B by using the implements keyword. There's no explicit way of declaring interfaces as, for example, in Java, where an interface contains only unimplemented methods (and eventually static variables). In Dart, since interfaces are defined by classes, the methods of the interface A may actually already be implemented, but the class that implements B still needs to override these implementations.

We can see that situation from the following piece of code:

class A {
  void m() {
    print("method m");
  }
}

// LINTER ERROR: Missing concrete implementation of A.m
// Try implementing missing method or make B abstract.
class B implements A {
}

In Dart, a mixin is also defined via ordinary class declarations...

... In principle, every class defines a mixin that can be extracted from it. However, in this proposal, a mixin may only be extracted from a class that has no declared constructors. This restriction avoids complications that arise due to the need to pass constructor parameters up the inheritance chain.

A mixin is basically a class that can define both unimplemented or implemented methods. It's a way to add methods to another class without needing to logically use inheritance. In Dart, a mixin is applied to a super class, that is extended via "normal" inheritance, like in the following example:

class A {
  void m() {
    print("method m");
  }
}

class MyMixin {
  void f(){
    print("method f");
  }
}

class B extends A with MyMixin {
}

In this case, we should note that B does not have to implement any further methods both of A and MyMixin.

There's a clear distinction between applying a mixin to a class and inheriting from a class, at least in a language that supports only single-parent inheritance, since, in that case, we could apply many mixins to a class, but a class could just inherit from another class.

There's also a clear distinction between implementing an interface and inheriting from a class. The class that implements an interface needs to mandatorily implement all methods defined by the interface.

So, in summary, the concept of implementing an interface is more about establishing a contract with the class that implements the interface, and the concept of mixins (as the name suggests) is more about reusing code (without recurring to an inheritance hierarchy).

When to use mixins and when to use interfaces in Dart? Are there some rules of thumb for at least special recurrent patterns when designing software where it would be better to define a mixin and applying it to a super class rather than making our class implement an interface? I would appreciate concrete examples of design decisions in a context where both interfaces and mixins could be used, but one is used over the other (for some reason).

回答1:

Mixins is all about how a class does what it does, it's inheriting and sharing concrete implementation. Interfaces is all about what a class is, it is the abstract signature and promises that the class must satisfy. It's a type.

Take a class that is implemented as class MyList<T> extends Something with ListMixin<T> .... You can use this class as MyList<int> l = new MyList<int>(); or List<int> l = new MyList<int>(), but you should never write ListMixin<int> l = new MyList<int>(). You can, but you shouldn't, because that is treating ListMixin as a type, and it really isn't intended as one. It's the same reason you should always write Map m = new HashMap(); and not HashMap m = new HashMap(); - the type is Map, it's an implementation detail that it's a HashMap.

If you mix in a class (or rather, the mixin derived from a class), then you get all the concrete members of that class in your new mixin class. If you implement a class (or rather, the implicit interface of a class), then you get no concrete members at all, but the abstract signature becomes part of your interface.

Some classes can be used as both, but you should only ever use a class as a mixin if it is intended to be used as a mixin (and documented as such). There are many changes that an class author can do to a class that would break their use as a mixin. We don't want to disallow any such change, which could be perfectly reasonable changes for a non-mixin class, so using a non-mixin class as a mixin is fragile and likely to break in the future.

On the other hand, a class intended to be used as a mixin is usually all about implementation, so it is likely that there is a similar interface declared as well, and that's what you should use in the implements clause.

So, if you want to implement a list, you can either implement the List class and do all the implementation yourself, or mix in the ListMixin class to reuse some base functionality. You can still write implements List<T>, but you get that by inheritance from ListMixin.

Mixins is not a way to get multiple inheritance in the classical sense. Mixins is a way to abstract and reuse a family of operations and state. It is similar to the reuse you get from extending a class, but it is compatible with single-inheritance because it is linear. If you have multiple inheritance, your class has two (or more) superclasses, and you need to handle conflicts between them, including diamond inheritance, in some way.

Mixins in Dart work by creating a new class that layers the implementation of the mixin on top of a superclass to create a new class - it is not "on the side" but "on top" of the superclass, so there is no ambiguity in how to resolve lookups.

Example:

class Counter {
  int _counter = 0;
  int next() => ++_counter;
}
class Operation {
  void operate(int step) { doSomething(); }
}
class AutoStepOperation extends Operation with Counter {
  void operate([int step]) {
    super.operate(step ?? super.next());
  }
}

What really happens is that you create a new class "Operation with Counter". It's equivalent to:

Example:

class Counter {
  int _counter = 0;
  int next() => ++_counter;
}
class Operation {
  void operate(int step) { doSomething(); }
}
class $OperationWithCounter = Operation with Counter;
class AutoStepOperation extends $OperationWithCounter {
  void operate([int step]) {
    super.operate(step ?? super.next());
  }
}

The mixin application of Counter to Operation create a new class, and that class appears in the superclass chain of AutoStepOperation.

If you do class X extends Y with I1, I2, I3 then you create four classes. If you just do class X extends Y implements I1, I2, I3 then you only create one class. Even if all of I1, I2 and I3 are completely empty abstract interfaces, using with to apply them is equivalent to:

class $X1 extends X implements I1 {}
class $X2 extends $X1 implements I2 {}
class $X3 extends $X2 implements I3 {}
class X extends $X3 {}

You wouldn't write that directly, so you should write it using with either

回答2:

Languages such as Java and C# use interfaces to have multiple inheritance of type in lieu of multiple implementation inheritance. There are complexity tradeoffs that languages with multiple implementation inheritance (such as Eiffel, C++, or Dart) have to deal with that the designers of Java and C# chose to avoid.

However, once you have multiple implementation inheritance, there is no real need to separately support multiple interface inheritance, as an interface then becomes just a special case of an abstract class with no instance variables and only abstract methods and interface inheritance is the same as inheriting from such a class.

Example:

abstract class IntA {
  void alpha();
}

abstract class IntB {
  void beta();
}

class C extends IntA with IntB {
  void alpha() => print("alpha");
  void beta() => print("beta");
}

void main() {
  var c = new C();
  IntA a = c;
  IntB b = c;
  a.alpha();
  b.beta();
}

Dart has multiple implementation inheritance (via mixins), so it doesn't also need multiple interface inheritance as a separate concept, nor a way to separately define interfaces as standalone entities. Implicit interfaces (via the implements clause) are used to be able to document or verify that one class implements at least the same interface as another. For example, Int8List implements List<int>, although the underlying implementation is completely different.

Use of inheritance/mixins and implicit interfaces obtained through implements is generally orthogonal; you will most likely use them in conjunction rather than in lieu of one another. For example, you may want to use implements Set<int> to describe the desired interface of a bitset implementation, and then use extends and/or with clauses to pull in the actual implementation for that interface. The reason is that your bitset will not share any of the actual implementation with Set<int>, but you still want to be able to use them interchangeably.

The collection library provides a SetMixin mixin for us that only requires us to implement some basic routines ourselves and supplies the rest of the Set<T> implementation based on those.

import "dart:collection";

class BitSetImpl {
  void add(int e) { ...; }
  void remove(int e) { ...; }
  bool contains(int e) { ...; }
  int lookup(int e) { ...; }
  Iterator<int> get iterator { ...; }
  int get length { ...; }
}

class BitSet extends BitSetImpl with SetMixin<int> implements Set<int> {
  BitSet() { ...; }
  Set<int> toSet() { return this; }
}