Implementing Your Own Operators

You can implement your own Observable operators. This page shows you how.

If your operator is designed to originate an Observable, rather than to transform or react to a source Observable, use the create( ) method rather than trying to implement Observable manually. Otherwise, follow the instructions below.

If your operator is designed to transform the items emitted by a source Observable, follow the instructions under Transformational Operators below. If your operator is designed to alter the source Observable as a whole, follow the instruction under Compositional Operators below.

(Note: in Xtend, a Groovy-like language, you can implement your operators as extension methods and can thereby chain them directly without using the methods described on this page. See RxJava and Xtend for details.)

Compositional Operators

The following example shows how you can use the lift( ) operator to chain your custom operator (in this example: myOperator) alongside standard RxJava operators like ofType and map:

fooObservable = barObservable.ofType(Integer).map({it*2}).lift(new myOperator<T>()).map({"transformed by myOperator: " + it});

The following section shows how you form the scaffolding of your operator so that it will work correctly with lift( ).

Implementing Your Operator

Define your operator as a public class that implements the Operator interface, like so:

public class myOperator<T> implements Operator<T> {
  public myOperator( /* any necessary params here */ ) {
    /* any necessary initialization here */
  }

  @Override
  public Subscriber<? super T> call(final Subscriber<? super T> s) {
    return new Subscriber<t>(s) {
      @Override
      public void onCompleted() {
        /* add your own onCompleted behavior here, or just pass the completed notification through: */
        if(!s.isUnsubscribed()) {
          s.onCompleted();
        }
      }

      @Override
      public void onError(Throwable t) {
        /* add your own onError behavior here, or just pass the error notification through: */
        if(!s.isUnsubscribed()) {
          s.onError(t);
        }
      }

      @Override
      public void onNext(T item) {
        /* this example performs some sort of simple transformation on each incoming item and then passes it along */
        if(!s.isUnsubscribed()) {
          transformedItem = myOperatorTransformOperation(item);
          s.onNext(transformedItem);
        }
      }
    };
  }
}

Other Considerations

• Your operator should check its Subscriber's isUnsubscribed( ) status before it emits any item to (or sends any notification to) the Subscriber. Do not waste time generating items that no Subscriber is interested in seeing.
• Your operator should obey the core tenets of the Observable contract:
  • It may call a Subscriber's onNext( ) method any number of times, but these calls must be non-overlapping.
  • It may call either a Subscriber's onCompleted( ) or onError( ) method, but not both, exactly once, and it may not subsequently call a Subscriber's onNext( ) method.
  • If you are unable to guarantee that your operator conforms to the above two tenets, you can add the serialize( ) operator to it to force the correct behavior.
• Do not block within your operator.
• It is usually best that you compose new operators by combining existing ones, to the extent that this is possible, rather than reinventing the wheel. RxJava itself does this with some of its standard operators, for example:
  • first( ) is defined as take(1).single( )
  • ignoreElements( ) is defined as filter(alwaysFalse( ))
  • reduce(a) is defined as scan(a).last( )
• If your operator uses functions or lambdas that are passed in as parameters (predicates, for instance), note that these may be sources of exceptions, and be prepared to catch these and notify subscribers via onError( ) calls.
  • Some exceptions are considered "fatal" and for them there's no point in trying to call onError( ) because that will either be futile or will just compound the problem. You can use the Exceptions.throwIfFatal(throwable) method to filter out such fatal exceptions and rethrow them rather than try to notify about them.
• In general, notify subscribers of error conditions immediately, rather than making an effort to emit more items first.

Transformational Operators

TBD