xstate-tree was born as an answer to the question "What would a UI framework that uses Actors as the building block look like?". Inspired by Thomas Weber's Master Thesis on the topic. XState was chosen to power the actors with React powering the UI derived from the actors.
xstate-tree was designed to enable modeling large applications as a single tree of xstate machines, with each machine being responsible for smaller and smaller sub sections of the UI. This allows modeling the entire UI structure in state machines, but without having to worry about the complexity of managing the state of the entire application in a single large machine.
Each machine has an associated, but loosely coupled, React view associated with it. The loose coupling allows the view to have no knowledge of the state machine for ease of testing and re-use in tools like Storybook. Actors views are composed together via "slots", which can be rendered in the view to provide child actors a place to render their views in the parent's view.
While xstate-tree manages your application state, it does not have a mechanism for providing global application state accessible by multiple actors, this must be provided with another library like Redux or GraphQL. It does provide a routing solution, outlined here.
At Koordinates we use xstate-tree for all new UI development. Our desktop application, built on top of Kart our Geospatial version control system, is built entirely with xstate-tree using GraphQL for global state.
A minimal example of a single machine tree (See CodeSandbox for slightly expanded multi-machine example):
import React from "react";
import { createRoot } from "react-dom/client";
import { createMachine } from "xstate";
import { assign } from "@xstate/immer";
import {
createXStateTreeMachine
buildRootComponent
} from "@koordinates/xstate-tree";
type Events =
| { type: "SWITCH_CLICKED" }
| { type: "INCREMENT"; amount: number };
type Context = { incremented: number };
// A standard xstate machine, nothing extra is needed for xstate-tree
const machine = createMachine<Context, Events>(
{
id: "root",
initial: "inactive",
context: {
incremented: 0
},
states: {
inactive: {
on: {
SWITCH_CLICKED: "active"
}
},
active: {
on: {
SWITCH_CLICKED: "inactive",
INCREMENT: { actions: "increment" }
}
}
}
},
{
actions: {
increment: assign((context, event) => {
if (event.type !== "INCREMENT") {
return;
}
context.incremented += event.amount;
})
}
}
);
const RootMachine = createXStateTreeMachine(machine, {
// Selectors to transform the machines state into a representation useful for the view
selectors({ ctx, canHandleEvent, inState }) {
return {
canIncrement: canHandleEvent({ type: "INCREMENT", amount: 1 }),
showSecret: ctx.incremented > 10,
count: ctx.incremented,
active: inState("active")
}
},
// Actions to abstract away the details of sending events to the machine
actions({ send, selectors }) {
return {
increment(amount: number) {
send({
type: "INCREMENT",
amount: selectors.count > 4 ? amount * 2 : amount
});
},
switch() {
send({ type: "SWITCH_CLICKED" });
}
}
},
// If this tree had more than a single machine the slots to render child machines into would be defined here
// see the codesandbox example for an expanded demonstration that uses slots
slots: [],
// A view to bring it all together
// the return value is a plain React view that can be rendered anywhere by passing in the needed props
// the view has no knowledge of the machine it's bound to
view({ actions, selectors }) {
return (
<div>
<button onClick={() => actions.switch()}>
{selectors.active ? "Deactivate" : "Activate"}
</button>
<p>Count: {selectors.count}</p>
<button
onClick={() => actions.increment(1)}
disabled={!selectors.canIncrement}
>
Increment
</button>
{selectors.showSecret && <p>The secret password is hunter2</p>}
</div>
);
},
});
// Build the React host for the tree
const XstateTreeRoot = buildRootComponent(RootMachine);
// Rendering it with React
const ReactRoot = createRoot(document.getElementById("root"));
ReactRoot.render(<XstateTreeRoot />);
A more complicated todomvc example
Each machine that forms the tree representing your UI has an associated set of selector, action, view functions, and "slots"
- Selector functions are provided with the current context of the machine, a function to determine if it can handle a given event and a function to determine if it is in a given state, and expose the returned result to the view.
- Action functions are provided with the
send
method bound to the machines interpreter and the result of calling the selector function - Slots are how children of the machine are exposed to the view. They can be either single slot for a single actor, or multi slot for when you have a list of actors.
- View functions are React views provided with the output of the selector and action functions, and the currently active slots
To assist in making xstate-tree easy to use with TypeScript there is the createXStateTreeMachine
function for typing selectors, actions and view arguments and stapling the resulting functions to the xstate machine
createXStateTreeMachine
accepts the xstate machine as the first argument and takes an options argument with the following fields, it is important the fields are defined in this order or TypeScript will infer the wrong types:
selectors
, receives an object withctx
,inState
, andcanHandleEvent
fields.ctx
is the machines current context,inState
is the xstatestate.matches
function to allow determining if the machine is in a given state, andcanHandleEvent
accepts an event object and returns whether the machine will do anything in response to that event in it's current stateactions
, receives an object withsend
andselectors
fields.send
is the xstatesend
function bound to the machine, andselectors
is the result of calling the selector functionview
, is a React component that receivesactions
,selectors
, andslots
as props.actions
andselectors
being the result of the action/selector functions andslots
being an object with keys as the slot names and the values the slots React component
Full API docs coming soon, see #20
Slots are how invoked/spawned children of the machine are supplied to the Machines view. The child machines get wrapped into a React component responsible for rendering the machine itself. Since the view is provided with these components it is responsible for determining where in the view output they show up. This leaves the view in complete control of where the child views are placed.
Slotted machines are determined based on the id of the invoked/spawned machine. There are two types of slots, single and multi. Single slots are for invoked machines, where there will only be a single machine per slot. Multi slots are for spawned machines where there are multiple children per slot, rendered as a group; think lists. There is a set of helper functions for creating slots which in turn can be used to get the id for the slot.
singleSlot
accepts the name of the slot as the first argument and returns an object with a method getId()
that returns the id of the slot.
multiSlot
accepts the name of the slot and returns an object with a method getId(id: string)
that returns the id of the slot
You should always use the getId
methods when invoking/spawning something into a slot. Each slot the machine has must be represented by a call to singleSlot
or multiSlot
and stored into an array of slots. These slots must be passed to the buildView
and buildXStateTreeMachine
functions.
Communicating between multiple independent xstate machines is done via the broadcast
function.
Any event broadcast via this function is sent to every machine that has the event in its nextEvents
array, so it won't get sent to machines that have no handler for the event.
To get access to the type information for these events in a machine listening for it, use the PickEvent
type to extract the events you are interested in
ie PickEvent<"FOO" | "BAR">
will return {type: "FOO" } | { type: "BAR" }
which can be added to your machines event union.
To provide the type information on what events are available you add them to the global XstateTreeEvents interface. This is done using declare global
declare global {
interface XstateTreeEvents {
BASIC: string;
WITH_PAYLOAD: { a: "payload" }
}
}
That adds two events to the system, a no payload event ({ type: "BASIC" }
) and event with payload ({ type: "WITH_PAYLOAD"; a: "payload" }
). These events will now be visible in the typings for broadcast
and PickEvent
. The property name is the type
of the event and the type of the property is the payload of the event. If the event has no payload, use string
.
These events can be added anywhere, either next to a component for component specific events or in a module for events that are for multiple machines. One thing that it is important to keep in mind is that these declare global
declarations must be loaded by the .d.ts
files when importing the component, otherwise the events will be missing. Which means
- If they are in their own file, say for a module level declaration, that file will need to be imported somewhere. Somewhere that using a component will trigger the import
- If they are tied to a component they need to be in the index.ts file that imports the view/selectors/actions etc and calls
buildXstateTreeMachine
. If they are in the file containing those functions the index.d.ts file will not end up importing them.
It is relatively simple to display xstate-tree views directly in Storybook. Since the views are plain React components that accept selectors/actions/slots/inState as props you can just import the view and render it in a Story
There are a few utilities in xstate-tree to make this easier
This is a simple Proxy object that renders a
This is not relevant if using the render-view-component approach. But useful if you are planning on rendering the view using the xstate-tree machine itself, or testing the machine via the view.
It's a simple function that takes a name argument and returns a basic xstate-tree machine that you can replace slot services with. It just renders a div containing the name supplied