A mirror of the `mobx-state-tree` API that allows constructing fast, read-only instances.

mobx-quick-tree is a wrapper around mobx-state-tree that adds a second, interface-identical type for each mobx-state-tree type you define that is high-performance, read-only, and no longer observable.

Why?

mobx-state-tree is great for modeling data and observing changes to it, but it adds a lot of runtime overhead! Raw mobx itself adds substantial overhead over plain old JS objects or ES6 classes, and mobx-state-tree adds more on top of that. If you want to use your MST data models in a non-reactive or non-observing context, all that runtime overhead for observability is just wasted, as nothing is ever-changing.

mobx-quick-tree implements the same API as MST and exposes the same useful observable instances for use in observable contexts, but adds a second option for instantiating a read-only instance that is 100x faster.

If mobx-state-tree instances are great for modeling within an "editor" part of an app where nodes and properties are changed all over the place, the performant, read-only instances constructed by mobx-quick-tree are great for use within a "read" part of an app that displays data in the tree without ever changing it. For a website builder for example, you might use MST in the page builder area where someone arranges components within a page, and then use MQT in the part of the app that needs to render those web pages frequently.

Two APIs

mobx-quick-tree has two APIs for building performant, read-only versions of your models:

• a 100% compatible, drop-in replacement for the mobx-state-tree API using types.model, types.compose, etc
• an ES6 class based API that performs even faster in read-only mode

Drop-in mobx-state-tree API compatibility

To begin using mobx-quick-tree, change all your import statements from importing mobx-state-tree to import mobx-quick-tree. mobx-quick-tree exports all the same utilities and objects and maintains the same robust TypeScript support mobx-state-tree users are used to.

For example, if you have a types.model defined, you can keep the definition entirely the same, but define it using the types object imported from mobx-quick-tree:

import { types } from "@gadgetinc/mobx-quick-tree"

// use types.model from MQT the exact same way you might use types.model from MST
const Todo = types.model("Todo", {
  name: types.string,
  done: types.boolean
}).actions(self => {
  setDone(done: boolean) {
    self.done = done;
  }
});

Once you have a mobx-quick-tree type defined, you can use it the same way you might use a mobx-state-tree type:

// create an observable instance with `create`, using mobx-state-tree under the hood
const instance = Todo.create({ name: "Hello", done: false });
// actions run like normal on observable instances
instance.setDone(true);

Each defined type also exposes a new .createReadOnly function for constructing read only versions of the type:

// create a readonly instance which is ~100x faster
const readOnlyInstance = Todo.createReadOnly({ name: "Hello read only", done: false });
readOnlyInstance.setDone(true); // will throw an error, the instance is readonly

.createReadOnly exists on models, arrays, maps, etc, and will be used throughout a tree started with a .createReadOnly call.

API coverage

mobx-quick-tree supports the same functionality as mobx-state-tree on observable instances, like:

• actions, views, and volatiles
• references
• snapshots
• patch streams
• middleware
• full type safety

Many of the pieces of functionality don't make sense to call on read only instances and will throw however. Functions that change data, like applyPatch or applySnapshot will work as documented in MST against observable instances, but will throw errors when run against read-only instances created with .createReadOnly.

Type-level functions like isModelType or isUnionType report type information correctly when run against types defined using mobx-quick-tree.

Mixing mobx-state-tree and mobx-quick-tree.

mobx-quick-tree does not support mixing types or instances defined with mobx-state-tree. mobx-quick-tree can co-exist in the same process, but will error if used in the same tree with mobx-state-tree.

If switching from mobx-state-tree to mobx-quick-tree, we recommend completely removing mobx-state-tree from your package.json, and switching all import statements over to import from mobx-quick-tree.

ES6 class-based Model API

If you need even more performance, mobx-quick-tree has an ES6 class-based API that replaces the pretty-but-slow MST API. The MST API design forces re-running each views or actions block for each and every instance defined, which adds a lot of overhead and is often not well handled by the JS VM. By using ES6 classes, each view and action can be defined only once at require time, and the prototype chain of the read-only objects can do a lot more of the heavy lifting.

These high-performance classes still allow accessing an observable instance that functions equivalently to a type defined using the mobx-state-tree API, so they can still be used in both observable and non-observable contexts.

Readonly instances created with the ES6 Class Model API from mobx-quick-tree are 3x faster than the readonly instances created with the MST style API, for a total of a 300x performance improvement.

To define an ES6 class model, you use a different API than mobx-state-tree that only mobx-quick-tree exports.

For example, a Todo class model can be defined like so:

import { register, ClassModel, action, types } from "@gadgetinc/mobx-quick-tree";

@register
class Todo extends ClassModel({
  name: types.string,
  done: types.boolean,
}) {
  @action
  setDone(done: boolean) {
    this.done = done;
  }
}

// create a readonly instance which is ~300x faster
const readOnlyInstance = Todo.createReadOnly({ name: "Hello read only", done: false });
readOnlyInstance.setDone(true); // will throw an error, the instance is readonly

// create an observable instance with `create`
const instance = Todo.create({ name: "Hello", done: false });
// actions run like normal on observable instances
instance.setDone(true);

The Class Model API works by using the class you define to power read only instances. Classes are fast to instantiate and are optimized well by JS VMs which makes them ideal for the high performance use case. For the observable instances, the Class Model API derives an equivalent type using mobx-state-tree's types.model, and configures it to have all the same properties, views, actions, and volatiles that the class does. The read-only class and writable observable type have identical interfaces at runtime, but will be instances of two different classes.

To access the derived mobx-state-tree type explicitly for a class model, you can use the static .mstType property.

Requirements for using Class Model API

To use the Class Model API and maintain compatibility with both read-only and observable instances, you must adhere to the following rules:

• All Class Models need to subclass a base class created by the ClassModel base class generator
• All Class Models need to be registered using the @register decorator
• All functions which mutate data must be decorated with the @action function. These are the functions that would be MST .actions() or raw mobx action functions.
• Any undecorated functions or getters on the class will become views. These functions can only read data, and aren't allowed to mutate it. These are the functions that would be MST .views() or raw mobx computed functions. Views can also be explicitly decorated with the @view decorator.
• All volatile properties must be registered with the @volatile decorator. These are the properties that would be modeled using MST's .volatile() API and are excluded from any snapshots.

Setting up a Class Model

Class models are defined using normal ES6 classes that are decorated and extend a special, dynamically generated base class. Class models store data by passing a list of typed properties to store to the ClassModel base class generator:

import { ClassModel, register, view, action } from "@gadgetinc/mobx-quick-tree";

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  // define a view with a function
  name() {
    // refer to properties of the model with `this`
    return `${this.year} ${this.model} ${this.make}`;
  }

  // define an action with a function, identify it as an action with the `@action` decorator
  @action
  setModel(model: string) {
    // set  properties of the model on `this`
    this.model = model;
  }
}

Each Class Model must be registered with the system using the @register decorator to be instantiated. @register is necessary for setting up the internal state of the class and generating the observable MST type.

Within Class Model class bodies, refer to the current instance using the standard ES6/JS this keyword. mobx-state-tree users tend to use self within view or action blocks, but Class Models return to using standard JS this for performance.

Creating instances of a class model

Once you have a Class Model class created, you can create both read-only instances and observable instances of it.

To create an observable instance of a Class Model, use the static .create function:

// create an observable instance
const observableInstance = Car.create({ make: "Toyota", model: "Prius", year: 2008 });

// access properties of the observable, will be observed if run within an observer
car.make; // => "Toyota"

// run actions to change data and fire observers by calling action functions
car.setModel("Camry");

To create an read only instance of a Class Model, use the static .createReadOnly function:

// create an read only instance
const readOnlyInstance = Car.createReadOnly({ make: "Toyota", model: "Prius", year: 2008 });

// access properties of the observable, will be observed if run within an observer
car.make; // => "Toyota"

// actions will throw if called on a readonly instance
// car.setModel("Camry"); => would throw an throw error

Differences in using a Class Model and a types.model

At runtime, observable instances of Class Models and types.model instances behave very similarly. Both are built atop mobx-state-tree, so views, mobx computeds, mobx-react observer components, and any other compatible component of the mobx ecosystem can observe properties on instances of either type.

At runtime, readonly instances of Class Models and readonly types.model instances behave similarly. Both are created with the .createReadOnly call.

If using TypeScript, Class Model instances and types.model instances should be treated slightly differently. mobx-state-tree uses the Instance helper to refer to instances of a type, like Instance<typeof Car>. Since Class Models are real classes, instances of them can be referred to with just the name of the class, like Car, without needing to use the Instance helper. This matches standard ES6 class behavior.

Quick reference for type-time differences:

Type	types.model model	Class Model model
Type of an instance	Instance<typeof Model>	Model (though Instance<typeof Model> works too)
Input snapshot of a model	SnapshotIn<typeof Model>	SnapshotIn<typeof Model>
Output snapshot of a model	SnapshotOut<typeof Model>	SnapshotOut<typeof Model>

Defining views

Class Models support views on instances, which are functions that derive new state from existing state. Class Model views mimic mobx-state-tree views defined using the .views() API on models defined with types.model. See the mobx-state-tree views docs for more information.

To define a view on a Class Model, define a function that takes no arguments or a getter within a Class Model body.

import { ClassModel, register, view } from "@gadgetinc/mobx-quick-tree";

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  // define a view with a function. any undecorated functions are automatically defined as views
  name() {
    // refer to properties of the model with `this`
    return `${this.year} ${this.model} ${this.make}`;
  }
}

const car = Car.createReadOnly({ make: "Toyota", model: "Prius", year: 2008 });
// run the view
car.name(); // => "2008 Toyota Prius"

Views can be defined as functions which don't take arguments like above, or as getter properties on the class body:

import { ClassModel, register, view } from "@gadgetinc/mobx-quick-tree";

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  // define a view as a property with a getter
  get name() {
    // refer to properties of the model with `this`
    return `${this.year} ${this.model} ${this.make}`;
  }
}

const car = Car.createReadOnly({ make: "Toyota", model: "Prius", year: 2008 });
// run the view
car.name; // => "2008 Toyota Prius"

Views are available on both read-only and observable instances of Class Models.

Views can also be made explicit with the @view decorator:

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  // define a view as a property with a getter and an explicit @view decorator
  @view
  name() {
    return `${this.year} ${this.model} ${this.make}`;
  }
}

Explicit decoration of views is exactly equivalent to an implicit declaration of views without a decorator.

Defining actions with @action

Class Models support actions on instances, which are functions that change the state of the instance or its children. Class Model actions are exactly the same as mobx-state-tree actions defined using the .actions() API on a types.model. See the mobx-state-tree actions docs for more information.

To define an action on a Class Model, define a function within a Class Model body, and register it as an action with @action.

import { ClassModel, register, action } from "@gadgetinc/mobx-quick-tree";

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  // define an action with a function
  @action
  setModel(model: string) {
    // set properties of the model on `this`
    this.model = model;
  }

  // actions don't need to take arguments
  @action
  addYear() {
    this.year = this.year + 1;
  }
}

const car = Car.create({ make: "Toyota", model: "Prius", year: 2008 });
car.year; // => 2008
car.addYear(); // run the action
car.year; // => 2009

Actions are only available on observable instances created with .create, and are present but will throw errors if created on instances created with .createReadOnly.

Asynchronous actions / flows

mobx-state-tree allows defining asynchronous actions using the flow helper. Asynchronous actions can't use async/await, and instead must use generator functions so that mobx-state-tree can wrap each chunk of execution with the appropriate wrappers. For more information on the generator-based async actions in mobx-state-tree, see the mobx-state-tree async actions docs.

To define an asynchronous action in a class model, wrap your action generator function in the flow() helper, assign it to the name on the class, and apply the @action decorator like you might with synchronous actions.

import { ClassModel, register, action } from "@gadgetinc/mobx-quick-tree";

@register
class Store extends ClassModel({
  data: types.string,
}) {
  // define an async action with a generator function, the flow() helper, and the @action decorator
  @action
  load = flow(function* (this: Store) {
    // use `yield` like you might use `await` to run promises
    this.data = yield getNewDataSomehow();
  });
}

const store = Store.create({ data: "" });

// call async actions with `await`
await store.load();

Creating asynchronous actions using generators works as follow:

• The action needs to be marked as generator, by postfixing the function keyword with a * and a name (which will be used by middleware), and wrapping it with flow
• The action still needs to be wrapped in the @action decorator
• For type safety, the action needs to explicitly take a this argument with the type of the model (this is a typescript limitation with type inference across these instance functions)

Your flow action function can do all the normal things that an async function can do, but you call async functions a bit differently.

• The action can be paused by using a yield statement. Yield always needs to return a Promise.
• If the promise resolves, the resolved value will be returned from the yield statement, and the action will continue to run
• If the promise rejects, the action continues and the rejection reason will be thrown from the yield statement
• Invoking the asynchronous action returns a promise. That will resolve with the return value of the function, or rejected with any exception that escapes from the asynchronous actions.

Defining volatile properties with @volatile

Class Models support volatile properties on instances, which are observable properties that are excluded from any snapshots. Volatiles last only for the lifetime of the object and are not persisted because they aren't serialized into snapshots or read out of incoming snapshots. Class Model volatiles are exactly the same as mobx-state-tree volatiles defined using the .volatile() API on a types.model.

Volatile tends to be most useful for implementation details, like timers, counters, transport objects like fetch requests, or other state that only matters for making other stuff work as opposed to saving source of truth data that might belong in a database. See the mobx-state-tree volatile docs for more information.

Volatiles can be referred to and used in views, and changed by actions. Volatile properties are observable, so you can rely on views recomputing if they change for observable instances.

To define an volatile property on a Class Model, define a type-only property within a Class Model body, and register it as a volatile @volatile that initializes the value:

import { ClassModel, register, action } from "@gadgetinc/mobx-quick-tree";

@register
class Loader extends ClassModel({}) {
  @volatile((_instance) => "ready");
  state!: string;

  @action
  reload() {
    this.state = "loading"
    // ... do something else
    this.state = "ready"
  }
}

const loader = Loader.create();
loader.reload();
loader.state // => "ready"

Volatile properties are initialized using the initializer function passed to the @volatile decorator. The initializer is passed the instance being initialized, and must return a value to be set as the value of the property.

Volatile properties are available on both read-only and observable instances. On read-only instances, volatiles will be initialized to the value returned by the initializer, and can't be changed after as actions are not available.

Readonly actions with @volatileAction

Readonly instances of MQT models don't support running actions -- they are read-only, so invoking any actions will throw. But, sometimes you want some measure of mutable state on a readonly instance that doesn't ever need to be persisted in a snapshot, but does need to change over the lifecycle of an object. Things like timers, references to external handles, DOM nodes, etc all don't belong in the snapshot, but do need to be referenced.

mobx-quick-tree supports a special kind of action which is available on readonly instances. These actions are only allowed to mutate volatile properties.

For example, we can create a stopwatch object that has a timer volatile property, and mutate the value of that volatile on both and observable class model instances.

import { ClassModel, register, action, volatileAction } from "@gadgetinc/mobx-quick-tree";

@register
class Stopwatch extends ClassModel({}) {
  @volatile(() => null);
  timer!: NodeJS.Timer | null;

  @volatileAction
  start(callback) {
    this.timer = setTimeout(callback, 5000)
  }

  @volatileAction
  stop() {
    clearTimeout(this.timeout)
    this.timeout = null;
  }
}

const watch = Stopwatch.createReadOnly()
watch.start(() => {}) // will work ok on both observable and readonly instances
watch.stop();

Note: Volatile actions will not trigger observers on readonly instances. Readonly instances are not observable because they are readonly (and for performance), and so volatiles aren't observable, and so volatile actions that change them won't fire observers. This makes volatile actions appropriate for reference tracking and implementation that syncs with external systems, but not for general state management. If you need to be able to observe state, use an observable instance.

Caching view values in snapshots with snapshottedView

For expensive views, mobx-quick-tree supports hydrating computed views from a snapshot. This allows read-only instances to skip re-computing the expensive view, and instead return a cached value from the snapshot quickly.

To hydrate a view's value from a snapshot, define a view with the @snapshottedView decorator.

import { ClassModel, register, view, snapshottedView } from "@gadgetinc/mobx-quick-tree";

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  @snapshottedView()
  get name() {
    console.log("computing name"); // pretend this is expensive
    return `${this.year} ${this.model} ${this.make}`;
  }
}

// create an observable instance
const car = Car.create({ make: "Toyota", model: "Prius", year: 2008 });
car.name; // => "2008 Toyota Prius" (logs "computing name")

// create a snapshot of the observable instance
const snapshot = {
  ...getSnapshot(car),
  name: car.name, // NOTE: you must add the snapshotted view's value to the snapshot manually
};

const readOnlyCar = Car.createReadOnly(snapshot);
readOnlyCar.name; // => "2008 Toyota Prius" (does not log "computing name")

Snapshotted views can transform the value from the snapshot before it is stored on the read-only instance. To transform the value of a snapshotted view, pass a createReadOnly function to the @snapshottedView decorator.

For example, for a view that returns a rich type like a URL, we can store the view's value as a string in the snapshot, and re-create the rich type when a read-only instance is created:

import { ClassModel, register, view, snapshottedView } from "@gadgetinc/mobx-quick-tree";

@register
class TransformExample extends ClassModel({ url: types.string }) {
  @snapshottedView<string>({
    createReadOnly(value, node) {
      return value ? new URL(value) : undefined;
    },
  })
  get withoutParams() {
    const url = new URL(this.url);
    for (const [key] of url.searchParams.entries()) {
      url.searchParams.delete(key);
    }
    return url;
  }

  @action
  setURL(url: string) {
    this.url = url;
  }
}

const example = TransformExample.create({ url: "https://example.com?foo=bar" });

const snapshot = {
  ...getSnapshot(example),
  withoutParams: example.withoutParams.toString(),
};

snapshot.withoutParams; // => "https://example.com"

const readOnlyExample = TransformExample.createReadOnly(snapshot);
readOnlyExample.withoutParams; // => URL { href: "https://example.com" }

If your snapshotted views need to emit patches when their values change, you can pass a shouldEmitPatchOnChange function that returns true to the @snapshottedView decorator, or you can pass true to setDefaultShouldEmitPatchOnChange to enable patch emission for all snapshotted views.

Snapshotted view semantics

Snapshotted views are a complicated beast, and are best avoided until your performance demands less computation on readonly instances.

On observable instances, snapshotted views act like normal views and are not populated from the snapshot.

On readonly instances, snapshotted views go through the following lifecycle:

• when a readonly instance is created, any snapshotted view values in the snapshot are memoized and stored in the readonly instance
• snapshotted views are never re-computed on readonly instances, and their value is always returned from the snapshot if present
• if the incoming snapshot does not have a value for the view, then the view is lazily computed on first access like a normal @view, and memoized forever after that

References to and from class models

Class Models support types.references within their properties as well as being the target of types.references on other models or class models.

For example, a Class Model can references another Class Model in it's properties with types.reference:

@register
class Make extends ClassModel({
  name: types.string,
}) {}

@register
class Car extends ClassModel({
  model: types.string,
  make: types.reference(Make),
}) {
  description() {
    return `${this.make.name} ${this.name}`;
  }
}

A Class model can also reference a MST API model defined with types.model:

const Make = types.model("Make", { name: types.string });

@register
class Car extends ClassModel({
  model: types.string,
  make: types.reference(Make),
}) {
  description() {
    return `${this.make.name} ${this.name}`;
  }
}

An MST API model defined with types.model can also reference a Class Model:

@register
class Make extends ClassModel({
  name: types.string,
}) {}

const Car = types
  .model({
    model: types.string,
    make: types.reference(Make),
  })
  .views((self) => ({
    description() {
      return `${self.make.name} ${self.name}`;
    },
  }));

Subclassing Class Models

Class Models support subclassing to build class heirarchies in the same way you might with normal ES6 classes. Class Model subclasses inherit properties, views, actions, and volatiles from the parent, and can add new properties, views, actions, and volatiles below. Subclasses are strongly typed and support the same observable/readonly as any other class model.

To subclass a Class Model, you can use the JavaScript builtin extends keyword to create a subclass:

@register
class Person extends ClassModel({
  name: types.string,
}) {
  @action
  setName(name: string) {
    this.name = name;
  }
}

@register
class Dancer extends Person {
  @action
  dance() {
    console.log("cool moves");
  }
}

const dancer = Dancer.create({ name: "Ni'jah" });
dancer.name; // #=> Ninjah , the inherited property from the parent class
dancer.dance();

Note: Subclasses of class models must be @register'd before use, just like their parents.

Adding observable properties in subclasses

To subclass a class model and add an observable property, you have to use a special .extends call when subclassing. extends accepts the same style of property declaration that the ClassModel constructor does, and will merge in the props passed with the original props from the parent.

@register
class Person extends ClassModel({
  name: types.string,
}) {
  @action
  setName(name: string) {
    this.name = name;
  }
}

@register
class Dancer extends Person.extend({ outfit: types.string }) {
  @action
  setOutfit(outfit: string) {
    this.outfit = outfit;
  }
  @action
  dance() {
    console.log(`cool moves in ${this.outfit}`);
  }
}

const dancer = Dancer.create({ name: "Ni'jah", outfit: "glam" });
dancer.name; // #=> Ninjah , the inherited property from the parent class
dancer.dance(); // cool moves in glam

dancer.setOutfit("catsuit");
dancer.dance(); // cool moves in catsuit

Dynamically defining Class Models using class expressions

Usually, Class Models are defined using top level ES6 classes exported from from a file. For advanced use-cases, classes can also be built dynamically within functions using ES6 class expressions. Generally, static classes defined with decorators are clearer and more performant, but for fancy class factories and the like you may want to use class expressions which MQT supports with slightly different syntax.

To define a class using a class expression, you can no longer use the decorator based API suggested above, as in the latest version of TypeScript, decorators are not valid within class expressions. They work just fine in named classes, but not in dynamically defined classes that are passed around as values.

Instead, you need to explicitly call the register function with the class, the list of decorators you'd like to apply to the class, and optionally a string class name:

// define an example function which returns a class model (a class factory)
const buildClass = () => {
  const klass = class extends ClassModel({
    key: types.string,
  }) {
    someView() {
      return this.key;
    }

    someAction(newKey: string) {
      this.key = newKey;
    }
  };

  return register(
    klass,
    {
      someView: view,
      someAction: action,
    },
    "Example",
  );
};

// invoke the class factory to define a class
const Example = buildClass();

const instance = Example.create({ key: "foo" });
instance.someAction("bar");

This pattern is most useful for class factories that create new classes dynamically. For example, we could build a class factory to define a Set of some other type:

import { ClassModel, types, register, view, action } from "@gadgetinc/mobx-quick-tree";

// define a class factory that produces a new class model implementing a set for any type
const buildSet = <T extends IAnyType>(type: T) => {
  const klass = class extends ClassModel({
    items: types.array(type),
  }) {
    has(item: Instance<T>) {
      return this.items.some((existing) => existing == item);
    }

    add(item: Instance<T>) {
      if (!this.has(item)) {
        this.items.push(item);
      }
    }

    remove(item: Instance<T>) {
      this.items.remove(item);
    }
  };

  return register(klass, {
    add: action,
    remove: action,
    has: view,
  });
};

// produce a set of numbers class
const NumberSet = buildSet(types.number);

// create an instance of the set
const set = NumberSet.create();
set.add(1);
set.add(2);

set.has(1); // => true
set.has(3); // => false

Dynamically subclassing Class Models

If you have bits of shared data or logic you want to re-use across classes in an MQT project, you can use the mixin pattern. MQT supports mixins by means of subclassing and the extends helper for adding new data properties to parent classes.

For example, we could create a Nameable mixin we can apply to two different classes to create two different subclasses, each getting some shared data or logic.

const Nameable = <Klass>(klass: Klass) => {
  class Named extends extend(klass, { name: types.string }) {
    get firstName() {
      return this.name.split(" ")[0];
    }
  }
};

class Human extends Nameable(ClassModel({ salutation: types.string })) {
  sayHello() {
    return `Hello ${this.firstName}!`;
  }
}

class Dog extends Nameable(ClassModel({ breed: types.string })) {
  pet() {
    return `Petted ${this.name} who is a very good ${breed}`;
  }
}

Class model snapshots

mobx-state-tree and mobx-quick-tree both support snapshotting the rich instances defined in JS land using the getSnapshot function, and both conform to the same set of rules. Snapshots are useful for persisting data from one place to another, and for later re-creating instances that match with applySnapshot or .create/.createReadOnly.

Snapshots in mobx-quick-tree apply in the same way as mobx-state-tree in that they can be serialized without issue to JSON. Instances are turned into snapshots according to the following rules:

• simple types like types.boolean or types.string are serialized as the equivalent JSON scalar type
• types.maybeNull will output null in the snapshot if no value is present
• types.maybe will be totally absent from the snapshot if no value is present
• types.array arrays are serialized as plain JS arrays
• types.map maps are turned into plain JS objects
• properties defined on models are all serialized into the snapshot
• actions, views, and volatiles on models will not be serialized at all into the snapshot
• references will be serialized to the snapshot as the value of the referenced node's identifier (and not re-serialize the whole referenced node)

Sharing functionality between class models

When converting types.model models that use types.compose, you will need a deeper refactoring. types.compose is an implementation of multiple inheritance which ES6 classes don't support. There are a couple ways to re-use functionality in Class Models:

• If the reusable chunk is only properties, you can define a constant of the properties, and spread it in the ClassModel({...}) base class definition. For example:

const SharedProps = {
  name: types.string,
  phoneNumber: types.string,
};

@register
class Student extends ClassModel({
  homeroom: types.string,
  ...SharedProps,
}) {}

@register
class Teacher extends ClassModel({
  email: types.string,
  ...SharedProps,
}) {}

• If the reusable chunk is just logic like views and actions, you can use a function to subclass. For example:

const addName = (klass) => {
  return class extends klass {
    name() {
      return this.firstName + " " + this.lastName;
    }
  };
};

@register
class Student extends addName(
  ClassModel({
    firstName: types.string,
    lastName: types.string,
    homeroom: types.string,
  }),
) {}

@register
class Teacher extends addName(
  ClassModel({
    firstName: types.string,
    lastName: types.string,
    email: types.string,
  }),
) {}

If the reusable chunk includes both properties and views or actions, you can combine the two techniques.

Converting a types.model to a Class Model

Changing a types.model into a Class Model requires two key changes:

• changing the syntax used to define the model
• switching any types.compose calls to become subclasses, or spread properties in the ClassModel base class factory.

Updating types.model syntax to become a ClassModel

To convert a types.model into a Class Model, you need to update the definition to use a class body. Here are the conversion rules:

• types.model("Name", {...properties...}) becomes class Name extends ClassModel({...properties...})
• the newly registered class needs to have the @register decorator
• any views defined in .views(...) blocks become functions defined on the class decorated with the @view decorator
• any actions defined in .actions(...) blocks become functions defined on the class decorated with the @action decorator, including flow() actions
• any volatile properties defined in .volatile() blocks become one @volatile property in the class per property.

For example, lets say we have this types.model model:

import { types } from "@gadgetinc/mobx-quick-tree";

const Car = types
  .model("Car", {
    make: types.string,
    model: types.string,
    year: types.number,
  })
  .views((self) => ({
    get name() {
      return `${self.year} ${self.make} ${self.model}`;
    },
    sku() {
      return `CAR-${self.year}-${self.model}`;
    },
  }))
  .actions((self) => ({
    setModel(model: string) {
      self.model = model;
    },
  }));

an equivalent Class Model would read:

import { types, register, action, view, ClassModel } from "@gadgetinc/mobx-quick-tree";

@register
class Car extends ClassModel({
  make: types.string,
  model: types.string,
  year: types.number,
}) {
  get name() {
    return `${self.year} ${self.make} ${self.model}`;
  }
  sku() {
    return `CAR-${self.year}-${self.model}`;
  }
  @action
  setModel(model: string) {
    self.model = model;
  }
}

Updating types.model volatiles to become ClassModel volatiles

In types.model models, volatiles are defined using the .volatile() call to add multiple properties, each with an initial value. In Class Models, volatiles are defined one at a time with the @volatile decorator.

For example, with this types.model model:

const Store = types.model("Store", {
  data: types.string;
}).volatile((self) => ({
  state: "not-started",
  finished: false
})).actions(self => ({
  load: flow(function *() {
    self.state = "loading"
    try {
      self.data = yield loadSomeData();
      self.state = "loaded";
    } catch (error) {
      self.state = "error";
    } finally {
      self.finished = true;
    }
  });
}));

we convert each volatile property into a @volatile call on the class body:

@register
class Store extends ClassModel({
  data: types.string;
}) {
  @volatile(() => "not-started")
  state: string

  @volatile(() => false)
  finished: boolean

  @action
  load = flow(function *(this: Store) {
    this.state = "loading"
    try {
      this.data = yield loadSomeData();
      this.state = "loaded";
    } catch (error) {
      this.state = "error;
    } finally {
      this.finished = true;
    }
  });
};