Turbine

A purely functional frontend framework based on functional reactive programming. Experimental.

Table of contents

• Why Turbine?
• Examples
• High-level overview
• Principles
• Installation
• More examples
• Tutorial
• API
• Contributing

Why Turbine?

The JavaScript world is full of frameworks. So why another one? Because we want something different. We want something that is purely functional without compromises. Something that takes the best lessons from existing JavaScript frameworks and couples them with the powerful techniques found in functional languages like Haskell. We want a framework that is highly expressive. Because when functional programming is at its best it gives you more power, not less. Turbine is supposed to be approachable for typical JavaScript developers while still preserving the benefits that comes from embracing purely functional programming.

We have done our best to realize our goal. But we are not done yet. We hope you will find Turbine interesting, try it and maybe even help us making it even better.

Examples

Email validator

See the example live and editable here.

const isValidEmail = (s: string) => /.+@.+\..+/i.test(s);

const app = component((on) => {
  const isValid = on.email.map(isValidEmail);
  return [
    span("Please enter an email address: "),
    input().use({ email: "value" }),
    div(["The address is ", isValid.map((b) => (b ? "valid" : "invalid"))])
  ];
});

// `runComponent` is the only impure function in application code
runComponent("#mount", main);

Counter

See the example live and editable here.

const counter = component((on, start) => {
  const count = start(accum((n, m) => n + m, 0, on.delta));
  return E.div([
    "Counter ",
    count,
    E.button({ class: "btn btn-default" }, " + ").use(o => ({
      delta: o.click.mapTo(1)
    })),
    E.button({ class: "btn btn-default" }, " - ").use(o => ({
      delta: o.click.mapTo(-1)
    }))
  ]);
});

// `runComponent` is the only impure function in application code
runComponent("#mount", counter);

See more examples here.

High-level overview

Here our some of our key features.

• Purely functional. A Turbine app is made up of only pure functions.
• Leverage TypeScript and runtime checking to improve the developing experience.
• Based on classic FRP. Behaviors represents values that change over time and streams provide reactivity. Turbine uses the FRP library Hareactive.
• A component-based architecture. Components are immutable, encapsulated and composable. Components are monads and are typically used and composed with do-notation (we implement do-notation with generators).
• Constructed DOM elements reacts directly to behaviors and streams. This avoids the overhead of using virtual DOM and should lead to great performance.
• Side-effects are expressed with a declarative IO monad. This allows for easy testing of code with side-effects. Furthermore, the IO-monad is integrated with FRP.
• The entire data flow through applications is explicit and easy to follow.
• Our libraries are available both as CommonJS and ES2015 modules. This allows for tree-shaking.

Here are some of the features we want to implement and goals we're working towards.

• Declarative and concise testing of time-dependent FRP code.
• Performance. We think Turbine can be made very efficient. But we are not yet at a point where we focus on performance.
• Support for server side rendering.
• Browser devtools for easier development and debugging.
• Hot-module replacement (if possible given our design).

Principles

This section describes some of the key principles and ideas underlying the design of Turbine.

Purely functional

Turbine is purely functional. We mean that in the most strict sense of the term. In a Turbine app, every single expression is pure. This gives a huge benefit in how easy it is to understand and maintain a Turbine app is.

One benefit of the complete purity is that every function in Turbine supports what is called "referential transparency". This means that an expression can always be replaced with its value.

As a simple example, say you have the following code:

const view = div([
  myComponent({ foo: "bar", something: 12 }),
  myComponent({ foo: "bar", something: 12 })
]);

One may notice that myComponent is called twice with the exact same arguments. Since all functions in a Turbine app are pure myComponent is no exception. Hence, we can make the following simple refactoring.

const component = myComponent({foo: "bar", something: 12}),
const view = div([
  component,
  component
]);

Such refactorings can always be safely done in Turbine.

Completely explicit data flow

One significant challenge when writing an interactive frontend application is how to manage the data flow through an application.

In Turbine we have strived to create an architecture where the data flow is easy to follow and understand. For us, this means that when looking at any piece of code it should be possible to see what other parts of the application it affects and what other parts it is affected by.

One manifestation of this principle is that in Turbine it is very simple to see how the model affects the view and how the view affects the model. The figure below illustrates this.

The arrows represent data flow between the model and the view. Note how these "conceptual arrows" are clearly expressed in the code. For instance, by looking at the buttons we can see exactly what output they produce.

Declarative models

Imperative programming is about doing. Functional programming is about being. This mean that ideally a functional program should be about defining what things are. That property is what makes functional programs declarative.

Below is a model from the counters example. Notice how the model consists of nothing but a series of const statements.

function* counterModel({ incrementClick, decrementClick, deleteClick }) {
  const increment = incrementClick.mapTo(1);
  const decrement = decrementClick.mapTo(-1);
  const deleteS = deleteClick.mapTo(id);
  const count = yield accum(add, 0, combine(increment, decrement));
  return { count, deleteS };
}

Each line is a declaration of a piece of the state. All models in Turbine follows this pattern. This makes state in a Turbine app very easy to understand. One can look at a single definition and be certain that it tells everything there is to know about that specific piece of state.

This is in sharp contrast to frameworks that mutate state or frameworks where state is stepped forward by reducer functions. With such approaches a single piece of state can potentially be affected and changed in several places. That can make it hard to understand how the state evolves. The benefits of having a definition as a single source of truth is lost.

Installation

npm install @funkia/turbine @funkia/hareactive

Hareactive is a peer dependency. It is the FRP library that that Turbine is based upon.

Alternatively, for quickly trying out Turbine you may want to see our Turbine starter kit.

More examples

Here is a series of examples that demonstrate how to use Turbine. Approximately listed in order of increasing complexity.

• Email validator — Very simple example of an email validator.
• Fahrenheit celsius — A converter between fahrenheit and celsius.
• Zip codes — A zip code validator. Shows one way of doing HTTP-requests with the IO-monad.
• Continuous time — Shows how to utilize continuous time.
• Counters — A list of counters. Demonstrates nested components, managing a list of components and how child components can communicate with parent components.
• Todo — An implementation of the classic TodoMVC application.

Tutorial

In this tutorial, we will build a simple application with a list of counters. The application will be simple but not completely trivial. Along the way, most of the key concepts in Turbine will be explained. We will see how to create HTML, how to create custom components, how a component can be nested and how it can share state with its parent.

Please open an issue if you have questions regarding the tutorial or ideas for improvements.

The final result and the intermediate states can be seen by cloning this git repository, going into the directory with the counters example and running webpack to serve the application.

git clone https://github.com/funkia/turbine/
cd turbine/examples/counters
npm run start

FRP

Turbine builds on top of the FRP library Hareactive. The two key concepts from FRP are behavior and stream. They are documented in more detail in the Hareactive readme. But the most important things to understand are behavior and stream.

• Behavior represents values that change over time. For instance, the position of the mouse or the number of times a user has clicked a button.
• Stream represents discrete events that happen over time. For instance click events.

What is Component

On top of the FRP primitives Turbine adds Component. Component is the key concept in Turbine. Once you understand Component—and how to use it—you understand Turbine. A Turbine app is just one big component.

Here is a high-level overview of what a component is.

• Components can contain logic expressed through operations on behaviors and streams.
• Components are encapsulated and have completely private state.
• Components contain output through which they selectively decide what state they share with their parent.
• Components write DOM elements as children to their parent. They can write zero, one or more DOM elements.
• Components can declare side-effects expressed as IO-actions.
• Components are composable—one component can be combined with another component and the result is a third component.

A Component in Turbine is pure and immutable. A Component can be thought of as a huge description of all of the above mentioned things. For instance, a Component contains a description about what its DOM look like. That part is a bit like virtual DOM. But, on top op that the description also explain how the DOM changes over time. The description also tells what output the Component contains. More on that later.

Creating HTML-elements

Turbine includes functions for creating components that represent standard HTML-elements. When you create your own components they will be made of these.

The element functions accept two arguments, both of which are optional. The first is an object describing various things like attributes, classes, etc. The second argument is a child component. For instance, to create a div with a span child we would write.

const myDiv = div({ class: "foo" }, span("Some text"));

The element functions are overloaded. So instead of giving span a component as child we can give it a string. The element functions also accept an array of child elements like this.

const myDiv = div({ class: "foo" }, [h1("A header"), p("Some text")]);

Using this we can build arbitrarily complex HTML. As an example we will build a simple view for a counter in our counter-application.

import { elements, runComponent } from "@funkia/turbine";
const { br, div, button } = elements;

// Counter
const counterView = div(["Counter ", 1, " ", button("+"), " ", button("-")]);

runComponent("body", counterView);

We define counterView as div-element with some text and two buttons inside. Since div returns a component counterView is a component. And a Turbine application is just a component so we have a complete application. We run the application on the last line when we call runComponent. It is an impure function that takes a selector, a component and runs the component with the found element as parent. You can view the entire code in version1.ts.

Dynamic HTML

The counterView above is completely static. The buttons do nothing and we hard-coded the value 1 into the view. Our next task is to make the program interactive.

Anywhere where we can give the element functions a constant value of a certain type we can alternatively give them a behavior with a value of that type. For instance, if we have a string-valued behavior we can use it like this

const mySpan = span(stringBehavior);

This will construct a component representing a span element with text content that is kept up to date with the value of the behavior.

To make the count in our counter view dynamic we turn it into a function that takes a behavior of a number and inserts it into the view.

const counterView = ({ count }: CounterViewInput) =>
  div(["Counter ", count, " ", button("+"), " ", button("-")]);

Because it will be easier going forward counterView takes an object with a count property.

Output from components

The above covers the input to the counter view. We now need to get output from it.

Remember that we mentioned how a Turbine component is a description about what the component will behave and look like. Part of that description also explains what output will come from the component.

To get a feel for what "output" means it may be helpful to mention a few examples.

• A button outputs, among other things, a stream of click events. So part of its output is a stream of the type Stream<ClickEvent>>.
• An input box's output includes a behavior of the text inside the input. The type would be Behavior<string>.
• A checkbox might output a behavior representing whether it is checked or not. It would have type Behavior<boolean>.

One way of looking at the output is that it is the information we would like to get from the view.

In practice a component will almost always output more than a single stream or behavior. By convention the output is therefore almost alway an object.

Components are represented by a generic type Component<O, A>. The A represents the available output of the component and the O represents the selected out of the component. The difference between selected and available output is highlighted in the example below.

Constructing an input element looks like this

const usernameInput = input({ placeholder: "Username" });

The type of the component constructed above is as follows ( the ... refer to the fact that we have omitted a lot of the output to keep things simple).

Component<{}, { value: Behavior<string>, click: Stream<ClickEvent>, ... }>

Among its available output an input element produces a string valued behavior named value that contains the current content of the input element.

Like this input component a newly constructed component always have {} as its selected output. This means that initially no output is selected. We can move output from the available output into the selected output by using the output method on components.

const usernameInput = input({
  attrs: { placeholder: "Username" }
}).output({ username: "value" });

Here usernameInput has the type

Component<{ username: Behavior<string> }, ...>

In the above code the invocation to output means: from the object of available output take the value property and add it to the object of selected output with the property name username.

The difference between available output and selected output matters when components are combined. In most cases, when components are composed or combined all their available output is discarded and only the selected output becomes part of the combined component.

For instance, in the code below the div is given two children.

div([
  button("Click me").output({ firstButtonClick: "click" }),
  button("Don't click me")
]);

The div element composes the two buttons. When doing so all output from the buttons except for the click stream from the first button is discarded.

Using the output method is a bit like adding event handlers in other UI frameworks. There are many events that one can add handlers to but on any given element only a few events are actually of interest and for these one will add event handlers. Similarly, in Turbine components have a lot of available output but only the piece of it that gets selected will be output in the end.

Back to the counters app. We want our counter view to produce two streams as output. One stream should be from whenever the first button is clicked and the other stream should contain clicks from the second button. That is, the view's output should have the type

{
  incrementClick: Stream<ClickEvent>,
  decrementClick: Stream<ClickEvent>
}

We can achieve that by using the output method in each button.

const counterView = ({ count }) =>
  div([
    "Counter ",
    count,
    " ",
    button("+").output({ incrementClick: "click" }),
    " ",
    button("-").output({ decrementClick: "click" })
  ]);

The call to output on each button tells them what output we are interested in. The first buttons selected output is then object with a stream named incrementClick and the later and object with one named decrementClick.

The div function then combines the selected output from the components in the array passed to it and output that as its own selected output. The result is that counterView returns a component that produces two streams as its output.

An analogy with promises

As mentioned above using the output method is a bit like adding event listeners in other frameworks. However, there are fundamental differences between the two things. If you are familiar with how asynchronous functions that takes callbacks differ from asynchronous function that returns promises then the following analogy may help understand this difference.

An asynchronous function for reading a file may look like this

readFileCallback("foo.txt", (file) => ...)

A similar function based on promises looks like this.

readFilePromise("foo.txt").then((file) => ...)

Notice that the readFileCallback function does not return the file that it reads. The file is instead passed to a callback that it gets as an argument. The readFilePromise function on the other hand returns the file wrapped in a promise of the type Promise<File>.

Most UI frameworks are similar to the readFileCallback function. In order to know when a button is pressed you do something like this.

<button onClick={(clickEvent) => ...}>Click me</button>

The click events on the button are not returned from the button function. Instead they are passed to a callback (or event handler) that the button function gets as an argumen.

The same thing in Turbine looks like this.

button("Click me").output({ click: "click" });

This is similar to the readFilePromise function. The button function does not take any callbacks but returns a stream of clicks wrapped in a component of the type Component<{ click: Stream<ClickEvent> }, ...>.

This example should give some intuition about how Turbine differs from most other frameworks. Other frameworks handle events similar to doing asynchronous computations with callbacks but Turbine handle events similarly to doing asynchronous computations with promises. In particular when creating components the output is returned as part of the component.

Adding a model

We now need to add a model with some logic to our counter view. The model needs to handle the increment and decrement stream and turn them into a behavior that represents the current count.

Turbine offers the function modelView for creating components with logic. modelView takes two arguments. The first describes the logic and the second the view. This keeps the logic neatly separated from the view.

The second argument to modelView, the view, is a function that returns a component. We already have such a function: counterView.

The first argument is a function that returns a Now-computation. You don't have to fully understand Now. One of the things it does is to make it possible to create stateful behaviors. The model function will as input receive the output from the component that the view function returns. The result of the Now-computation will be passed on to the view function and will be the output of the component that modelView returns. Here is how we use to create our counter component.

function* counterModel({ incrementClick, decrementClick }: CounterModelInput) {
  const increment = incrementClick.mapTo(1);
  const decrement = decrementClick.mapTo(-1);
  const changes = combine(increment, decrement);
  const count = yield accum((n, m) => n + m, 0, changes);
  return { count };
}

const counter = modelView(counterModel, counterView)();

Note that there is a cyclic dependency between the model and the view. The figure below illustrates this.

We now have a fully functional counter. You have now seen how to create a simple component with encapsulated state and logic. The current code can be seen in version2.ts.

Creating a list of counters

Our next step is to create a list of counters. To do that we will create a new component called counterList. The component will contain a list of counter components as well as a button for adding counters to the list.

Let's begin by defining a view function that creates a header and a button.

function* counterListView() {
  yield h1("Counters");
  const { click: addCounter } = yield button(
    { class: "btn btn-primary" },
    "Add counter"
  );
  return { addCounter };
}

We hook the view up to a model using modelView. Again, the model function receives the return value from the view function.

const counterList = modelView(counterListModel, counterListView);

const counterListModel = fgo(function*({ addCounter, listOut }) {
  const nextId = yield scan(add, 2, addCounter.mapTo(1));
  const appendCounterFn = map(
    (id) => (ids: number[]) => ids.concat([id]),
    nextId
  );
  const counterIds = yield accum(apply, [0], appendCounterFn);
  return { counterIds };
});

const counterListView = ({ sum, counterIds }) => [
  h1("Counters"),
  button({ class: "btn btn-primary" }, "Add counter").output({
    addCounter: "click"
  }),
  ul(list(counter, counterIds).output((o) => ({ listOut: o })))
];

const counterList = modelView(counterListModel, counterListView);

To create a dynamic list of counters we have to use the list function.

Documentation

Understanding generator functions

Turbine's use of generator functions may seem a bit puzzling at first. For instance, it may seem like generator functions serve two different purposes. One when they're used in the model and another when they're used in the view

But, what they do under the hood is exactly the same in both cases. The key to understand is that generator functions is just sugar for calling chain several times in succession.

When we use chain on components we can combine elements and pipe output from one component into the next. The code below combines two input elements with a span element that shows the concatenation of the text in the two input fields.

input({ attrs: { placeholder: "foo" } }).chain(({ value: aValue }) =>
  input().chain(({ value: bValue }) => {
    const concatenated = lift((a, b) => a + b, aValue, bValue);
    return span(["Concatenated text: ", concatenated]).mapTo({ concatenated });
  })
);

However, the above code is very awkward as each invocation of chain adds an extra layer of nesting. To solve this problem we use generators.

go(function*() {
  const { value: aValue } = yield input();
  const { value: bValue } = yield input();
  const concatenated = lift((a, b) => a + b, aValue, bValue);
  yield span(["Concatenated text: ", concatenated]);
  return { concatenated };
});

The above code does exactly the same as the previous example. But it is a lot easier to read!

The go function works like this. We yield a value with a chain method. go then calls chain on the yielded value. go calls chain with a function that continues the generator function with the value that chain passes it. The end result is a value of the same type that we yield inside the generator function. When we yield a Component<A> we will get an A back inside the generator function.

Finally we return a value and that value will be the output of the component that go returns.

Here is another example. The following code uses chain explicitly.

const view = button("Accept").chain(({ click: acceptClick }) =>
  button("Reject").map(({ click: rejectClick }) => ({
    acceptClick,
    rejectClick
  }))
);

The above code is equivalent to the following.

const view = go(function*() {
  const { click: acceptClick } = yield button("Accept");
  const { click: rejectClick } = yield button("Reject");
  return { acceptClick, rejectClick };
});

Again, the code that uses generator functions is a lot easier to read. This is why they're useful in Turbine.

Component is not the only type in Turbine that has a chain method. Now and Behavior does as well. And since go is only sugar for calling chain it works with these types as well.

API

The API documentation is incomplete. See also the examples, the tutorial, the Hareactive documentation and this tutorial about IO.

Component

Component#map

Mapping over a component is a way of applying a function to the output of a component. If a component has output of type A then we can map a function from A to B over the component and get a new component whose output is of type B.

In the example below input creates a component with an object as output. The object contains a behavior named value. The function given to map receives the output from the component.

We then call map on the behavior value and take the length of the string. The result is that usernameInput has the type Component<Behavior<number>> because it's mapped output is a number-valued behavior whose value is the current length of the text in the input element.

const usernameInput = input({ class: "form-control" }).map((output) =>
  output.value.map((s) => s.length)
);

Component#chain

map makes it possible to transform and change the output from a component. However, it does not make it possible to take output from one component and pipe it into another component. That is where chain enters the picture. The type of the chain method is as follows.

chain((output: Output) => Component<NewOutput>): Component<NewOutput>;

The chain method on a components with output Output takes a function that takes Output as argument and returns a new component. Here is an example. An invocation component.chain(fn) returns a new component that works like this:

• The output from component is passed to fn.
• fn returns a new component, let's call it component2
• The DOM-elements from component and component2 are both added to the parent.
• The output is the output from component2.

Here is an example.

input().chain((inputOutput) => span(inputOutput.value));

The above example boils down to this:

Create input component   Create span component with text content
  ↓                             ↓
input().chain((inputOutput) => span(inputOutput.value));
                   ↑                              ↑
      Output from input-element       Behavior of text in input-element

The result is an input element followed by a span element. When something is written in the input the text in the span element is updated accordingly.

loop

Sometimes situations arise where there is a cyclic dependency between two components.

For instance, you may have a function that creates a component that shows the value of an input string-value behavior and outputs a string-valued behavior.

const myComponent = (b: Behavior<string>) => span(b).chain((_) => input());

Now we'd have a cyclic dependency if we wanted to construct two of these views so that the first showed the output from the second and the second showed output from the first. With loop we can do it like this:

loop(({ output1, output2 }) =>
  go(function*() {
    const output1_ = yield myComponent(output2);
    const output2_ = yield myComponent(output1);
    return { output1: output1_, output2: output2_ };
  })
);

The loop functional seems pretty magical. It has the following signature (slightly simplified):

loop<A extends ReactiveObject>(f: (a: A) => Component<A>): Component<A>

I.e. loop takes a function that returns a component whose output has the same type as the argument to the function. loop then passes the output in as argument to the function. That is, f will as argument receive the output from the component it returns. The only restriction is that the output from the component must be an object with streams and/or behaviors as values.

Visually it looks like this.

modelView

The modelView functions makes it possible to create components where the view is decoupled from the model and its logic.

modelView takes two arguments:

• The model which is a function that returns a Now computation. The Now computation is run when the component is being created.
• The view which is a function that returns a Component.

modelView establishes a circular dependency between the model and the view. The model returns a Now computation and the result of this computation is passed into the view function. The view function then returns a component. The output of the component is passed to the model function.

Visually the circular dependency looks like this.

modelView returns a function that returns a component. The arguments given to this function will be passed along to both the model and the view functions. This makes it easy to create components that take input.

const myComponent = modelView(
  (outputFromView, arg1, arg2) => ...,
  (outputFromModel, arg1, arg2) => ...
);

myComponent("foo", "bar");

list

The list function is used to create dynamic lists in the UI.

Note: If you are familiar with frameworks like Angular or Vue then you can think of list as being similar to ngRepeat in Angular 1, ngFor in Angular 2, and v-for in Vue.

The list function has the following type.

function list<A, O>(
  componentCreator: (a: A) => Component<O, any>,
  listB: Behavior<A[]>,
  getKey: (a: A, index: number) => number | string = id
): Component<{}, Behavior<O[]>>;

The first parameter, componentCreator, is a function that takes a value of type A and returns a component. This function will be invoked to create the elements of the dynamic list. The second argument, listB, is a behavior of an array where the elements in the array are of some type A.

The list function will return a component that at any given point is time is equivalent to applying componentCreator to the current array in listB and then showing the resulting components one after another.

Whenever listB changes the component returned by list will react to those changes and keep the displayed list up-to-date. To do this, the last argument, the getKey function, is used to figure out how elements are moved, removed, or added. Therefore getKey should return a value that is unique for each element.

The following example illustrates the above. Let us say we have a list of users where each user is an object with an id and a username:

type User = {
  id: number;
  username: string;
};

The current list of users is represented by a behavior users: Behavior<User[]>. We want to display the users in a list with their username being editable. This can be achieved with the list function.

list((user) => input({ value: user.username }), users, (user) => user.id);

If the users behavior starts out with the value

[{ username: "foo", id: 1 }, { username: "bar", id: 2 }];

Then the component created by calling list will produce HTML like this

<input value="foo" /> <input value="bar" />

Now, if the value of users changes into

[
  { username: "baz", id: 3}
  { username: "bar", id: 2 }
  { username: "foo", id: 1 },
]

Then list will reorder the two existing input elements and insert a new input element in the beginning. Thanks to the getKey function list can efficiently do this by applying getKey to the old and the current value of the list and figure out how the elements have moved around.

SVG

You can use embed SVG in Turbine in much the same way you'd embed it in HTML:

svg({ height: "100", width: "100" }, [
  circle({
    cx: "50",
    cy: "50",
    r: "40",
    fill: "red"
  }),
  svgText({ x: 100, y: 30 }, "Hello SVG!")
]);

The only element with a different name is svgText because text in Turbine is an HTML Text Node.

Contributing

Turbine is developed by Funkia. We write functional libraries. You can be a part of it too. Share your feedback and ideas. We also love PRs.

Run tests once with the below command. It will additionally generate an HTML coverage report in ./coverage.

npm test

Continuously run the tests with

npm run test-watch