Hareactive is a purely functional reactive programming (FRP) library for JavaScript and TypeScript. It is simple to use, powerful, and performant.

• Simple and precise semantics. This means that everything in the library can be understood based on a very simple mental model. This makes the library easy to use and free from surprises.
• Purely functional API.
• Based on classic FRP. This means that the library makes a distinction between behaviors and streams.
• Supports continuous time for expressive and efficient creation of time-dependent behavior.
• Integrates with declarative side-effects in a way that is pure, testable and uses FRP for powerful handling of asynchronous operations.
• Declarative testing. Hareactive programs are easy to test synchronously and declaratively.
• Great performance.

Hareactive is simple. It aims to have an API that is understandable and easy to use. It does that by making a clear distinction between semantics and implementation details. This means that the library implements a very simple mental model. By understanding this conceptual model the entire API can be understood.

This means that to you use Hareactive you do not have to worry about things such as "lazy observables", "hot vs cold observables" and "unicast vs multicast observables". These are all unfortunate concepts that confuse people and make reactive libraries harder to use. In Hareactive we consider such things implementation detail that users should never have to think about.

Hareactive implements what is called classic FRP. This means that it makes a distinction between two types of time dependent concepts. This makes code written in Hareactive more precise and easier to understand.

Hareactive is powerful. It features all the typical methods found in other FRP libraries. But on top of that it comes with many unique features that are rarely found elsewhere. For instance, continuous time.

• Installation
• Conceptual overview
• Tutorial/cookbook
• API documentation
• Contributing
• Benchmark

Hareactive can be installed from npm. The package ships with both CommonJS modules and ES6 modules

npm install @funkia/hareactive

Hareactive contains four key concepts: Behavior, stream, future and now. This section will describe each of these at conceptual level.

For a practical introduction into using Hareactive see the tutorial. Unless you're already familiar with classic FRP you should at least read the sections on behavior, stream and now before you dive into the tutorial.

A behavior is a value that changes over time. For instance, the current position of the mouse or the value of an input field is a behavior. Conceptually a behavior is a function from a point in time to a value. A behavior always has a value at any given time.

Since a behavior is a function of time we can visualize it by plotting it as a graph. The figure below shows two examples of behaviors. The left behavior is what we call a continuous behavior since it changes infinitely often. The right behavior only changes at specific moments, but it's still a function of time. Hareactive is implemented so that both types of behavior can be represented efficiently.

It is important to understand that behaviors are not implemented as functions. Although, in theory, they could be. All operations that Hareactive offers on behaviors can be explained and defined based on the understanding that a behavior is a function of time. It is a mental model that can be used to understand the library.

A Stream is a series of values that arrive over time. Conceptually it is a list of values where each value is associated with a moment in time.

An example could be a stream of keypresses that a user makes. Each keypress happens at a specific moment in time and with a value indicating which key was pressed.

Similarily to behaviors a stream can be visualized. But, in this case we wont get a graph. Instead we will get some points in time. Each point is called an occurrence. The value of an occurrence can be anything. For instance, the figure to the left may represent a stream of booleans where all the "low" stars represents an occurrence with the value false and the "high" stars represents true.

The difference between a stream and a behavior is pretty clear when we see them visually. A behavior has a value at all points in time where a stream is a series of events that happens at specific moments in time.

To understand why Hareactive features both behavior and stream you may want to read the blog post Behaviors and streams, why both?.

A future is a value associated with a certain point in time. For instance, the result of an HTTP-request is a future since it occurs at a specific time (when the response is received) and contains a value (the response itself).

Future has much in common with JavaScript's Promises. However, it is simpler. A future has no notion of resolution or rejection. That is, a specific future can be understood simply as a time and a value. Conceptually one can think of it as being implemented simply like this.

{time: 22, value: "Foo"}

The relationship between Future and Stream is the same as the relationship between having a variable of a type and a variable that is a list of that type. You wouldn't store a username as ["username"] because there is always exactly one username.

Similarly in Hareactive we don't use Stream to express the result of a HTTP-request since a HTTP-request only delivers a response exactly once. It is more precise to use a Future for things where there is exactly one occurrence and Stream where there may be zero or more.

At first, the difference between the three things may be tricky to understand. Especially if you're used to other libraries where all three are represented as a single structure (maybe called "stream" or "observable"). The key is to understand that the three types represent things that are fundamentally different. And that expressing different things with different structures is beneficial.

You could forget about future and use a stream where you'd otherwise use a future. Because stream is more powerful than future. In the same way you could always use arrays of values instead of just single values. But you don't do that because username = "foo" expresses that only one username exists whereas username = ["foo"] gives the impression that a user can have more than one username. Similarly one could forget about numbers and just use strings instead. But saying amount = 22 is obviously better than amount = "22" because it's more precise.

This is how to figure out if a certain thing is a future, a stream or a behavior:

1. Ask the question: "does the thing always have a current value?". If yes, you're done, the thing should be represented as a behavior.
2. Ask the question: "does the thing happen exactly once?". If yes, the thing should be represented as a future. If no, you should use a stream.

Below are some examples:

• The time remaining before an alarm goes off: The remaining time always have a current value, therefore it is a behavior.
• The moment where the alarm goes off: This has no current value. And since the alarm only goes off a single time this is a future.
• User clicking on a specific button: This has no notion of a current value. And the user may press the button more than once. Thus a stream is the proper representation.
• Whether or not a button is currently pressed: This always has a current value. The button is always either pressed or not pressed. This should be represented as a behavior.
• The tenth time a button is pressed: This happens once at a specific moment in time. Use a future.

Now represents a computation that should be run in the present moment. Hence the name "now". Now is perhaps the most difficult concept in Hareactive.

A value of type Now is a description of something that we'd like to do. Such a description can declare that it wants to do one of two things.

• Get the current value of behavior. This is done with the sample function. Since a Now-computation will always be run in the present it is impossible to sample a behavior in the past.
• Describe side-effects. This is done with functions such as perform and performStream. With these functions we can describe things that should happen when a stream occurs.

Most Hareactive programs are bootstrapped by a Now-computation. That is, they take the form.

const main = ...

runNow(main);

Now is closely tied to the concept of stateful behaviors which is the topic of the next section.

A notorious problem in FRP is how to implement functions that return behaviors or streams that depend on the past. Such behaviors or streams are called "stateful"

For instance accumFrom creates a behavior that accumulates values over time. Clearly such a behavior depends on the past. Thus we say that accumFrom returns a stateful behavior.

Implementing stateful methods such as accumFrom in a way that is both intuitive to use, pure and memory safe is very tricky.

When implementing functions such as accumFrom most reactive libraries in JavaScript do one of these two things:

• Calling accumFrom doesn't begin accumulating state at all. Only when someone starts observing the result of accumFrom is state accumulated. This is very counter intuitive behavior.
• Calling accumFrom starts accumulating state from when accumFrom is called. This is pretty easy to understand. But it makes accumFrom impure as it will not return the same behavior when called at different time.

To solve this problem Hareactive uses a solution invented by Atze van der Ploeg and presented in his paper "Principled Practical FRP". His brilliant idea gives Hareactive the best of both worlds. Intuitive behavior and purity.

The solution means that some functions return a value that, compared to what one might expect, is wrapped in an "extra" behavior. This "behavior wrapping" is applied to all functions that return a result that depends on the past. The before mentioned accumFrom, for instance, returns a value of type Behavior<Behavior<A>>.

Remember that a behavior is a value that depends on time. It is a function from time. Therefore a behavior of a behavior is like a value that depends on two moments in time. This makes sense for accumFrom because the result of accumulating depends both on when we start accumulating and where we are now.

To get rid of the extra layer of nesting we often use sample. The sample function returns a Now-computation that asks for the current value of a behavior. It has the type (b: Behavior<A>) => Now<A>. Using sample with accumFrom looks like this.

const count = sample(accumFrom((acc, inc) => acc + inc, 0, incrementStream));

Here count has type Now<Behavior<A>> and it represents a Now-computation that will start accumulating from the present moment.

The definition of higher-order FRP is that it allows for FRP primitives nested inside other FRP primitives. Combinations like streams of streams, behaviors of behaviors, streams of futures, and any others are possible.

The benefit of higher-order FRP is increased expressiveness that makes it possibe to express many real-world scenarios with ease. One example would be an application with a list of counters. Each counter has a value which can be represented as a Behavior<number>. A list of counters would then have the type Array<Behavior<number>>. If additionally the list itself can change (maybe new counters can be added) then the type whould be Behavior<Array<Behavior<number>>. This higher-order type nicely captures that we have a changing list of changing numbers.

The downside of higher-order FRP is that sometimes dealing with these nested types can be tricky. Hareactive provides a number of functions to help with this. The table below gives an overview.

This cookbook will demonstrate how to use Hareactive. The examples gradually increase in complexity. Reading from the top serves as an tutorial about the library.

Please open an issue if anything is unclear from the explanations given.

You can use the map method. For instance, if you have a behavior of a number you can square the number as follows. map returns a new behavior with all values of the original behavior passed through the function:

behaviorOfNumber.map((n) => n * n);

map is also available as a function instead of a method.

map((n) => n * n, behaviorOfNumber);

Yes. Streams also have a map method.

streamOfNumbers.map((n) => n * n);

The map function also works with streams.

map((n) => n * n, streamOfNumbers);

This is done with the combine method or the combine function.

combine(firstStream, secondStream);

You can similarly combine any number of streams:

combine(firstStream, secondStream, thirdStream, etcStream);

Behaviors always have a current value. So to combine them you will have to specify how to turn the two values from the two behaviors into a single value. You do that with the lift function.

For instance, if you have two behaviors of numbers you can combine them by adding their values together.

lift((n, m) => n + m, behaviorN, behaviorM);

You can also combine in this fashion any number of behaviors, which has to match the number of the function arguments:

lift((n, m, q) => (n + m) / q, behaviorN, behaviorM, behaviorQ);

You probably want stepperFrom:

const b = stepperFrom(initial, stream);

We've though of that. Hareactive comes with a function for doing just that:

streamFromEvent(domElement, "click");

Definitely. Yes. fromFunction takes an impure function and turns it into a behavior whose value at any time is equal to what the impure function would return at that time:

const localStorageBehavior = fromFunction(() => localStorage.getItem("foobar"));

Both streams and behaviors have a log method that logs to the console when something happens.

misbehavingStream.log();

Converts any value into a future that has "always occurred". Semantically Future.of(a) is equivalent to (-Infinity, a).

Converts a promise to a future.

Returns true if f is a future and false otherwise.

Adds a consumer as listener to a future. If the future has already occurred the consumer is immediately pushed to.

Empty stream.

This function does not exist. Use empty to create a dummy stream for testing purposes.

Returns true if s is a stream and false otherwise.

Applies a function-valued behavior to a stream. Whenever the stream has an occurrence the value is passed through the current function of the behavior.

Returns a stream with all the occurrences from s for which predicate returns true.

const stream = testStreamFromArray([1, 3, 2, 4, 1]);
const filtered = stream.filter((n) => n > 2);
filtered.semantic(); //=> [{ time: 1, value: 3 }, { time: 3, value: 4 }]

Returns a pair of streams. The first contains all occurrences from stream for which predicate returns true and the other the occurrences for which predicate returns false.

const whereTrue = stream.filter(predicate);
const whereFalse = stream.filter((v) => !predicate(v));
// is equivalent to
const [whereTrue, whereFalse] = split(predicate, stream);

Filters a stream by applying the predicate-valued behavior to all occurrences.

Whenever stream has an occurrence the current value of behavior is considered. If it is true then the returned stream also has the occurrence—otherwise it doesn't. The behavior works as a filter that decides whether or not values are let through.

A stateful scan.

Creates a stream that occurs exactly when s occurs. Every time the stream s has an occurrence the current value of b is sampled. The value in the occurrence is then replaced with the sampled value.

const stream = testStreamFromObject({
  1: 0,
  4: 0,
  8: 0,
  12: 0
});
const shot = snapshot(time, stream);
const result = testStreamFromObject({
  1: 1,
  4: 4,
  8: 8,
  12: 12
});
// short == result

Returns a stream that occurs whenever s occurs. At each occurrence the value from s and the value from b is passed to f and the return value is the value of the returned streams occurrence.

Takes a stream valued behavior and returns a stream that emits values from the current stream at the behavior. I.e. the returned stream always "shifts" to the current stream at the behavior.

function shift<A>(s: Stream<Stream<A>>): Now<Stream<A>>;

Takes a stream of a stream and returns a stream that emits from the last stream.

function shiftFrom<A>(s: Stream<Stream<A>>): Behavior<Stream<A>>;

Takes a stream of a stream and returns a stream that emits from the last stream.

changes<A>(b: Behavior<A>, comparator: (v: A, u: A) => boolean = (v, u) => v === u): Stream<A>;

Takes a behavior and returns a stream that has an occurrence whenever the behaviors value changes.

The second argument is an optional comparator that will be used to determine equality between values of the behavior. It defaults to using ===. This default is only intended to be used for JavaScript primitives like booleans, numbers, strings, etc.

Combines two streams into a single stream that contains the occurrences of both a and b sorted by the time of their occurrences. If two occurrences happens at the exactly same time then the occurrence from a comes first.

const s1 = testStreamFromObject({ 0: "#1", 2: "#3" });
const s2 = testStreamFromObject({ 1: "#2", 2: "#4", 3: "#5" });
const combined = combine(s1, s2);
assert.deepEqual(combined.semantic(), [
  { time: 0, value: "#1" },
  { time: 1, value: "#2" },
  { time: 2, value: "#3" },
  { time: 2, value: "#4" },
  { time: 3, value: "#5" }
]);

Returns true if obj is a stream and false otherwise.

isStream(empty); //=> true
isStream(12); //=> false

Returns a stream that occurs ms milliseconds after s occurs.

Returns a stream that after occurring, ignores the next occurrences in ms milliseconds.

The log method on streams logs the value of every occurrence using console.log. It is intended to be used for debugging streams during development.

The option prefix argument will be logged along with every value if specified.

myStream.log("myStream:");

Converts any value into a constant behavior.

This takes an impure function that varies over time and returns a pull-driven behavior. This is particularly useful if the function is contionusly changing, like Date.now.

Returns true if b is a behavior and false otherwise.

Takes a boolean valued behavior an returns a behavior that at any point in time contains a future that occurs in the next moment where b is true.

Creates a future than on occurence samples the current value of the behavior and occurs with that value. That is, the original value of the future is overwritten with the behavior value at the time when the future occurs.

From an initial value and a future value, stepTo creates a new behavior that has the initial value until next occurs, after which it has the value of the future.

Creates a new behavior that acts exactly like initial until next occurs after which it acts like the behavior it contains.

A behavior of a behavior that switches to the latest behavior from s.

A behavior of a behavior that switches to the latest behavior from s.

Creates a behavior whose value is the last occurrence in the stream.

The returned behavior initially has the initial value, on each occurrence in source the function is applied to the current value of the behaviour and the value of the occurrence, the returned value becomes the next value of the behavior.

Constructs a behavior based on a function. At any point in time the value of the behavior is equal to the result of applying the function to a sampling function. The sampling function returns the current value of any behavior.

moment is a powerful function that can do many things and sometimes it can do them in a way that is a lot easier than other functions. A typical usage of moment has the following form.

moment((at) => {
  ...
})

Above, the at function above can be applied to any behavior and it will return the current value of the behavior. The following example adds together the values of three behaviors of numbers.

const sum = moment((at) => at(aBeh) + at(bBeh) + at(cBeh));

The above could also be achieved with lift. However, moment can give better performance when used with a function which dynamically switches which behaviors it depends on. To understand this, consider the following contrived example.

const lifted = lift((a, b, c, d) => (a && b ? c : d), aB, bB, cB, dB);

Here the resulting behavior will always depend on both aB, bB, cB, dB. This means that if any of them changes then the value of lifted will be recomputed. But, if for instance, aB is false then the function actually only uses aB and there is no need to recompute lifted if any of the other behaviors changes. However, lift can't know this since the function given to it is just a "black box".

If, on the other hand, we use moment:

const momented = moment((at) => (at(aB) && at(bB) ? at(cB) : at(dB)));

Then moment can simply check which behaviors are actually sampled inside the function passed to it, and it uses this information to figure out which behaviors momented depends upon in any given time. This means that when aB is false the implementation can figure out that, currently, momented only depends on atB and there is no need to recompute momented when any of the other behaviors changes.

moment can also be very useful with behaviors nested inside behaviors. If persons is a behavior of an array of persons and is of the type Behavior<{ age: Behavior<number>, name: string }[]> then the following code creates a behavior that at any time is equal to the name of the first person in the array whose age is greater than 20.

const first = moment((at) => {
  for (const person of at(persons)) {
    if (at(person.age) > 20) {
      return person.name;
    }
  }
});

Achieving something similar without moment would be quite tricky.

A behavior whose value is the number of milliseconds elapsed since UNIX epoch. I.e. its current value is equal to the value got by calling Date.now.

The now-computation results in a behavior that tracks the time passed since its creation.

A behavior giving access to continuous time. When sampled the outer behavior gives a behavior with values that contain the difference between the current sample time and the time at which the outer behavior was sampled.

Integrate behavior with respect to time.

The value of the behavior is treated as a rate of change per millisecond.

Integrate behavior with respect to time.

The value of the behavior is treated as a rate of change per millisecond.

The log method on behaviors logs the value of the behavior whenever it changes using console.log. It is intended to be used for debugging behaviors during development.

If the behavior is a pull behavior (i.e. it may change infinitely often) then changes will only be logged every ms milliseconds.

The option prefix argument will be logged along with every value if specified.

myBehavior.log("myBehavior:");
time.map(t => t * t).log("Time squared is:", 1000);

The Now monad represents a computation that takes place in a given moment and where the moment will always be now when the computation is run.

Converts any value into the Now monad.

Run an asynchronous IO action and return a future in the Now monad that resolves with the eventual result of the IO action once it completes. This function is what allows the Now monad to execute imperative actions in a way that is pure and integrated with FRP.

Returns the current value of a behavior in the Now monad. This is possible because computations in the Now monad have an associated point in time.

Takes a stream of IO actions and return a stream in a now computation. When run the now computation executes each IO action and delivers their result into the created stream.

A variant of performStream where outdated IO results are ignored.

A variant of performStream where IO results occur in the same order.

Convert a future now computation into a now computation of a future. This function is what allows a Now-computation to reach beyond the current moment that it is running in.

Run the given Now-computation. The returned promise resolves once the future that is the result of running the now computation occurs. This is an impure function and should not be used in normal application code.

Contributions are very welcome. Development happens as follows:

Install dependencies.

npm install

Run tests.

npm test

Running the tests will generate an HTML coverage report in ./coverage/.

Continuously run the tests with

npm run test-watch

We also use tslint for ensuring a coherent code-style.

Get set up to running the benchmarks:

npm run build
./benchmark/prepare-benchmarks.sh

Run all benchmarks with:

npm run bench

Run a single benchmark with:

node benchmark/<name-of-benchmark>

For example

node benchmark/scan.suite