• Public
  • Public/Protected
  • All



npm versionnpm downloads Twitter Follow

This project is part of the @thi.ng/umbrella monorepo.


Reactive streams & subscription primitives for constructing dataflow graphs / pipelines.

This library provides & uses three key building blocks for reactive programming:

  • Stream sources: event targets, iterables, timers, promises,watches, workers, manual-push...
  • Subscriptions: chained stream processors, each subscribable (one-tmany) itself
  • Transducers: stream transformers, either as individual subscription or to transform incoming values for a single subscription. See packages/transducers) for 100+ composable operators.
  • Recursive teardown: Whenever possible, and depending on configuration, unsubscriptions initiate cleanup and propagate to parent(s).
  • Workers: highly configurable, web worker integration for concurrent / parallel stream processing (fork-join, tunneled stream processing, etc.)

Conceptual differences to RxJS

(No value judgments implied - there's room for both approaches!)

  • Streams are not the same as Observables: I.e. stream sources are NOT (often just cannot) re-run for each new sub added. Only the first sub is guaranteed to receive all values. Subs added at a later time MIGHT not receive earlier emitted values, but only the most recent emitted and any future values
  • Every subscription supports any number of subscribers, which can be added/removed at any time
  • Depending on configuration options, every unsubscription recursively triggers upstream unsubscriptions (provided a parent has no other active child subscriptions)
  • Every subscription can have its own transducer transforming incoming values (possibly into multiple new ones)
  • Transducers can create streams themselves (only for merge() /sync())
  • Transducers can cause early stream termination and subsequent unwinding for its parent and downstream subscriptions.
  • Values can be manually injected into the stream pipeline / graph at any point
  • Unhandled errors in a subscription will move the subscription into an error state and cause unsubscription from parent (if any). Unhandled errors in stream sources will cancel the stream.
  • Much smaller API surface, since most common & custom operations can be solved via available transducers. Therefore there's less of a need to provide specialized functions (map / filter etc.) and gain more flexibility in terms of composing new operations.
  • IMHO less confusing naming / terminology (only streams (producers) & subscriptions (consumers))


STABLE - used in production

Search or submit any issues for this package

New features & breaking changes in 6.0.0

Completely revised & improved error handling, stronger distinction between .subscribe() and .transform() methods & internal simplification of their implementations.

  1. All error handlers now MUST return a boolean to indicate if the error was recoverable from or should put the subscription into the error state. See error handling for details.

  2. The options given to .transform() and .map() can now include an error handler:

// transform stream with given transducer(s)
// and forward any errors to `handleError` (user defined fn)
src.transform(xf1, xf2,..., { error: (e) => { ... } });

// or, also new, provide everything as single options object
// (for this version, see note (1) below)
src.transform({ xform: map(...), error: handleError });
  1. The .subscribe(sub, xform, opts) signature has been removed and the xform (transducer) must now be given as part of the options object:
const src = reactive(1);

// old
src.subscribe(trace("foo"), filter((x) => x < 10), { id: "child-sub" });

// new, see note (1) below
src.subscribe(trace("foo"), { xform: filter((x) => x < 10), id: "child-sub" });
  1. Added generics for PubSub topics, added .transformTopic() and updated signatures for .subscribeTopic(), both in similarity to above.
type Event = { id: string; value: any; };

const src = pubsub<Event>({ topic: (e) => e.id });

// transform topic stream with given transducer (see note (1) below)
// and forward any errors to `handleError` (user defined fn)
src.transformTopic("foo", map((e) => e.value), { error: handleError })


  • (1): If using multiple transducers, they must be pre-composed with comp(). Other signatures of .transform() method support up to 4 transducers and composes them automatically.

Support packages

Related packages

  • @thi.ng/atom - Mutable wrappers for nested immutable values with optional undo/redo history and transaction support
  • @thi.ng/hdom - Lightweight vanilla ES6 UI component trees with customizable branch-local behaviors
  • @thi.ng/rdom - Lightweight, reactive, VDOM-less UI/DOM components with async lifecycle and @thi.ng/hiccup compatible
  • @thi.ng/transducers - Lightweight transducer implementations for ES6 / TypeScript


yarn add @thi.ng/rstream

ES module import:

<script type="module" src="https://cdn.skypack.dev/@thi.ng/rstream"></script>

Skypack documentation

For NodeJS (v14.6+):

node --experimental-specifier-resolution=node --experimental-repl-await

> const rstream = await import("@thi.ng/rstream");

Package sizes (gzipped, pre-treeshake): ESM: 5.92 KB


Usage examples

Several demos in this repo's /examples directory are using this package.

A selection:

ScreenshotDescriptionLive demoSource
Interactive image processing (adaptive threshold)DemoSource
Figlet-style bitmap font creation with transducersDemoSource
Canvas based dial widgetDemoSource
Probabilistic color theme generatorDemoSource
Basic crypto-currency candle chart with multiple moving averages plotsDemoSource
Color palette generation via dominant color extraction from uploaded imagesDemoSource
Interactive inverse FFT toy synthDemoSource
Mouse gesture / stroke analysis, simplification, corner detectionDemoSource
Interactive pattern drawing demo using transducersDemoSource
Various hdom-canvas shape drawing examples & SVG conversion / exportDemoSource
Canvas based Immediate Mode GUI componentsDemoSource
Worker based, interactive Mandelbrot visualizationDemoSource
Minimal Markdown to Hiccup to HTML parser / transformerDemoSource
Basic rstream-gestures multi-touch demoDemoSource
Parser grammar livecoding editor/playground & codegenDemoSource
Interactive pixel sorting tool using thi.ng/color & thi.ng/pixelDemoSource
Demonstates various rdom usage patternsDemoSource
Dynamically loaded images w/ preloader stateDemoSource
rdom & hiccup-canvas interop testDemoSource
Full umbrella repo doc string search w/ paginated resultsDemoSource
rdom powered SVG graph with draggable nodesDemoSource
Animated Voronoi diagram, cubic splines & SVG downloadDemoSource
Minimal demo of using rstream constructs to form an interceptor-style event loopDemoSource
Interactive grid generator, SVG generation & export, undo/redo supportDemoSource
rstream based UI updates & state handlingDemoSource
rstream based spreadsheet w/ S-expression formula DSLDemoSource
Fork-join worker-based raymarch rendererDemoSource
hdom based slide deck viewer & slides from my ClojureX 2018 keynoteDemoSource
Transducer & rstream based hdom UI updatesDemoSource
WebGL cube maps with async texture loadingDemoSource


Generated API docs

Common configuration options

Since version 3.0.0 all stream and subscription factory functions take an optional object of configuration options with at least these keys (each optional):

interface CommonOpts {
* Internal ID associated with this stream. If omitted, an
* autogenerated ID will be used.
id: string;
* If false or `CloseMode.NEVER`, the stream stays active even if
* all inputs are done. If true (default) or `CloseMode.LAST`, the
* stream closes when the last input is done. If `CloseMode.FIRST`,
* the instance closes when the first input is done.
* @defaultValue CloseMode.LAST
closeIn: CloseMode;
* If false or `CloseMode.NEVER`, the stream stays active once there
* are no more subscribers. If true (default) or `CloseMode.LAST`,
* the stream closes when the last subscriber has unsubscribed. If
* `CloseMode.FIRST`, the instance closes when the first subscriber
* disconnects.
* @defaultValue CloseMode.LAST
closeOut: CloseMode;
* If true (default), stream caches last received value and pushes
* it to new subscriberswhen they subscribe. If false, calling
* `.deref()` on this stream will always return `undefined`.
* @defaultValue true
cache: boolean;

Stream creation


Source: stream()

Creates a new Stream instance, optionally with given StreamSource function and / or ID. If a src function is provided, the function will be only called (with the Stream instance as single argument) once the first subscriber has attached to the stream. If the function returns another function, it will be used for cleanup purposes if the stream is cancelled, e.g. if the last subscriber has unsubscribed. Streams are intended as (primarily async) data sources in a dataflow graph and are the primary construct for the various from*() functions provided by the package. However, streams can also be triggered manually (from outside the stream), in which case the user should call stream.next() to cause value propagation.

a = stream<number>((s) => {
// a 1
// a 2
// a done

// as reactive value mechanism
b = stream<number>();
// or alternatively
// b = subscription();


// external trigger
// b1 42
// b2 42

IDeref support

Stream (like all other types of Subscription) implements the @thi.ng/api IDeref interface which provides read access to a stream's last received value. This is useful for various purposes, e.g. in combination with @thi.ng/hdom, which supports direct embedding of streams (i.e. their values) into UI components (and will be deref'd automatically). If the stream has not yet emitted a value or if the stream is already done, it will deref to undefined.

Furthermore, all subscription types can be configured (via the cache option) to NOT retain their last emitted value, in which case .deref() will always return undefined.


Source: subscription()

Creates a new Subscription instance, the fundamental datatype & building block provided by this package (Streams are Subscriptions too). Subscriptions can be:

  • linked into directed graphs (if async, not necessarily DAGs)
  • transformed using transducers (incl. early termination)
  • can have any number of subscribers (optionally each w/ their own transducer)
  • recursively unsubscribe themselves from parent after their last subscriber unsubscribed
  • will go into a non-recoverable error state if NONE of the subscribers has an error handler itself
  • implement the @thi.ng/api IDeref interface
// as reactive value mechanism (same as with stream() above)
s = subscription<any, any>();
s.subscribe(trace("s2"), tx.filter((x) => x > 25));

// external trigger
// s1 23
// s2 42
// s1 42

Other stream creation helpers

Meta streams

Source: metaStream()

MetaStreams are streams of streams. A MetaStream is a subscription type which transforms each incoming value into a new stream, subscribes to it (via an hidden / internal subscription) and then only passes values from that stream to its own subscribers. If a new value is received, the meta stream first unsubscribes from the possibly still active stream created from the previous input, before creating and subscribing to the new stream. Hence this stream type is useful for cases where streams need to be dynamically and invisibly created & inserted into an existing dataflow topology without changing it, and with the guarantee that never more than one of these is active at the same time. Similar behavior (without the restriction in number) can be achieved using merge() (see further below).

The user supplied factory function will be called for each incoming value and is responsible for creating the new stream instances. If the function returns null / undefined, no further action will be taken (acts like a filter transducer, i.e. the incoming value is simply ignored).

// transform each received odd number into a stream
// producing 3 copies of that number in the metastream
// even numbers are ignored
a = metastream<number, string>(
(x) => (x & 1)
? fromIterable(tx.repeat("odd: " + x, 3), { delay: 100 })
: null


// odd: 23
// odd: 23
// odd: 23

a.next(42) // not odd, ignored by meta factory fn

// odd: 43
// odd: 43
// odd: 43

The factory function does NOT need to create new streams, but too can merely return other existing streams, and so making the meta stream act like a switch / stream selector.

If the meta stream is the only subscriber to these input streams, you'll need to use the closeOut: CloseMode.NEVER option when creating the inputs. This keeps them alive and allows for dynamic switching between them.

// infinite inputs
a = fromIterable(
{ delay: 1000, closeOut: CloseMode.NEVER }

b = fromIterable(
{ delay: 1000, closeOut: CloseMode.NEVER }

// stream selector / switch
m = metaStream((x) => x ? a : b);
m.subscribe(trace("meta from: "));

// meta from: a

// meta from: b

// meta from: a

Stream merging

Unordered merge from multiple inputs (dynamic add/remove)

Source: merge()


Returns a new StreamMerge instance, a subscription type consuming inputs from multiple inputs and passing received values on to any subscribers. Input streams can be added and removed dynamically. By default, StreamMerge calls done() when the last active input is done, but this behavior can be overridden via the close option, using CloseMode enums.

// input streams w/ different frequencies
src: [
fromIterable([1, 2, 3], { delay: 10 }),
fromIterable([10, 20, 30], { delay: 21 }),
fromIterable([100, 200, 300], { delay: 7 })
// 100
// 1
// 200
// 10
// 2
// 300
// 3
// 20
// 30

Use the labeled() transducer for each input to create a stream of labeled values and track their provenance:

src: [
fromIterable([1, 2, 3]).transform(tx.labeled("a")),
fromIterable([10, 20, 30]).transform(tx.labeled("b")),
// ["a", 1]
// ["b", 10]
// ["a", 2]
// ["b", 20]
// ["a", 3]
// ["b", 30]

See StreamMergeOpts for further reference of the various behavior options.

Adding inputs automatically

If the StreamMerge receives a Subscription-like value from any of its inputs, it will not be processed as usual, but instead will be added as new input to the merge and then automatically remove once that stream is exhausted.

// stream source w/ transducer mapping values to new streams
a = stream().transform(tx.map((x) => fromIterable(tx.repeat(x, 3))));
// simple 1Hz counter
b = fromInterval(1000);

merge({ src: [a, b] }).subscribe(trace());
// 0
// 1
// 2

// sent "a" will be transformed into stream via above transducer
// and then auto-added as new input to the StreamMerge
// a
// a
// a
// 3
// 4

Synchronized merge and labeled tuple objects

Source: sync()


Similar to StreamMerge above, but with extra synchronization of inputs. Before emitting any new values, StreamSync collects values until at least one has been received from all inputs. Once that's the case, the collected values are sent as labeled tuple object to downstream subscribers. Each value in the emitted tuple objects is stored under their input stream's ID. Only the last value received from each input is passed on. After the initial tuple has been emitted, you can choose from two possible behaviors:

  1. Any future change in any input will produce a new result tuple. These tuples will retain the most recently read values from other inputs. This behavior is the default and illustrated in the above schematic.
  2. If the reset option is true, every input will have to provide at least one new value again until another result tuple is produced.

Any done inputs are automatically removed. By default, StreamSync calls done() when the last active input is done, but this behavior can be overridden via the close constructor option, using CloseMode enums.

const a = stream();
const b = stream();
s = sync<any,any>({ src: { a, b } }).subscribe(trace("result: "));
// result: { a: 1, b: 2 }

Input streams can be added and removed dynamically and the emitted tuple size adjusts to the current number of inputs (the next time a value is received from any input).

If the reset option is enabled, the last emitted tuple is allowed to be incomplete, by default. To only allow complete tuples, also set the all option to false.

The synchronization is done via the partitionSync() transducer from the @thi.ng/transducers package. See this function's docs for further details.

See StreamSyncOpts for further reference of the various behavior options.

Stream splitting

Topic based splitting

Source: pubsub()


Topic based stream splitter. Applies topic function to each received value and only forwards it to child subscriptions for returned topic. The actual topic (return value from topic fn) can be of any type, apart from undefined. Complex topics (e.g objects / arrays) are allowed and they're matched with registered topics using @thi.ng/equiv by default (but customizable via equiv option). Each topic can have any number of subscribers.

If a transducer is specified for the PubSub, it is always applied prior to passing the input to the topic function. I.e. in this case the topic function will receive the transformed inputs.

PubSub supports dynamic topic subscriptions and unsubscriptions via subscribeTopic() and unsubscribeTopic(). However, the standard subscribe() / unsubscribe() methods are NOT supported (since meaningless here) and will throw an error! unsubscribe() can only be called WITHOUT argument to unsubscribe the entire PubSub instance (incl. all topic subscriptions) from the parent stream.

Splitting via predicate

Source: bisect()

Returns a new PubSub instance using given predicate pred as boolean topic function and a & b as subscribers for truthy (a) and falsy b values.

fromIterable([1, 2, 3, 4]).subscribe(
bisect((x) => !!(x & 1), trace("odd"), trace("even"))
// odd 1
// even 2
// odd 3
// even 4
// odd done
// even done

If a or b need to be subscribed to directly, then a / b MUST be first created as Subscription (if not already) and a reference kept prior to calling bisect().

const odd = subscription();
const even = subscription();
odd.subscribe(trace("odd x10"), tx.map((x) => x * 10));

fromIterable([1, 2, 3, 4]).subscribe(bisect((x) => !!(x & 1), odd, even));
// odd x10 10
// odd 1
// even 2
// odd x10 30
// odd 3
// even 4
// odd done
// odd x10 done
// even done


Input chunking / buffering, controlled by sidechain

Source: sidechainPartition()


Buffers values from src until side chain fires, then emits buffer (unless empty) and repeats process until either input is done. By default, the value read from the side chain is ignored, however the optional predicate can be used to only trigger for specific values / conditions.

// merge various event streams
fromEvent(document, "mousemove"),
fromEvent(document, "mousedown"),
fromEvent(document, "mouseup")
// queue event processing to only execute during the
// requestAnimationFrame cycle (RAF)

Input toggling, controlled by sidechain

Source: sidechainToggle()


Filters values from input based on values received from side chain. By default, the value read from the side chain is ignored, however the optional predicate can be used to only trigger for specific values/conditions. Every time the predicate fn returns true, the filter will be toggled on/off. Whilst switched off, no input values will be forwarded.

// use slower interval stream to toggle main stream on/off
// 0
// 3
// 4
// 7
// 8

Worker support

Parallel stream processing via workers

Source: forkJoin()


const $self: Worker = <any>self;
self.addEventListener("message", (e) => {
const { buf, factor } = e.data;
$self.postMessage(buf.map((x) => x * factor));
const src = stream<number[]>();

// fork worker jobs & re-join results
src: src,
// worker job preparation
// this function is called for each worker ID and the results
// of that function are the messages sent to the workers...
fork: (id, numWorkers, buf) => {
const size = (buf.length / numWorkers) | 0;
return {
buf: id < numWorkers - 1
? buf.slice(id * size, (id + 1) * size)
: buf.slice(id * size),
factor: id * 10
// re-join worker results
join: (parts) => <number[]>Array.prototype.concat.apply([], parts),
// worker script
worker: "./worker.js",
// default: navigator.hardwareConcurrency
numWorkers: 4

src.next(new Array(16).fill(1));

// result: [0, 0, 0, 0, 10, 10, 10, 10, 20, 20, 20, 20, 30, 30, 30, 30]

Stream processing via workers

Source: tunnel()

Delegate stream value processing to workers and pass on their responses to downstream subscriptions. Supports multiple worker instances and worker termination / restart for each new stream value received.

Source: postWorker()

Send values to workers (incl. optional (inline) worker instantiation)

Source: fromWorker()

Create value stream from worker messages.

Other subscription ops

  • debounce - ignore high frequency interim values
  • resolve - resolve on-stream promises
  • trace - debug helper
  • transduce - transduce or just reduce an entire stream into a promise
  • tween - stream interpolation

Error handling

Detailed information, discussion & diagrams about the new error handling can be found in this issue

The ISubscriber interface supports optional error handlers which will be called if code in the next() or done() handlers throws an error. If no error handler is defined for a subscriber, the wrapping Subscription's own error handler will be called, which might put this subscription into an error state and stop it from receiving new values.

src = subscription({ next(x) { throw x; } });

// triggers error, caught by subscription wrapper
// sub-0 unhandled error: 1

src.getState() === State.ERROR
// true

// no error, but also inputs won't be received/processed either

// another sub with error handler & indicating error could be handled
src = subscription({
next(x) { throw x; },
error(x) { console.warn("eeek", x); return true; }

// error caught by given handler
// eeek 1

// sub still usable, no error
src.getState() !== State.ERROR
// true

// further inputs still accepted
// eeek 2




If this project contributes to an academic publication, please cite it as:

  title = "@thi.ng/rstream",
  author = "Karsten Schmidt and others",
  note = "https://thi.ng/rstream",
  year = 2017


© 2017 - 2021 Karsten Schmidt // Apache Software License 2.0

Generated using TypeDoc