Model Driven State Management
TypeScript JavaScript Other
Fetching latest commit…
Cannot retrieve the latest commit at this time.

README.md

npm version Build Status Coverage Status Join the community on Spectrum


mobx-state-tree

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


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.

Getting started

See the Getting started tutorial.

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 on 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 generated automatically.

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 But even cooler: because it supports snapshots, middleware and replayable actions out of the box, it is even possible to replace a Redux store and reducer with a MobX state tree. This makes it even possible to connect the Redux devtools to MST. See the Redux / MST TodoMVC example.

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 pretty unique feature of MST is that it offers liveliness guarantees; it 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 pretty 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

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

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 of using plain mobx with your own model classes. Which is perfectly fine as well.

  1. Each object in a 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 a MST tree is a tree in itself. Any operation that can be invoked on the complete tree can also be applied to a sub tree.
  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 amount 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 an 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
        }
    })

Or, shorter 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 listens to any action that is invoked on the model or any of its descendants.
  • 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:

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 }
})

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 further 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 amountOfChildren() {
            return self.users.filter(user => user.age < 18).length
        },
        amountOfPeopleOlderThan(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.amountOfChildren, " children")
})
autorun(() => {
    console.log("There are now ", userStore.amountOfPeopleOlderThan(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
  • revertPatch(model, patch) reverse applies a patch (or array of patches) to the provided model. This replays the inverse of a set of patches to a model, which can be used to bring it back to its original state

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.identifier(types.refinement(types.string, 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.maybe(
    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"
})

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)
})

But, 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 => {
        const 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:

const Todo =  types.model({}).extend(self => {
    let localState = 3

    return {
        views: {
            get x() {
                return localState
            }
        },
        actions: {
            setX(value) {
                localState = value
            }
        }
    }
})

model.volatile

In many cases it is useful to have volatile state that is observable (in terms of Mobx observables) and readable from outside the instance. In that case, in the above example, localState could have been declared as const localState = observable.box(3). Since this is such a common pattern, there is a shorthand to declare such properties, and the example above could be rewritten to:

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. The 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)
  2. The volatile properties will be only observable be 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.

Primitive types

  • types.string
  • types.number
  • 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(dispatcher?, types...) create a union of multiple types. If the correct type cannot be inferred unambiguously from a snapshot, provide a dispatcher function of the form (snapshot) => Type.
  • types.optional(type, defaultValue) marks an value as being optional (in e.g. a model). If a value is not provided 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.
  • 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(null)), null).
  • 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 Accepts any kind of serializable value (both primitive and complex), but assumes that the value itself is immutable and serializable.
  • types.compose(name?, type1...typeX), creates a new model type by taking a bunch of existing types and combining them into a new one

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(subType?) 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.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.

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 is the preProcessSnapshot hook. Because it is needed before instantiating model elements, it needs to be defined on the type itself:

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!")
        }
    }))
Hook Meaning
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!
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
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.
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

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

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

Api overview

See the full API docs for more details.

signature
addDisposer(node, () => void) Function to be invoked whenever the target node is to be destroyed
addMiddleware(node, middleware: (actionDescription, next) => any, includeHooks) Attaches middleware to a node. See middleware. Returns disposer.
applyAction(node, actionDescription) Replays an action on the targeted node
applyPatch(node, jsonPatch) Applies a JSON patch, or array of patches, to a node in the tree
applySnapshot(node, snapshot) Updates a node with the given snapshot
createActionTrackingMiddleware Utility to make writing middleware that tracks async actions less cumbersome
clone(node, keepEnvironment?: true | false | newEnvironment) Creates a full clone of the given node. By default preserves the same environment
decorate(handler, function) Attaches middleware to a specific action (or flow)
destroy(node) Kills node, making it unusable. Removes it from any parent in the process
detach(node) Removes node from its current parent, and lets it live on as standalone tree
flow(generator) creates an asynchronous flow based on a generator function
getChildType(node, property?) Returns the declared type of the given property of node. For arrays and maps property can be omitted as they all have the same type
getEnv(node) Returns the environment of node, see environments
getParent(node, depth=1) Returns the intermediate parent of the node, or a higher one if depth > 1
getParentOfType(node, type) Return the first parent that satisfies the provided type
getPath(node) Returns the path of node in the tree
getPathParts(node) Returns the path of node in the tree, unescaped as separate parts
getRelativePath(base, target) Returns the short path, which one could use to walk from node base to node target, assuming they are in the same tree. Up is represented as ../
getRoot(node) Returns the root element of the tree containing node
getIdentifier(node) Returns the identifier of the given element
getSnapshot(node, applyPostProcess) Returns the snapshot of the node. See snapshots
getType(node) Returns the type of node
hasParent(node, depth=1) Returns true if node has a parent at depth
hasParentOfType(node, type) Returns true if the node has a parent that satisfies the provided type
isAlive(node) Returns true if node is alive
isStateTreeNode(value) Returns true if value is a node of a mobx-state-tree
isProtected(value) Returns true if the given node is protected, see actions
isRoot(node) Returns true if node has no parents
joinJsonPath(parts) Joins and escapes the given path parts into a JSON path
onAction(node, (actionDescription) => void) A built-in middleware that calls the provided callback with an action description upon each invocation. Returns disposer
onPatch(node, (patch) => void) Attach a JSONPatch listener, that is invoked for each change in the tree. Returns disposer
onSnapshot(node, (snapshot) => void) Attach a snapshot listener, that is invoked for each change in the tree. Returns disposer
process(generator) DEPRECATED – replaced with flow
protect(node) Protects an unprotected tree against modifications from outside actions
recordActions(node) Creates a recorder that listens to all actions in node. Call .stop() on the recorder to stop this, and .replay(target) to replay the recorded actions on another tree
recordPatches(node) Creates a recorder that listens to all patches emitted by the node. Call .stop() on the recorder to stop this, and .replay(target) to replay the recorded patches on another tree
getMembers(node) Returns the model name, properties, actions, views, volatiles
resolve(node, path) Resolves a path (json path) relatively to the given node
resolveIdentifier(type, target, identifier) resolves an identifier of a given type in a model tree
resolvePath(target, path) resolves a JSON path, starting at the specified target
splitJsonPath(path) Splits and unescapes the given JSON path into path parts
typecheck(type, value) Typechecks a value against a type. Throws on errors. Use this if you need typechecks even in a production build.
tryResolve(node, path) Like resolve, but just returns null if resolving fails at any point in the path
unprotect(node) Unprotects node, making it possible to directly modify any value in the subtree, without actions
walk(startNode, (node) => void) Performs a depth-first walk through a tree
escapeJsonPath(path) escape special characters in an identifier, according to http://tools.ietf.org/html/rfc6901
unescapeJsonPath(path) escape special characters in an identifier, according to http://tools.ietf.org/html/rfc6901

A disposer is a function that cancels the effect it was created for.

Tips

Building with production environment

MobX-state-tree provides a lot of dev-only checks. They check the correctness of function calls and perform runtime type-checks over your models. It is recommended to disable them in production builds. To do so, you should use webpack's DefinePlugin to set environment as production and remove them. More information could be found in the official webpack guides.

Generate MST models from JSON

The following service can generate MST models based on JSON: https://transform.now.sh/json-to-mobx-state-tree

optionals and default value functions

types.optional can take an optional function parameter which will be invoked each time a default value is needed. This is useful to generate timestamps, identifiers or even complex objects, for example:

createdDate: types.optional(types.Date, () => new Date())

toJSON() for debugging

For debugging you might want to use getSnapshot(model, applyPostProcess) to print the state of a model. But if you didn't import getSnapshot while debugging in some debugger; don't worry, model.toJSON() will produce the same snapshot. (For API consistency, this feature is not part of the typed API)

Handle circular dependencies between files using late

On the exporting file:

export function LateStore() {
    return types.model({
        title: types.string
    })
}

In the importing file

import { LateStore } from "./circular-dep"

const Store = types.late(LateStore)

Thanks to function hoisting in combination with types.late, this lets you have circular dependencies between types, across files.

Simulate inheritance by using type composition

There is no notion of inheritance in MST. The recommended approach is to keep references to the original configuration of a model in order to compose it into a new one, for example by using types.compose (which combines two types) or producing fresh types using .props|.views|.actions. An example of classical inheritance could be expressed using composition as follows:

const Square = types
    .model(
        "Square",
        {
            width: types.number
        }
    )
    .views(self => ({
        surface() {
            return self.width * self.width
        }
    }))

// create a new type, based on Square
const Box = Square
    .named("Box")
    .views(self => {
        // save the base implementation of surface
        const superSurface = self.surface

        return {
            // super contrived override example!
            surface() {
                return superSurface() * 1
            },
            volume() {
                return self.surface * self.width
            }
        }
    }))

// no inheritance, but, union types and code reuse
const Shape = types.union(Box, Square)

Similarly, compose can be used to simply mixin types:

const CreationLogger = types.model().actions(self => ({
    afterCreate() {
        console.log("Instantiated " + getType(self).name)
    }
}))

const BaseSquare = types
    .model({
        width: types.number
    })
    .views(self => ({
        surface() {
            return self.width * self.width
        }
    }))

export const LoggingSquare = types
    .compose(
        // combine a simple square model...
        BaseSquare,
        // ... with the logger type
        CreationLogger
    )
    // ..and give it a nice name
    .named("LoggingSquare")

FAQ

When not to use MST?

MST makes state management very tangible by offering access to snapshots, patches and by providing interceptable actions. Also it fixes the this problem. All these features have the downside that they incur a little runtime overhead. Although in many places the MST core can still be optimized significantly, there will always be a constant overhead. If you have a performance critical application that handles huge amounts of mutable data, you will probably be better off by using 'raw' mobx. Which has predictable and well-known performance characteristics, and has much less overhead.

Likewise, if your application is mainly dealing with stateless information (such as a logging system) MST doesn't add much values.

Where is the any type?

MST doesn't offer an any type because it can't reason about it. For example, given a snapshot and a field with any, how should MST know how to deserialize it? Or apply patches to it? Etc. etc. If you need any there are following options

  1. Use types.frozen. Frozen values need to be immutable and serializable (so MST can treat them verbatim)
  2. Use volatile state. Volatile state can store anything, but won't appear in snapshots, patches etc.
  3. If your type is regular, and you just are too lazy to type the model, you could also consider generating the type at runtime once (after all, MST types are just JS...). But you will loose static typing and any confusion it causes is up to you to handle :-).

How does reconciliation work?

  • When applying snapshots, MST will always try to reuse existing object instances for snapshots with the same identifier (see types.identifier()).
  • If no identifier is specified, but the type of the snapshot is correct, MST will reconcile objects as well if they are stored in a specific model property or under the same map key.
  • In arrays, items without an identifier are never reconciled.

If an object is reconciled, the consequence is that localState is preserved and postCreate / attach life-cycle hooks are not fired because applying a snapshot results just in an existing tree node being updated.

Creating async flows

See creating asynchronous flow.

Using mobx and mobx-state-tree together

egghead.io lesson 5: Render mobx-state-tree Models in React

Yep, perfectly fine. No problem. Go on. observer, autorun etc. will work as expected.

Should all state of my app be stored in mobx-state-tree?

No, or, not necessarily. An application can use both state trees and vanilla MobX observables at the same time. State trees are primarily designed to store your domain data, as this kind of state is often distributed and not very local. For local component state, for example, vanilla MobX observables might often be simpler to use.

Can I use Hot Module Reloading?

egghead.io lesson 10: Restore the Model Tree State using Hot Module Reloading when Model Definitions Change

Yes, with MST it is pretty straight forward to setup hot reloading for your store definitions, while preserving state. See the todomvc example

TypeScript & MST

TypeScript support is best-effort, as not all patterns can be expressed in TypeScript. But except for assigning snapshots to properties we get pretty close! As MST uses the latest fancy Typescript features it is recommended to use TypeScript 2.3 or higher, with noImplicitThis and strictNullChecks enabled.

We recommend using TypeScript together with MST, but since the type system of MST is more dynamic than the TypeScript system, there are cases that cannot be expressed neatly and occassionally you will need to fallback to any or manually adding type annotations.

Using a MST type at design time

When using models, you write an interface, along with its property types, that will be used to perform type checks at runtime. What about compile time? You can use TypeScript interfaces to perform those checks, but that would require writing again all the properties and their actions!

Good news! You don't need to write it twice! Using the typeof operator of TypeScript over the .Type property of a MST Type will result in a valid TypeScript Type!

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

type ITodo = typeof Todo.Type // => ITodo is now a valid TypeScript type with { title: string; setTitle: (v: string) => void }

Due to the way typeof operator works, when working with big and deep models trees, it might make your IDE/ts server takes a lot of CPU time and freeze vscode (or others) A partial solution for this is to turn the .Type into an interface.

type ITodoType = typeof Todo.Type;
interface ITodo extends ITodoType {};

Snapshot types are limited

Until conditionally mapped types are available (scheduled for TS 2.8), the types of snapshots cannot be inferred correctly. But you will get some type assistence when using getSnaphot with types, like this:

const snapshot = getSnapshot<typeof Car.SnapshotType>(car)

Tip: recycle the interface

type ICarSnapshot = typeof Car.SnapshotType
const snapshot = getSnapshot<ICarSnapshot>(car)

For lazy folks:

const snapshot = getSnapshot<any>(car)

note: Even when typing snapshots, they will still not be as accurate as they could be until TS 2.8

Typing self in actions and views

The type of self is what self was before the action or views blocks starts, and only after that part finishes, the actions will be added to the type of self.

Sometimes you'll need to take into account where your typings are available and where they aren't. The code below will not compile: TypeScript will complain that self.upperProp is not a known property. Computed properties are only available after .views is evaluated.

For example:

const Example = types
  .model('Example', {
    prop: types.string,
  })
  .views(self => ({
    get upperProp(): string {
      return self.prop.toUpperCase();
    },
    get twiceUpperProp(): string {
      return self.upperProp + self.upperProp; // Compile error: `self.upperProp` is not yet defined
    },
  }));

You can circumvent this situation by declaring the views in two steps:

const Example = types
  .model('Example', { prop: types.string })
  .views((self: typeof Example.Type) => { // override the inferred type of self
      const views = {
        get upperProp(): string {
            return self.prop.toUpperCase();
        },
        get twiceUpperProp(): string {
            return views.upperProp + views.upperProp;
        }
      }
      return views
  }))

Note that you can also declare multiple .views block, in which case the self parameter gets extended after each block

const Example = types
  .model('Example', { prop: types.string })
  .views(self => {
    get upperProp(): string {
      return self.prop.toUpperCase();
    },
  }))
  .views(self => ({
    get twiceUpperProp(): string {
      return self.upperProp + self.upperProp;
    },
  }));

Similarly, when writing actions or views one can use helper functions:

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

const Example = types
  .model('Example', { prop: types.string })
  .actions(self => {
    // Don't forget that async operations HAVE
    // to use `flow( ... )`.
    const fetchData = flow(function *fetchData() {
      yield doSomething()
    })

    return {
      fetchData,
      afterCreate() {
        // Notice that we call the function directly
        // instead of using `self.fetchData()`. This is
        // because Typescript doesn't know yet about `fetchData()`
        // being part of `self` in this context.
        fetchData()
      }
    }
  });

Snapshots can be used to write values

Everywhere where you can modify your state tree and assign a model instance, you can also just assign a snapshot, and MST will convert it to a model instance for you. However, that is simply not expressible in static type systems atm (as the type written to a value differs to the type read from it). So the only work around is to upcast to any:

const Task = types.model({
    done: false
})
const Store = types.model({
    tasks: types.array(Task),
    selection: types.maybe(Task)
})

const s = Store.create({ tasks: [] })
// `{}` is a valid snapshot of Task, and hence a valid task, MST allows this, but TS doesn't, so need any cast
s.tasks.push({} as any)
s.selection = {} as any

Known Typescript Issue 5938

Theres a known issue with typescript and interfaces as described by: https://github.com/Microsoft/TypeScript/issues/5938

This rears its ugly head if you try to define a model such as:

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

export const Todo = types.model({
    title: types.string
});

export type ITodo = typeof Todo.Type

And you have your tsconfig.json settings set to:

{
  "compilerOptions": {
    ...
    "declaration": true,
    "noUnusedLocals": true
    ...
  }
}

Then you will get errors such as:

error TS4023: Exported variable 'Todo' has or is using name 'IModelType' from external module "..." but cannot be named.

Until Microsoft fixes this issue the solution is to re-export IModelType:

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

export type __IModelType = IModelType<any,any>;

export const Todo = types.model({
    title: types.string
});

export type ITodo = typeof Todo.Type

It ain't pretty, but it works.

Optional/empty maps

Optional parameters, including "empty" maps, should be either a valid snapshot or a MST instance. To fix type errors such as Error while converting {} to map, define your type as such:

  map: types.optional(types.map(OtherType), {})

How does MST compare to Redux

So far this might look a lot like an immutable state tree as found for example in Redux apps, but there're are only so many reasons to use redux as per article linked at the very top of redux guide that MST covers too, meanwhile:

  • Like Redux, and unlike MobX, MST prescribes a very specific state architecture.
  • mobx-state-tree allows direct modification of any value in the tree; it is not necessary to construct a new tree in your actions.
  • mobx-state-tree allows for fine-grained and efficient observation of any point in the state tree.
  • mobx-state-tree generates JSON patches for any modification that is made.
  • mobx-state-tree provides utilities to turn any MST tree into a valid Redux store.
  • Having multiple MSTs in a single application is perfectly fine.

Contributing

  1. Clone this repository
  2. Run yarn run bootstrap and yarn run build once.
  3. Extensive pull requests are best discussed in an issue first
  4. Have fun!

Thanks!

  • Mendix for sponsoring and providing the opportunity to work on exploratory projects like MST.
  • Dan Abramov's work on Redux has strongly influenced the idea of snapshots and transactional actions in MST.
  • Giulio Canti's work on tcomb and type systems in general has strongly influenced the type system of MST.
  • All the early adopters encouraging to pursue this whole idea and proving it is something feasible.