Contract Definition – Multicast/Unicast – Simplified Types

README.md

@@ -19,7 +19,7 @@ The latest preview release is available on Maven Central as
 
 Handling streams of data—especially “live” data whose volume is not predetermined—requires special care in an asynchronous system. The most prominent issue is that resource consumption needs to be carefully controlled such that a fast data source does not overwhelm the stream destination. Asynchrony is needed in order to enable the parallel use of computing resources, on collaborating network hosts or multiple CPU cores within a single machine.
 
-The main goal of Reactive Streams is to govern the exchange of stream data across an asynchronous boundary—think passing  elements on to another thread or thread-pool—while ensuring that the receiving side is not forced to buffer arbitrary amounts of data. In other words, backpressure is an integral part of this model in order to allow the queues which mediate between threads to be bounded. The benefits of asynchronous processing would be negated if the communication of backpressure were synchronous (see also the [Reactive Manifesto](http://reactivemanifesto.org/)), therefore care has been taken to mandate fully non-blocking and asynchronous behavior of all aspects of a Reactive Streams implementation.
+The main goal of Reactive Streams is to govern the exchange of stream data across an asynchronous boundary – think passing  elements on to another thread or thread-pool — while ensuring that the receiving side is not forced to buffer arbitrary amounts of data. In other words, backpressure is an integral part of this model in order to allow the queues which mediate between threads to be bounded. The benefits of asynchronous processing would be negated if the communication of backpressure were synchronous (see also the [Reactive Manifesto](http://reactivemanifesto.org/)), therefore care has been taken to mandate fully non-blocking and asynchronous behavior of all aspects of a Reactive Streams implementation.
 
 It is the intention of this specification to allow the creation of many conforming implementations, which by virtue of abiding by the rules will be able to interoperate smoothly, preserving the aforementioned benefits and characteristics across the whole processing graph of a stream application.
 
@@ -34,67 +34,101 @@ In summary, Reactive Streams is a standard and specification for Stream-oriented
 
 The Reactive Streams specification consists of the following parts:
 
-**The SPI** defines the interoperablility layer between different implementations.
-
-**The API** specifies the types that the users of Reactive Stream libraries use.
+**The API** specifies the types to implement Reactive Streams and achieve interoperablility between different implementations.
 
 ***The Technology Compatibility Kit (TCK)*** is a standard test suite for conformance testing of implementations.
 
-Implementations are free to implement additional features not covered by the specification as long as they conform to the API and SPI requirements and pass the tests in the TCK.
+Implementations are free to implement additional features not covered by the specification as long as they conform to the API requirements and pass the tests in the TCK.
 
-#### Comparison with related technologies ####
+### API Components ###
 
-In contrast to reactive streams described in this document, a Future represents exactly one element (or a failure) that is produced asynchronosly while streams can provide a potentially unbounded number of elements.
+The API consists of the following components that are required to be provided by Reactive Stream implementations:
 
-Compared to Rx, the SPI described here prescribes a mandatory, non-blocking way to handle back-pressure and requires the processing of an element by a dowstream component to be dispatched asynchronously.
+ - Publisher
+ - Subscriber
+ - Subscription
 
-Iteratees are an abstraction used for consuming a stream, often for parsing it. In this sense they are not a stream transformation or combination tool in themselves.
+A *Publisher* is a provider of a potentially unbounded number of sequenced elements, publishing them according to the demand received from its Subscriber(s). 
 
-### SPI Components ###
+In response to a call to `Publisher.subscribe(Subscriber)` the possible invocation sequences for methods on the `Subscriber` are given by the following protocol:
 
-The SPI consists of components that are required to be provided by Reactive Stream implementations but these interfaces should not be exposed to libraries or user code that *use* a Reactive Streams implementation. The reason for this is that the methods used on the SPI level have very strict and rather complex semantic requirements which are likely to be violated by end users.
+```
+onError | (onSubscribe onNext* (onError | onComplete)?)
+```
 
-The components of the SPI are:
+- The number of `onNext` events emitted by a `Publisher` to a `Subscriber` will at no point in time exceed the cumulative demand that has been signaled via that `Subscriber`’s `Subscription`.
+- A `Publisher` can send less events than requested and terminate the `Subscription` by calling `onComplete` or `onError`.
+- Events sent to a `Subscriber` can only be sent sequentially (no concurrent notifications).
+- If a `Publisher` fails it must emit an `onError`.
+- If a `Publisher` terminates successfully (finite stream) it must emit an `onComplete`.
+- If a Publisher signals either `onError` or `onComplete` on a `Subscriber`, that `Subscriber`’s `Subscription` must be considered canceled.
+- Once a terminal state has been signaled (`onError`, `onNext`) no further events can be sent.
+- Upon receiving a `Subscription.cancel` request it should stop sending events as soon as it can. 
+- `Subscription`'s which have been canceled should not receive subsequent `onError` or `onComplete` events, but implementations will not be able to strictly guarantee this in all cases due to the intrinsic race condition between actions taken concurrently by `Publisher` and `Subscriber`.
+- The `Publisher.subscribe` method can be called as many times as wanted as long as it is with a different `Subscriber` each time. It is up to the `Publisher` whether underlying streams are shared or not. In other words, a `Publisher` can support multi-subscribe and then choose whether each `Subscription` is unicast or multicast.
+- A `Publisher` can reject calls to its `subscribe` method if it is unable or unwilling to serve them (e.g. because it is overwhelmed or bounded by a finite number of underlying resources, etc...). It does this by calling `onError` on the `Subscriber` passed to `Publisher.subscribe` instead of calling `onSubscribe`".
+- A `Publisher` should not throw an `Exception`. The only legal way to signal failure (or reject a `Subscription`) is via the `Subscriber.onError` method.
+- The `Subscription.request` method must behave asynchronously (separate thread, event loop, trampoline, etc) so as to not cause a StackOverflow since `Subscriber.onNext` -> `Subscription.request` -> `Subscriber.onNext` can recurse infinitely. This allows a `Subscriber` to directly invoke `Subscription.request` and isolate the async responsibility to the `Subscription` instance which has responsibility for scheduling events.
+
+
+A *`Subscriber`* is a component that accepts a sequenced stream of elements provided by a `Publisher`. At any given time a `Subscriber` might be subscribed to at most one `Publisher`. It provides the callback `onNext` to be called by the upstream `Publisher`, accepting an element that is to be processed or enqueued without blocking the `Publisher`. 
+
+- `Subscriber` can be used once-and-only-once to subscribe to a `Publisher`.
 
- - Publisher
- - Subscriber
- - Subscription
 
-A *Publisher* is a provider of a potentially unbounded number of sequenced elements, publishing them according to the demand received from its Subscriber(s). A Publisher can serve multiple subscribers subscribed dynamically at various points in time. In the case of multiple subscribers the Publisher should respect the processing rates of all of its subscribers (possibly allowing for a bounded drift between them). It must eventually clean up its resources after all of its subscribers have been unsubscribed and shut down. A Publisher will typically support fanning out to multiple Subscribers in order to support the dynamic assembly of processing networks from building blocks that can freely be shared.
+A `Subscriber` communicates demand to the `Publisher` via a *`Subscription`* which is passed to the `Subscriber` after the subscription has been established. The `Subscription` exposes the `request(int)` method that is used by the `Subscriber` to signal demand to the `Publisher`. 
 
-A *Subscriber* is a component that accepts a sequenced stream of elements provided by a Publisher. At any given time a Subscriber might be subscribed to at most one Publisher. It provides the callback onNext to be called by the upstream Producer, accepting an element that is to be asynchronously processed or enqueued without blocking the Producer. 
+- A `Subscription` can be used once-and-only-once to represent a subscription by a `Subscriber` to a `Publisher`.
+- Calls from a `Subscriber` to `Subscription.request(int n)` can be made directly since it is the responsibility of `Subscription` to handle async dispatching.
 
-A Subscriber communicates demand to the Publisher via a *Subscription* which is passed to the Subscriber after the subscription has been established. The Subscription exposes the requestMore(int) method that is used by the Subscriber to signal demand to the Publisher. For each of its subscribers the Publisher obeys the following invariant:
+For each of its subscribers the `Publisher` obeys the following invariant:
 
-*If N is the total number of demand tokens handed to the Publisher P by a Subscriber S during the time period up to a time T, then the number of onNext calls that are allowed to be performed by P on S before T must be less than or equal to N. The number of pending demand tokens must be tracked by the Producer separately for each of its subscribers.*
+*If N is the total number of demand tokens handed to the `Publisher` P by a `Subscriber` S during the time period up to a time T, then the number of `onNext` calls that are allowed to be performed by P on S before T must be less than or equal to N. The number of pending demand tokens must be tracked by the `Producer` separately for each of its subscribers.*
 
-Subscribers that do not currently have an active subscription may subscribe to a Publisher. The only guarantee for subscribers attached at different points in time is that they all observe a common suffix of the stream, i.e. they receive the same elements after a certain point in time but it is not guaranteed that they see exactly the same number of elements. This obviously only holds if the subscriber does not cancel its subscription before the stream has been terminated.
+`Subscriber`s that do not currently have an active subscription may subscribe to a `Publisher`. The only guarantee for subscribers attached at different points in time is that they all observe a common suffix of the stream, i.e. they receive the same elements after a certain point in time but it is not guaranteed that they see exactly the same number of elements. This obviously only holds if the subscriber does not cancel its subscription before the stream has been terminated.
 
 > In practice there is a difference between the guarantees that different publishers can provide for subscribers attached at different points in time. For example Publishers serving elements from a strict collection (“cold”) might guarantee that all subscribers see *exactly* the same elements (unless unsubscribed before completion) since they can replay the elements from the collection at any point in time. Other publishers might represent an ephemeral source of elements (e.g. a “hot” TCP stream) and keep only a limited output buffer to replay for future subscribers.
 
-At any time the Publisher may signal that it is not able to provide more elements. This is done by invoking onComplete on its subscribers.
+At any time the `Publisher` may signal that it is not able to provide more elements. This is done by invoking `onComplete` on its subscribers.
 
-> For example a Publisher representing a strict collection signals completion to its subscriber after it provided all the elements. Now a later subscriber might still receive the whole collection before receiving onComplete.
+> For example a `Publisher` representing a strict collection signals completion to its subscriber after it provided all the elements. Now a later subscriber might still receive the whole collection before receiving onComplete.
 
-### API components ###
+### Asynchronous vs Synchronous Processing ###
 
-The purpose of the API is to provide the types that users interact with directly. SPI methods and interfaces should not be exposed expect for the purpose of writing Reactive Streams implementations.
+The Reactive Streams API prescribes that all processing of elements (`onNext`) or termination signals (`onError`, `onComplete`) do not *block* the `Publisher`. Each of the `on*` handlers can process the events synchronously or asynchronously. 
 
-The API counterpart for Publisher is *Producer* and for Subscriber is *Consumer*. The combination of these two—a stream processing element with asynchronous input and output—is called *Processor*.
+For example, this `onNext` implementation does synchronous transformation and enqueues the result for further asynchronous processing:
 
-The only operation supported by any Producer–Consumer pair is their ability to establish a connection for the purpose of transferring the stream of elements from Producer to Consumer; this is achieved by the method `produceTo()`. Concrete implementations of Reactive Streams are expected to offer a rich set of combinators and transformations, but these are not the subject of this specification. The reason is that implementations shall have the freedom to formulate the end-user API in an idiomatic fashion for the respective platform, language and use-case they target.
+```java
+void onNext(T t) {
+  queue.offer(transform(t));
+}
+```
 
-In addition there is one method each on Producer and Consumer to obtain a reference to the underlying Publisher or Subscriber, respectively. These are necessary for implementations, but is not to be considered end-user API.
+In a push-based model such as this doing asynchronous processing, back-pressure needs to be provided otherwise buffer bloat can occur.
 
-### Asynchronous processing ###
+In contrast to communicating back-pressure by blocking the publisher, a non-blocking solution needs to communicate demand through a dedicated control channel. This channel is provided by the `Subscription`: the `Subscriber` controls the maximum amount of future elements it is willing receive by sending explicit demand tokens (by calling `request(int)`).
 
-The Reactive Streams SPI prescribes that all processing of elements (onNext) or termination signals (onError, onComplete) happens outside of the execution stack of the Publisher. This is achieved by scheduling the processing to run asynchronously, possibly on a different thread. The Subscriber should make sure to minimize the amount of processing steps used to initiate this process, meaning that all its SPI-mandated methods shall return as quickly as possible.
+Expanding on the `onNext` example above, as the queue is drained and processed asynchronously it would signal demand such as this:
 
-In contrast to communicating back-pressure by blocking the publisher, a non-blocking solution needs to communicate demand through a dedicated control channel. This channel is provided by the Subscription: the subscriber controls the maximum amount of future elements it is willing receive by sending explicit demand tokens (by calling requestMore(int)).
+```java
+// TODO replace with fully functioning code example rather than this pseudo-code snippet
+void process() {
+   eventLoop.schedule(() -> {
+	    T t;
+   		while((t = queue.poll()) != null) {
+			doWork(t);
+			if(queue.size() < THRESHOLD) {
+				subscription.request(queue.capacity());
+			}
+		}
+   })
+}
+```
 
 #### Relationship to synchronous stream-processing ####
 
-This document describes asynchronous, non-blocking backpressure boundaries but in between those boundaries any kind of synchronous stream processing model is permitted. This is useful for performance optimization (eliminating inter-thread synchronization) and it conveniently transports backpressure implicitly (the calling method cannot continue while the call lasts). As an example consider a section consisting of three connected Processors, A, B and C:
+This document defines a protocol for asynchronous, non-blocking backpressure boundaries but in between those boundaries any kind of synchronous stream processing model is permitted. This is useful for performance optimization (eliminating inter-thread synchronization) and it conveniently transports backpressure implicitly (the calling method cannot continue while the call lasts). As an example consider a section consisting of three connected Processors, A, B and C:
 
     (...) --> A[S1 --> S2] --> B[S3 --> S4 --> S5] --> C[S6] --> (...)