@builder.io/mobx-state-tree

Opinionated, transactional, MobX powered state container

Usage no npm install needed!

<script type="module">
  import builderIoMobxStateTree from 'https://cdn.skypack.dev/@builder.io/mobx-state-tree';
</script>

README

logo

mobx-state-tree

Opinionated, transactional, MobX powered state container combining the best features of the immutable and mutable world for an optimal DX

npm version CircleCI Coverage Status Join the community on Spectrum

Mobx and MST are amazing pieces of software, for me it is the missing brick when you build React based apps. Thanks for the great work!

Nicolas Galle full post Introduction blog post The curious case of MobX state tree

MobX state tree is a community driven project, but is looking for active maintainers! See #700

Contents

Installation

Typescript typings are included in the packages. Use module: "commonjs" or moduleResolution: "node" to make sure they are picked up automatically in any consuming project.

Supported browsers:

  • MobX-state-tree runs on any ES5 environment
  • However, for MobX version 4 or 5 can be used. MobX 4 will run on any environment, MobX 5 only on modern browsers. See for more details the MobX readme

Getting started

See the Getting started tutorial or follow the free egghead.io course (note however that the course is for MST v2, so it might be a bit outdated).

Talks & blogs

Philosophy & Overview

mobx-state-tree is a state container that combines the simplicity and ease of mutable data with the traceability of immutable data and the reactiveness and performance of observable data.

Simply put, mobx-state-tree tries to combine the best features of both immutability (transactionality, traceability and composition) and mutability (discoverability, co-location and encapsulation) based approaches to state management; everything to provide the best developer experience possible. Unlike MobX itself, mobx-state-tree is very opinionated about how data should be structured and updated. This makes it possible to solve many common problems out of the box.

Central in MST (mobx-state-tree) is the concept of a living tree. The tree consists of mutable, but strictly protected objects enriched with runtime type information. In other words, each tree has a shape (type information) and state (data). From this living tree, immutable, structurally shared, snapshots are automatically generated.

import { types, onSnapshot } from "mobx-state-tree"

const Todo = types
    .model("Todo", {
        title: types.string,
        done: false
    })
    .actions(self => ({
        toggle() {
            self.done = !self.done
        }
    }))

const Store = types.model("Store", {
    todos: types.array(Todo)
})

// create an instance from a snapshot
const store = Store.create({
    todos: [
        {
            title: "Get coffee"
        }
    ]
})

// listen to new snapshots
onSnapshot(store, snapshot => {
    console.dir(snapshot)
})

// invoke action that modifies the tree
store.todos[0].toggle()
// prints: `{ todos: [{ title: "Get coffee", done: true }]}`

By using the type information available, snapshots can be converted to living trees, and vice versa, with zero effort. Because of this, time travelling is supported out of the box, and tools like HMR are trivial to support example.

The type information is designed in such a way that it is used both at design- and run-time to verify type correctness (Design time type checking works in TypeScript only at the moment; Flow PR's are welcome!)

[mobx-state-tree] Value '{\"todos\":[{\"turtle\":\"Get tea\"}]}' is not assignable to type: Store, expected an instance of Store or a snapshot like '{ todos: { title: string; done: boolean }[] }' instead.

Runtime type error

typescript error

Designtime type error

Because state trees are living, mutable models, actions are straight-forward to write; just modify local instance properties where appropriate. See toggleTodo() above or the examples below. It is not necessary to produce a new state tree yourself, MST's snapshot functionality will derive one for you automatically.

Although mutable sounds scary to some, fear not, actions have many interesting properties. By default trees can only be modified by using an action that belongs to the same subtree. Furthermore, actions are replayable and can be used to distribute changes (example).

Moreover, because changes can be detected on a fine grained level, JSON patches are supported out of the box. Simply subscribing to the patch stream of a tree is another way to sync diffs with, for example, back-end servers or other clients (example).

patches

Since MST uses MobX behind the scenes, it integrates seamlessly with mobx and mobx-react. See also this egghead.io lesson: Render mobx-state-tree Models in React. Even cooler, because it supports snapshots, middleware and replayable actions out of the box, it is possible to replace a Redux store and reducer with a MobX state tree. This makes it possible to connect the Redux devtools to MST. See the Redux / MST TodoMVC example.

For futher reading: the conceptual difference between snapshots, patches and actions in relation to distributing state changes is extensively discussed in this blog post

devtools

Finally, MST has built-in support for references, identifiers, dependency injection, change recording and circular type definitions (even across files). Even fancier, it analyses liveliness of objects, failing early when you try to access accidentally cached information! (More on that later)

A unique feature of MST is that it offers liveliness guarantees. MST will throw when reading or writing from objects that are no longer part of a state tree. This protects you against accidental stale reads of objects still referred by, for example, a closure.

const oldTodo = store.todos[0]
store.removeTodo(0)

function logTodo(todo) {
    setTimeout(() => console.log(todo.title), 1000)
}

logTodo(store.todos[0])
store.removeTodo(0)
// throws exception in one second for using an stale object!

Despite all that, you will see that the API is quite straightforward!

Another way to look at mobx-state-tree is to consider it, as argued by Daniel Earwicker, to be "React, but for data". Like React, MST consists of composable components, called models, which captures a small piece of state. They are instantiated from props (snapshots) and after that manage and protect their own internal state (using actions). Moreover, when applying snapshots, tree nodes are reconciled as much as possible. There is even a context-like mechanism, called environments, to pass information to deep descendants.

An introduction to the philosophy can be watched here. Slides. Or, as markdown to read it quickly.

mobx-state-tree "immutable trees" and "graph model" features talk, "Next Generation State Management" at React Europe 2017. Slides.

Examples

To run the examples:

  1. clone this repository
  2. navigate to the example folder (e.g. packages/mst-example-bookshop)
  3. run yarn install and yarn start
  • Bookshop Example webshop application with references, identifiers, routing, testing, etc.
  • Boxes Example app where one can draw, drag, and drop boxes. With time-travelling and multi-client synchronization over websockets.
  • TodoMVC Classic example app using React and MST.
  • Redux TodoMVC Redux TodoMVC application, except that the reducers are replaced with a MST. Tip: open the Redux devtools; they will work!

Concepts

With MobX state tree, you build, as the name suggests, trees of models.

Trees, types and state

Each node in the tree is described by two things: Its type (the shape of the thing) and its data (the state it is currently in).

The simplest tree possible:

import { types } from "mobx-state-tree"

// declaring the shape of a node with the type `Todo`
const Todo = types.model({
    title: types.string
})

// creating a tree based on the "Todo" type, with initial data:
const coffeeTodo = Todo.create({
    title: "Get coffee"
})

The types.model type declaration is used to describe the shape of an object. Other built-in types include arrays, maps, primitives, etc. See the types overview. The type information will be used for both.

Creating models

egghead.io lesson 1: Describe Your Application Domain Using mobx-state-tree(MST) Models

The most important type in MST is types.model, which can be used to describe the shape of an object. An example:

const TodoStore = types
    // 1
    .model("TodoStore", {
        loaded: types.boolean, // 2
        endpoint: "http://localhost", // 3
        todos: types.array(Todo), // 4
        selectedTodo: types.reference(Todo) // 5
    })
    .views(self => {
        return {
            // 6
            get completedTodos() {
                return self.todos.filter(t => t.done)
            },
            // 7
            findTodosByUser(user) {
                return self.todos.filter(t => t.assignee === user)
            }
        }
    })
    .actions(self => {
        return {
            addTodo(title) {
                self.todos.push({
                    id: Math.random(),
                    title
                })
            }
        }
    })

When defining a model, it is advised to give the model a name for debugging purposes (see // 1). A model takes additionally object argument defining the properties.

The properties argument is a key-value set where each key indicates the introduction of a property, and the value its type. The following types are acceptable:

  1. A type. This can be a simple primitive type like types.boolean, see // 2, or a complex, possibly pre-defined type (// 4)
  2. A primitive. Using a primitive as type is syntactic sugar for introducing a property with a default value. See // 3, endpoint: "http://localhost" is the same as endpoint: types.optional(types.string, "http://localhost"). The primitive type is inferred from the default value. Properties with a default value can be omitted in snapshots.
  3. A computed property, see // 6. Computed properties are tracked and memoized by MobX. Computed properties will not be stored in snapshots or emit patch events. It is possible to provide a setter for a computed property as well. A setter should always invoke an action.
  4. A view function (see // 7). A view function can, unlike computed properties, take arbitrary arguments. It won't be memoized, but its value can be tracked by MobX nonetheless. View functions are not allowed to change the model, but should rather be used to retrieve information from the model.

Tip: (self) => ({ action1() { }, action2() { }}) is ES6 syntax for function (self) { return { action1: function() { }, action2: function() { } }}. In other words, it's short way of directly returning an object literal. For that reason a comma between each member of a model is mandatory, unlike classes which are syntactically a totally different concept.

types.model creates a chainable model type, where each chained method produces a new type:

  • .named(name) clones the current type, but gives it a new name
  • .props(props) produces a new type, based on the current one, and adds / overrides the specified properties
  • .actions(self => object literal with actions) produces a new type, based on the current one, and adds / overrides the specified actions
  • .views(self => object literal with view functions) produces a new type, based on the current one, and adds / overrides the specified view functions
  • .preProcessSnapshot(snapshot => snapshot) can be used to pre-process the raw JSON before instantiating a new model. See Lifecycle hooks or alternatively types.snapshotProcessor
  • .postProcessSnapshot(snapshot => snapshot) can be used to post-process the raw JSON before getting a model snapshot. See Lifecycle hooks or alternatively types.snapshotProcessor

Note that views and actions don't define actions and views directly, but rather they should be given a function. The function will be invoked when a new model instance is created. The instance will be passed in as the first and only argument typically called self. This has two advantages:

  1. All methods will always be bound correctly, and won't suffer from an unbound this
  2. The closure can be used to store private state or methods of the instance. See also actions and volatile state.

Quick example:

const TodoStore = types
    .model("TodoStore", {
        /* props */
    })
    .actions(self => {
        const instantiationTime = Date.now()

        function addTodo(title) {
            console.log(`Adding Todo ${title} after ${(Date.now() - instantiationTime) / 1000}s.`)
            self.todos.push({
                id: Math.random(),
                title
            })
        }

        return { addTodo }
    })

It is perfectly fine to chain multiple views, props calls etc in arbitrary order. This can be a great way to structure complex types, mix-in utility functions, etc. Each call in the chain creates a new, immutable type which can itself be stored and reused as part of other types, etc.

It is also possible to define lifecycle hooks in the actions object. These are actions with a predefined name that are run at a specific moment. See Lifecycle hooks.

Tree semantics in detail

MST trees have very specific semantics. These semantics purposefully constrain what you can do with MST. The reward for that is all kinds of generic features out of the box like snapshots, replayability, etc. If these constraints don't suit your app, you are probably better off using plain MobX with your own model classes, which is fine as well.

  1. Each object in an MST tree is considered a node. Each primitive (and frozen) value is considered a leaf.
  2. MST has only three types of nodes: model, array and map.
  3. Every node tree in an MST tree is a tree in itself. Any operation that can be invoked on the complete tree can also be applied to a subtree.
  4. A node can only exist exactly once in a tree. This ensures it has a unique, identifiable position.
  5. It is however possible to refer to another object in the same tree by using references
  6. There is no limit to the number of MST trees that live in an application. However, each node can only live in exactly one tree.
  7. All leaves in the tree must be serializable. It is not possible to store, for example, functions in a MST.
  8. The only free-form type in MST is frozen, with the requirement that frozen values are immutable and serializable so that the MST semantics can still be upheld.
  9. At any point in the tree it is possible to assign a snapshot to the tree instead of a concrete instance of the expected type. In that case an instance of the correct type, based on the snapshot, will be automatically created for you.
  10. Nodes in the MST tree will be reconciled (the exact same instance will be reused) when updating the tree by any means, based on their identifier property. If there is no identifier property, instances won't be reconciled.
  11. If a node in the tree is replaced by another node, the original node will die and become unusable. This makes sure you are not accidentally holding on to stale objects anywhere in your application.
  12. If you want to create a new node based on an existing node in a tree, you can either detach that node, or clone it.

These egghead.io lessons nicely leverage the specific semantics of MST trees:

egghead.io lesson 6: Build Forms with React to Edit mobx-state-tree Models
egghead.io lesson 7: Remove Model Instances from the Tree
egghead.io lesson 8: Create an Entry Form to Add Models to the State Tree

Composing trees

In MST every node in the tree is a tree in itself. Trees can be composed by composing their types:

const TodoStore = types.model({
    todos: types.array(Todo)
})

const storeInstance = TodoStore.create({
    todos: [
        {
            title: "Get biscuit"
        }
    ]
})

The snapshot passed to the create method of a type will recursively be turned in MST nodes. So, you can safely call:

storeInstance.todos[0].setTitle("Chocolate instead plz")

Because any node in a tree is a tree in itself, any built-in method in MST can be invoked on any node in the tree, not just the root. This makes it possible to get a patch stream of a certain subtree, or to apply middleware to a certain subtree only.

Actions

egghead.io lesson 2: Attach Behavior to mobx-state-tree Models Using Actions

By default, nodes can only be modified by one of their actions, or by actions higher up in the tree. Actions can be defined by returning an object from the action initializer function that was passed to actions. The initializer function is executed for each instance, so that self is always bound to the current instance. Also, the closure of that function can be used to store so called volatile state for the instance or to create private functions that can only be invoked from the actions, but not from the outside.

const Todo = types
    .model({
        title: types.string
    })
    .actions(self => {
        function setTitle(newTitle) {
            self.title = newTitle
        }

        return {
            setTitle
        }
    })

Shorter form if no local state or private functions are involved:

const Todo = types
    .model({
        title: types.string
    })
    .actions(self => ({
        // note the `({`, we are returning an object literal
        setTitle(newTitle) {
            self.title = newTitle
        }
    }))

Actions are replayable and are therefore constrained in several ways:

  • Trying to modify a node without using an action will throw an exception.
  • It's recommended to make sure action arguments are serializable. Some arguments can be serialized automatically such as relative paths to other nodes
  • Actions can only modify models that belong to the (sub)tree on which they are invoked
  • You cannot use this inside actions. Instead, use self. This makes it safe to pass actions around without binding them or wrapping them in arrow functions.

Useful methods:

  • onAction listens to any action that is invoked on the model or any of its descendants.
  • addMiddleware adds an interceptor function to any action invoked on the subtree.
  • applyAction invokes an action on the model according to the given action description

Asynchronous actions

egghead.io lesson 12: Defining Asynchronous Processes Using Flow

Asynchronous actions have first class support in MST and are described in more detail here. Asynchronous actions are written by using generators and always return a promise. For a real working example see the bookshop sources. A quick example to get the gist:

Warning: don't import flow from "mobx", but from "mobx-state-tree" instead!

import { types, flow } from "mobx-state-tree"

someModel.actions(self => {
    const fetchProjects = flow(function*() {
        // <- note the star, this a generator function!
        self.state = "pending"
        try {
            // ... yield can be used in async/await style
            self.githubProjects = yield fetchGithubProjectsSomehow()
            self.state = "done"
        } catch (error) {
            // ... including try/catch error handling
            console.error("Failed to fetch projects", error)
            self.state = "error"
        }
        // The action will return a promise that resolves to the returned value
        // (or rejects with anything thrown from the action)
        return self.githubProjects.length
    })

    return { fetchProjects }
})

Note that, since MST v3.9, TypeScript correctly infers flow arguments and usually infers correctly flow return types, but one exception to this case is when a Promise is returned as final value. In this case (and only in this case) this construct needs to be used:

return castFlowReturn(somePromise)

Action listeners versus middleware

The difference between action listeners and middleware is: middleware can intercept the action that is about to be invoked, modify arguments, return types, etc. Action listeners cannot intercept and are only notified. Action listeners receive the action arguments in a serializable format, while middleware receives the raw arguments. (onAction is actually just a built-in middleware).

For more details on creating middleware, see the docs.

Disabling protected mode

This may be desired if the default protection of mobx-state-tree doesn't fit your use case. For example, if you are not interested in replayable actions or hate the effort of writing actions to modify any field, unprotect(tree) will disable the protected mode of a tree allowing anyone to directly modify the tree.

Views

egghead.io lesson 4: Derive Information from Models Using Views

Any fact that can be derived from your state is called a "view" or "derivation". See the Mobx concepts & principles for some background.

Views come in two flavors: views with arguments and views without arguments. The latter are called computed values, based on the computed concept in MobX. The main difference between the two is that computed properties create an explicit caching point, but later they work the same and any other computed value or MobX based reaction like @observer components can react to them. Computed values are defined using getter functions.

Example:

import { autorun } from "mobx"

const UserStore = types
    .model({
        users: types.array(User)
    })
    .views(self => ({
        get numberOfChildren() {
            return self.users.filter(user => user.age < 18).length
        },
        numberOfPeopleOlderThan(age) {
            return self.users.filter(user => user.age > age).length
        }
    }))

const userStore = UserStore.create(/* */)

// Every time the userStore is updated in a relevant way, log messages will be printed
autorun(() => {
    console.log("There are now ", userStore.numberOfChildren, " children")
})
autorun(() => {
    console.log("There are now ", userStore.numberOfPeopleOlderThan(75), " pretty old people")
})

If you want to share volatile state between views and actions, use .extend instead of .views + .actions. See the volatile state section.

Snapshots

egghead.io lesson 3: Test mobx-state-tree Models by Recording Snapshots or Patches
egghead.io lesson 9: Store Store in Local Storage
egghead.io lesson 16: Automatically Send Changes to the Server by Using onSnapshot

Snapshots are the immutable serialization, in plain objects, of a tree at a specific point in time. Snapshots can be inspected through getSnapshot(node, applyPostProcess). Snapshots don't contain any type information and are stripped from all actions, etc., so they are perfectly suitable for transportation. Requesting a snapshot is cheap as MST always maintains a snapshot of each node in the background and uses structural sharing.

coffeeTodo.setTitle("Tea instead plz")

console.dir(getSnapshot(coffeeTodo))
// prints `{ title: "Tea instead plz" }`

Some interesting properties of snapshots:

  • Snapshots are immutable
  • Snapshots can be transported
  • Snapshots can be used to update models or restore them to a particular state
  • Snapshots are automatically converted to models when needed. So, the two following statements are equivalent: store.todos.push(Todo.create({ title: "test" })) and store.todos.push({ title: "test" }).

Useful methods:

  • getSnapshot(model, applyPostProcess): returns a snapshot representing the current state of the model
  • onSnapshot(model, callback): creates a listener that fires whenever a new snapshot is available (but only one per MobX transaction).
  • applySnapshot(model, snapshot): updates the state of the model and all its descendants to the state represented by the snapshot

Patches

egghead.io lesson 3: Test mobx-state-tree Models by Recording Snapshots or Patches

Modifying a model does not only result in a new snapshot, but also in a stream of JSON-patches describing which modifications were made. Patches have the following signature:

export interface IJsonPatch {
    op: "replace" | "add" | "remove"
    path: string
    value?: any
}
  • Patches are constructed according to JSON-Patch, RFC 6902
  • Patches are emitted immediately when a mutation is made and don't respect transaction boundaries (like snapshots)
  • Patch listeners can be used to achieve deep observing
  • The path attribute of a patch contains the path of the event relative to the place where the event listener is attached
  • A single mutation can result in multiple patches, for example when splicing an array
  • Patches can be reverse applied, which enables many powerful patterns like undo / redo

Useful methods:

  • onPatch(model, listener) attaches a patch listener to the provided model, which will be invoked whenever the model or any of its descendants is mutated
  • applyPatch(model, patch) applies a patch (or array of patches) to the provided model

References and identifiers

egghead.io lesson 13: Create Relationships in your Data with mobx-state-tree Using References and Identifiers

References and identifiers are a first-class concept in MST. This makes it possible to declare references and keep the data normalized in the background, while you interact with it in a denormalized manner.

Example:

const Todo = types.model({
    id: types.identifier,
    title: types.string
})

const TodoStore = types.model({
    todos: types.array(Todo),
    selectedTodo: types.reference(Todo)
})

// create a store with a normalized snapshot
const storeInstance = TodoStore.create({
    todos: [
        {
            id: "47",
            title: "Get coffee"
        }
    ],
    selectedTodo: "47"
})

// because `selectedTodo` is declared to be a reference, it returns the actual Todo node with the matching identifier
console.log(storeInstance.selectedTodo.title)
// prints "Get coffee"

Identifiers

  • Each model can define zero or one identifier() properties
  • The identifier property of an object cannot be modified after initialization
  • Each identifier / type combination should be unique within the entire tree
  • Identifiers are used to reconcile items inside arrays and maps - wherever possible - when applying snapshots
  • The map.put() method can be used to simplify adding objects that have identifiers to maps
  • The primary goal of identifiers is not validation, but reconciliation and reference resolving. For this reason identifiers cannot be defined or updated after creation. If you want to check if some value just looks as an identifier, without providing the above semantics; use something like: types.refinement(types.string, v => v.match(/someregex/))

_Tip: If you know the format of the identifiers in your application, leverage types.refinement to actively check this, for example the following definition enforces that identifiers of Car always start with the string "Car_":

const Car = types.model("Car", {
    id: types.refinement(types.identifier, identifier => identifier.indexOf("Car_") === 0)
})

References

References are defined by mentioning the type they should resolve to. The targeted type should have exactly one attribute of the type identifier. References are looked up through the entire tree but per type, so identifiers need to be unique in the entire tree.

Customizable references

The default implementation uses the identifier cache to resolve references (See resolveIdentifier). However, it is also possible to override the resolve logic and provide your own custom resolve logic. This also makes it possible to, for example, trigger a data fetch when trying to resolve the reference (example).

Example:

const User = types.model({
    id: types.identifier,
    name: types.string
})

const UserByNameReference = types.maybeNull(
    types.reference(User, {
        // given an identifier, find the user
        get(identifier /* string */, parent: any /*Store*/) {
            return parent.users.find(u => u.name === identifier) || null
        },
        // given a user, produce the identifier that should be stored
        set(value /* User */) {
            return value.name
        }
    })
)

const Store = types.model({
    users: types.array(User),
    selection: UserByNameReference
})

const s = Store.create({
    users: [{ id: "1", name: "Michel" }, { id: "2", name: "Mattia" }],
    selection: "Mattia"
})

Reference validation: isValidReference, tryReference, onInvalidated hook and types.safeReference

Accessing an invalid reference (a reference to a dead/detached node) triggers an exception.

In order to check if a reference is valid, MST offers the isValidReference(() => ref): boolean function:

const isValid = isValidReference(() => store.myRef)

Also, if you are unsure if a reference is valid or not you can use the tryReference(() => ref): ref | undefined function:

// the result will be the passed ref if ok, or undefined if invalid
const maybeValidRef = tryReference(() => store.myRef)

The options parameter for references also accepts an optional onInvalidated hook, which will be called when the reference target node that the reference is pointing to is about to be detached/destroyed. It has the following signature:

const refWithOnInvalidated = types.reference(Todo, {
    onInvalidated(event: {
        // what is causing the target to become invalidated
        cause: "detach" | "destroy" | "invalidSnapshotReference"
        // the target that is about to become invalidated (undefined if "invalidSnapshotReference")
        invalidTarget: STN | undefined
        // the identifier that is about to become invalidated
        invalidId: string | number
        // parent node of the reference (not the reference target)
        parent: IAnyStateTreeNode
        // a function to remove the reference from its parent (or set to undefined in the case of models)
        removeRef: () => void
        // a function to set our reference to a new target
        replaceRef: (newRef: STN | null | undefined) => void
    }) {
        // do something
    }
})

Note that invalidation will only trigger while the reference is attached to a parent (be it a model, an array, a map, etc.).

A default implementation of such onInvalidated hook is provided by the types.safeReference type. It is like a standard reference, except that once the target node becomes invalidated it will:

  • If its parent is a model: Set its own property to undefined
  • If its parent is an array: Remove itself from the array
  • If its parent is a map: Remove itself from the map

In addition to the options possible for a plain reference type, the optional options parameter object also accepts a parameter named acceptsUndefined, which is set to true by default, so it is suitable for model properties. When used inside collections (arrays/maps) it is recommended to set this option to false so it can't take undefined as value, which is usually the desired in those cases.

Strictly speaking, safeReference with acceptsUndefined set to true (the default) is implemented as

types.maybe(
    types.reference(Type, {
        ...customGetSetIfAvailable,
        onInvalidated(ev) {
            ev.removeRef()
        }
    })
)

and with acceptsUndefined set to false as

types.reference(Type, {
    ...customGetSetIfAvailable,
    onInvalidated(ev) {
        ev.removeRef()
    }
})

const Todo = types.model({ id: types.identifier })
const Store = types.model({
    todos: types.array(Todo),
    selectedTodo: types.safeReference(Todo),
    multipleSelectedTodos: types.array(types.safeReference(Todo, { acceptsUndefined: false }))
})

// given selectedTodo points to a valid Todo and that Todo is later removed from the todos
// array, then selectedTodo will automatically become undefined, and if it is included in multipleSelectedTodos
// then it will be removed from the array

Listening to observables, snapshots, patches or actions

MST is powered by MobX. This means that it is immediately compatible with observer components or reactions like autorun:

import { autorun } from "mobx"

autorun(() => {
    console.log(storeInstance.selectedTodo.title)
})

Because MST keeps immutable snapshots in the background, it is also possible to be notified when a new snapshot of the tree is available. This is similar to .subscribe on a redux store:

onSnapshot(storeInstance, newSnapshot => {
    console.dir("Got new state: ", newSnapshot)
})

However, sometimes it is more useful to precisely know what has changed rather than just receiving a complete new snapshot. For that, MST supports json-patches out of the box.

onPatch(storeInstance, patch => {
    console.dir("Got change: ", patch)
})

storeInstance.todos[0].setTitle("Add milk")
// prints:
{
    path: "/todos/0",
    op: "replace",
    value: "Add milk"
}

Similarly, you can be notified whenever an action is invoked by using onAction.

onAction(storeInstance, call => {
    console.dir("Action was called: ", call)
})

storeInstance.todos[0].setTitle("Add milk")
// prints:
{
    path: "/todos/0",
    name: "setTitle",
    args: ["Add milk"]
}

It is even possible to intercept actions before they are applied by adding middleware using addMiddleware:

addMiddleware(storeInstance, (call, next) => {
    call.args[0] = call.args[0].replace(/tea/gi, "Coffee")
    return next(call)
})

A more extensive middleware example can be found in this code sandbox. For more details on creating middleware and the exact specification of middleware events, see the docs.

Finally, it is not only possible to be notified about snapshots, patches or actions. It is also possible to re-apply them by using applySnapshot, applyPatch or applyAction!

Volatile state

egghead.io lesson 15: Use Volatile State and Lifecycle Methods to Manage Private State

MST models primarily aid in storing persistable state. State that can be persisted, serialized, transferred, patched, replaced, etc. However, sometimes you need to keep track of temporary, non-persistable state. This is called volatile state in MST. Examples include promises, sockets, DOM elements, etc. - state which is needed for local purposes as long as the object is alive.

Volatile state (which is also private) can be introduced by creating variables inside any of the action initializer functions.

Volatile is preserved for the life-time of an object and not reset when snapshots are applied, etc. Note that the life time of an object depends on proper reconciliation, see the how does reconciliation work? section below.

The following is an example of an object with volatile state. Note that volatile state here is used to track a XHR request and clean up resources when it is disposed. Without volatile state this kind of information would need to be stored in an external WeakMap or something similar.

const Store = types
    .model({
        todos: types.array(Todo),
        state: types.enumeration("State", ["loading", "loaded", "error"])
    })
    .actions(self => {
        let pendingRequest = null // a Promise

        function afterCreate() {
            self.state = "loading"
            pendingRequest = someXhrLib.createRequest("someEndpoint")
        }

        function beforeDestroy() {
            // abort the request, no longer interested
            pendingRequest.abort()
        }

        return {
            afterCreate,
            beforeDestroy
        }
    })

Some tips:

  1. Note that multiple actions calls can be chained. This makes it possible to create multiple closures with their own protected volatile state.

  2. Although in the above example the pendingRequest could be initialized directly in the action initializer, it is recommended to do this in the afterCreate hook, which will only once the entire instance has been set up (there might be many action and property initializers for a single type).

  3. The above example doesn't actually use the promise. For how to work with promises / asynchronous flows, see the asynchronous actions section above.

  4. It is possible to share volatile state between views and actions by using extend. .extend works like a combination of .actions and .views and should return an object with a actions and views field:

Here's an example of how to do your own volatile state using an observable:

// if your local state is part of a view getter (computed) then
// it is important to make sure that state used such getters are observable,
// or else the value returned by the view would become stale upon observation
const Todo = types.model({}).extend(self => {
    const localState = observable.box(3)

    return {
        views: {
            // note this one IS a getter (computed value)
            get x() {
                return localState.get()
            }
        },
        actions: {
            setX(value) {
                localState.set(value)
            }
        }
    }
})

And here's an example of how to do your own volatile state not using an observable (but if you do this make sure the local state will never be used in a computed value first and bear in mind it won't be reactive!):

// if not using an observable then make sure your local state is NOT part of a view getter or computed value of any kind!
// also changes to it WON'T be reactive
const Todo = types.model({}).extend(self => {
    let localState = 3

    return {
        views: {
            // note this one is NOT a getter (NOT a computed value)
            // if this were a getter this value would get stale upon observation
            getX() {
                return localState
            }
        },
        actions: {
            setX(value) {
                localState = value
            }
        }
    }
})

model.volatile

Since the pattern above (having a volatile state that is observable (in terms of Mobx observables) and readable from outside the instance) is such a common pattern there is a shorthand to declare such properties. The example above can be rewritten as:

const Todo = types
    .model({})
    .volatile(self => ({
        localState: 3
    }))
    .actions(self => ({
        setX(value) {
            self.localState = value
        }
    }))

The object that is returned from the volatile initializer function can contain any piece of data and will result in an instance property with the same name. Volatile properties have the following characteristics:

  1. They can be read from outside the model (if you want hidden volatile state, keep the state in your closure as shown in the previous section, and only if it is not used on a view consider not making it observable)
  2. The volatile properties will be only observable, see observable references. Values assigned to them will be unmodified and not automatically converted to deep observable structures.
  3. Like normal properties, they can only be modified through actions
  4. Volatile props will not show up in snapshots, and cannot be updated by applying snapshots
  5. Volatile props are preserved during the lifecycle of an instance. See also reconciliation
  6. Changes in volatile props won't show up in the patch or snapshot stream
  7. It is currently not supported to define getters / setters in the object returned by volatile

Dependency injection

When creating a new state tree it is possible to pass in environment specific data by passing an object as the second argument to a .create call. This object should be (shallowly) immutable and can be accessed by any model in the tree by calling getEnv(self).

This is useful to inject environment or test-specific utilities like a transport layer, loggers, etc. This is also very useful to mock behavior in unit tests or provide instantiated utilities to models without requiring singleton modules. See also the bookshop example for inspiration.

import { types, getEnv } from "mobx-state-tree"

const Todo = types
    .model({
        title: ""
    })
    .actions(self => ({
        setTitle(newTitle) {
            // grab injected logger and log
            getEnv(self).logger.log("Changed title to: " + newTitle)
            self.title = newTitle
        }
    }))

const Store = types.model({
    todos: types.array(Todo)
})

// setup logger and inject it when the store is created
const logger = {
    log(msg) {
        console.log(msg)
    }
}

const store = Store.create(
    {
        todos: [{ title: "Grab tea" }]
    },
    {
        logger: logger // inject logger to the tree
    }
)

store.todos[0].setTitle("Grab coffee")
// prints: Changed title to: Grab coffee

Types overview

egghead.io lesson 11: More mobx-state-tree Types: map, literal, union, and enumeration
egghead.io lesson 17: Create Dynamic Types and use Type Composition to Extract Common Functionality

These are the types available in MST. All types can be found in the types namespace, e.g. types.string. See Api Docs for examples.

Complex types

  • types.model(properties, actions) Defines a "class like" type with properties and actions to operate on the object.
  • types.array(type) Declares an array of the specified type.
  • types.map(type) Declares a map of the specified type.

Note that since MST v3 types.array and types.map are wrapped in types.optional by default, with [] and {} set as their default values, respectively.

Primitive types

  • types.string
  • types.number
  • types.integer
  • types.boolean
  • types.Date
  • types.custom creates a custom primitive type. This is useful to define your own types that map a serialized form one-to-one to an immutable object like a Decimal or Date.

Utility types

  • types.union(options?: { dispatcher?: (snapshot) => Type, eager?: boolean }, types...) create a union of multiple types. If the correct type cannot be inferred unambiguously from a snapshot, provide a dispatcher function to determine the type. When eager flag is set to true (default) - the first matching type will be used, if set to false the type check will pass only if exactly 1 type matches.
  • types.optional(type, defaultValue, optionalValues?) marks a value as being optional (in e.g. a model). If a value is not provided/undefined (or set to any of the primitive values passed as an optional optionalValues array) the defaultValue will be used instead. If defaultValue is a function, it will be evaluated. This can be used to generate, for example, IDs or timestamps upon creation.
  • types.literal(value) can be used to create a literal type, where the only possible value is specifically that value. This is very powerful in combination with unions. E.g. temperature: types.union(types.literal("hot"), types.literal("cold")).
  • types.enumeration(name?, options: string[]) creates an enumeration. This method is a shorthand for a union of string literals. If you are using typescript and want to create a type based on an string enum (e.g. enum Color { ... }) then use types.enumeration<Color>("Color", Object.values(Color)), where the "Color" name argument is optional.
  • types.refinement(name?, baseType, (snapshot) => boolean) creates a type that is more specific than the base type, e.g. types.refinement(types.string, value => value.length > 5) to create a type of strings that can only be longer then 5.
  • types.maybe(type) makes a type optional and nullable, shorthand for types.optional(types.union(type, types.literal(undefined)), undefined).
  • types.maybeNull(type) like maybe, but uses null to represent the absence of a value.
  • types.null the type of null.
  • types.undefined the type of undefined.
  • types.late(() => type) can be used to create recursive or circular types, or types that are spread over files in such a way that circular dependencies between files would be an issue otherwise.
  • types.frozen(subType? | defaultValue?) Accepts any kind of serializable value (both primitive and complex), but assumes that the value itself is immutable and serializable. frozen can be invoked in a few different ways:
    • types.frozen() - behaves the same as types.frozen in MST 2.
    • types.frozen(subType) - provide a valid MST type and frozen will check if the provided data conforms the snapshot for that type. Note that the type will not actually be instantiated, so it can only be used to check the shape of the data. Adding views or actions to SubType would be pointless.
    • types.frozen(someDefaultValue) - provide a primitive value, object or array, and MST will infer the type from that object, and also make it the default value for the field
    • (Typescript) types.frozen<TypeScriptType>(...) - provide a typescript type, to help in strongly typing the field (design time only)
  • types.compose(name?, type1...typeX), creates a new model type by taking a bunch of existing types and combining them into a new one.
  • types.reference(targetType) creates a property that is a reference to another item of the given targetType somewhere in the same tree. See references for more details.
  • types.safeReference(targetType) is like a standard reference, except that it accepts the undefined value by default and automatically sets itself to undefined (when the parent is a model) / removes itself from arrays and maps when the reference it is pointing to gets detached/destroyed. See references for more details.
  • types.snapshotProcessor(type, processors, name?) runs a pre snapshot / post snapshot processor before/after serializing a given type. Example:
    const Todo1 = types.model({ text: types.string })
// in the backend the text type must be null when empty
interface BackendTodo {
    text: string | null
}
const Todo2 = types.snapshotProcessor(Todo1, {
    // from snapshot to instance
    preProcessor(sn: BackendTodo) {
        return {
            text: sn.text || "";
        }
    },
    // from instance to snapshot
    postProcessor(sn): BackendTodo {
        return {
            text: !sn.text ? null : sn.text
        }
    }
})

Property types

Property types can only be used as a direct member of a types.model type and not further composed (for now).

  • types.identifier Only one such member can exist in a types.model and should uniquely identify the object. See identifiers for more details. subType should be either types.string or types.number, defaulting to the first if not specified.
  • types.identifierNumber Similar to types.identifier. However, during serialization, the identifier value will be parsed from / serialized to a number.

LifeCycle hooks for types.model

egghead.io lesson 14: Loading Data from the Server after model creation

All of the below hooks can be created by returning an action with the given name, like:

const Todo = types.model("Todo", { done: true }).actions(self => ({
    afterCreate() {
        console.log("Created a new todo!")
    }
}))

The exception to this rule are the preProcessSnapshot and postProcessSnapshot hooks (see types.snapshotProcessor as an alternative):

types
    .model("Todo", { done: true })
    .preProcessSnapshot(snapshot => ({
        // auto convert strings to booleans as part of preprocessing
        done: snapshot.done === "true" ? true : snapshot.done === "false" ? false : snapshot.done
    }))
    .actions(self => ({
        afterCreate() {
            console.log("Created a new todo!")
        }
    }))

Note: pre and post processing are just meant to convert your data into types that are more acceptable to MST. Typically it should be the case that `postProcess(preProcess(snapshot)) === snapshot. If that isn't the case, consider whether you shouldn't be using a dedicated a view instead to normalize your snapshot to some other format you need.

Hook Meaning
afterCreate Immediately after an instance is created and initial values are applied. Children will fire this event before parents. You can't make assumptions about the parent safely, use afterAttach if you need to.
afterAttach As soon as the direct parent is assigned (this node is attached to another node). If an element is created as part of a parent, afterAttach is also fired. Unlike afterCreate, afterAttach will fire breadth first. So, in afterAttach one can safely make assumptions about the parent, but in afterCreate not
beforeDetach As soon as the node is removed from the direct parent, but only if the node is not destroyed. In other words, when detach(node) is used
beforeDestroy Called before the node is destroyed, as a result of calling destroy, or by removing or replacing the node from the tree. Child destructors will fire before parents
preProcessSnapshot Before creating an instance or applying a snapshot to an existing instance, this hook is called to give the option to transform the snapshot before it is applied. The hook should be a pure function that returns a new snapshot. This can be useful to do some data conversion, enrichment, property renames, etc. This hook is not called for individual property updates. **Note 1: Unlike the other hooks, this one is _not created as part of the actions initializer, but directly on the type!**_ _**Note 2: The preProcessSnapshot transformation must be pure; it should not modify its original input argument!**_
postProcessSnapshot This hook is called every time a new snapshot is being generated. Typically it is the inverse function of preProcessSnapshot. This function should be a pure function that returns a new snapshot. **Note: Unlike the other hooks, this one is _not created as part of the actions initializer, but directly on the type!**_

Note, except for preProcessSnapshot and postProcessSnapshot, all hooks should be defined as actions.

All hooks can be defined multiple times and can be composed automatically.

LifeCycle hooks for types.array/types.map

Hooks for types.array/types.map can be defined by using the .hooks(self => ({})) method.

Calling .hooks(...) produces new type, same as calling .actions() for types.model.

Available hooks are:

Hook Meaning