This addon provides an alternative model implementation to DS.Model that is compatible with the rest of the ember-data ecosystem.

Ember-data users define their schemas via DS.Model classes which explicitly state what attributes and relationships they expect. Having many such classes each explicitly defining their schemas provides a lot of clarity and a pleasant environment for implementing standard object oriented principles.

However, it can be an issue in environments where the API responses are not easily known in advance, or where they are so varied as to require thousands of DS.Models which can be a burden both to developer ergonomics as well as runtime performance.

ember-m3 lets you use a single class for many API endpoints, inferring the schema from the payload and API-specific conventions.

For example, if your API returns responses like the following:

{
  "data": {
    "id": "isbn:9780439708180",
    "type": "com.example.bookstore.Book",
    "attributes": {
      "name": "Harry Potter and the Sorcerer's Stone",
      "author": "urn:Author:3",
      "chapters": [{
        "name": "The Boy Who Lived",
        "mentionedCharacters": ["urn:Character:harry"],
        "readerComments": [{
          "id": "urn:ReaderComment:1",
          "type": "com.example.bookstore.ReaderComment",
          "name": "Someone or Other",
          "body": "I have it on good authority that this is part of a book of some kind",
        }]
      }],
    },
  },
  "included": [{
    "id": "urn:author:3",
    "type": "com.example.bookstore.Author",
    "attributes": {
      "name": "JK Rowling",
    },
  }],
}

You could support it with the following schema:

// app/services/m3-schema.js
//
// generated via `ember generate service m3-schema`
import DefaultSchema from 'ember-m3/services/m3-schema';

const BookStoreRegExp = /^com\.example\.bookstore\./;
const ISBNRegExp = /^isbn:/;
const URNRegExp = /^urn:/;

function computeValue(key, value, modelName, schemaInterface) {
   // the value is a reference
    if (typeof value === 'string' && (ISBNRegExp.test(value) || URNRegExp.test(value))) {
      return schemaInterface.reference({
        type: null,
        id: value,
      });
    }
    // The value is a nested model
    if (typeof value === 'object' && value !== null && typeof value.$type === 'string') {
      return {
        id: value.isbn,
        type: value.$type,
        attributes: value,
      };
    }
    // Otherwise return the raw value
    return value;
}

export default class Schema extends DefaultSchema {
  includesModel(modelName) {
    return BookStoreRegExp.test(modelName);
  }

  computeAttribute(key, value, modelName, schemaInterface) {
    if (Array.isArray(value)) {
      return schemaInterface.managedArray(value.map((v) => computeValue(key, v, modelName, schemaInterface)));
    }
  } else {
    return computeValue(key, value, modelName, schemaInterface);
  }
}

Notice that in this case, the schema doesn't specify anything model-specific and would work whether the API returns 3 different kinds of models or 3,000.

Model-specific information is still needed to handle cases that cannot be generally inferred from the payload (such as distinguishing Date fields). See the Schema section for details.

The benefits of using ember-m3 over DS.Model are:

• handle dynamic schemas whose structure is not known in advance
• handle relationship references at arbitrary points in the payload seamlessly (eg relationship references within POJO attributes)
• limit the payload size of schema information by inferring as much as possible from the structure of the payload itself
• more easily query arbitrary URLs, especially when the types of the returned models are not known in advance

The trade-offs made for this include:

• Having only one model class prevents the use of some OOP patterns: You can't add computed properties to only one model for instance, and will need to rely on a different pattern of helpers and utility functions
• Inferring the schema from the payload can make the client side code less clear as is often the case in "static" vs. "dynamic" tradeoffs

• ember install ember-m3

The existing store API works as expected. findRecord, queryRecord &c., will build a URL using the -ember-m3 adapter and create a record for the returned response using MegamorphicModel. Note that the actual name queried will be passed to the adapter so you can build URLs correctly.

For example

store.findRecord('com.example.bookstore.book', 'isbn:9780439708180');

Results in an adapter call

import MegamorphicModel from 'ember-m3/model';

findRecord(store, modelClass, id, snapshot) {
  modelClass === MegamorphicModel;
  snapshot.modelName === 'com.example.bookstore.book';
  id === 'isbn:9780439708180';
}

ember-m3 does not define an -ember-m3 adapter but you can define one in your app. Otherwise the default adapter lookup rules are followed (ie your application adapter will be used).

ember-m3 also adds store.queryURL. This is helpful for one-off endpoints or endpoints where the type returned is not known and you just want a thin wrapper around the API response that knows how to look up relationships.

store.queryURL(url, options);

Returns a promise that will resolve to

1. A MegamorphicModel if the [primary data][json-api:primary-data] of the normalized response is a resource.
2. A RecordArray of MegamorphicModels if the [primary data][json-api:primary-data] of the normalized response is an array of resources.

The raw API response is normalized via the -ember-m3 serializer. M3 does not define such a serializer but you can add one to your app if your API requires normalization to JSON API.

• url The URL path to query. The -ember-m3 adapter is consulted for its host and namespace properties.

  • When url is an absolute URL, (eg http://bookstore.example.com/books) or a network-path reference (eg //books), the adapter's host and namespace properties are ignored.
  • When url is an absolute path reference (eg /books) it is prefixed with the adapter's host and/or namespace if they are present.
  • When url is a relative path reference it is prefixed with the adapter's host and/or namespace, whichever is present. It is an error to call queryURL when url is a relative path reference and the adapter specifies neither host nor namespace.

• options additional options. All are optional, as is the options object itself.

  • options.method defaults to GET. The HTTP method to use.

  • options.params defaults to null. The parameters to include, either in the URL (for GET requests) or request body (for others).

  • options.queryParams defaults to null. The parameters that will be converted to query string and appended to the URL, especially useful for POST method which needs both URL with query string(options.queryParams) and payload data(options.params).

  • options.cacheKey defaults to null. A string to uniquely identify this request. null or undefined indicates the result should not be cached. It is passed to serializer.normalizeResponse as the id parameter, but the serializer is free to ignore it.

  • options.reload defaults to false. If true, make a request even if an entry was found under cacheKey. Do not resolve the returned promise until that request completes.

  • options.backgroundReload defaults to false. If true, make a request even if an entry was found under cacheKey. If true and a cached entry was found, resolve the returned promise immediately with the cached entry and update the store when the request completes.

  • options.adapterOptions defaults to undefined. The custom options to pass along to the queryURL function on the adapter.

When cacheKey is provided, the response is cached under cacheKey.

If the response contains a model with an id, that model will be cached under that id as well as under the cacheKey. The entry under the model's id and under the cacheKey will point to the same model. Changes to the model will be reflected in both the models retrieved by cacheKey and the models retreived by the model's id.

Using cacheKey with queryURL can be useful to show, eg dashboard data or any other data that changes over time.

Consider the following:

store.queryURL('/newsfeed/latest', { cacheKey: 'newsfeed.latest', backgroundReload: true });

In this example, the first time the user visits a route that makes this query, the promise will wait to resolve until the request completes. The second time the request is made the promise will resolve immediately with the cached values while loading fresh values in the background.

Note that what is actually cached is the result: ie either a MegamorphicModel or, more likely, a RecordArray of MegamorphicModels.

It is possible to do the same thing in stock Ember Data by making a @ember-data/model class to wrap your search results and querying via:

// app/models/news-feed.js
import Model, { hasMany } from '@ember-data/model';
export class NewsFeed extends Model {
  @hasMany('feed-item')
  feedItems;
}

// somewhere, presumably in a route
store.findRecord('news-feed', 'latest', { backgroundReload: true });

As with ember-m3 generally, similar functionality is provided without the need to create models and relationships within your app code.

Because models (or RecordArrays of models) are cached, the cache can be emptied automatically when the models are unloaded. In the case of RecordArrays of models, the entire cache entry is evicted if any of the member models is unloaded.

In cases where we need to manually insert into the cache, we can use cacheURL. As an example, we may need to compute a secondary cache key once we receive response from our API.

store.queryURL('/foo', { cacheKey }).then((result) => {
  const secondaryCacheKey = computeSecondaryCacheKey(result);
  store.cacheURL(secondaryCacheKey, result);
});

When we unload the model, we will evict both the initial cacheKey as well as secondaryCacheKey.

store.queryURL('/foo', { cacheKey: 'foo' }).then((result) => {
  store.cacheURL('bar', result);

  // Cache conceptually looks like: { foo: ..., bar: ...' }
  result.unloadRecord();
  // Cache is now empty
});

You have to register a schema to tell ember-m3 what types it should be enabled for, as well as information that cannot be inferred from the response payload. You can think of the schema as a service that represents the same information, more or less, as all of your DS.Model files.

When modeling a payload it is necessary to know what properties of the payload are attributes that should be accessible to the application & templates, what properties are relationships that should look up other models, and what properties should be ignored.

DS.Model achieves this by enumerating the attributes and relationships on a DS.Model subclass inside your app/models directory.

By contrast ember-m3 relies on application-wide conventions to know the difference between attributes and relationships and otherwise reports all properties returned by the API as accessible to the application.

For APIs with many models, the ember-m3 approach can produce a substantially smaller application. Similarly the approach uses fewer classes which reduces the runtime cost of relationships.

Schema is a service registered from app/services/m3-schema.js. For convenience, you can extend a default schema from ember-m3/services/schema. The schema should have following properties.

• includesModel(modelName) Whether or not ember-m3 should handle this modelName. It's fine to just return true here but this hook allows ember-m3 to work alongside DS.Model.

• computeAttribute(key, value, modelName, schemaInterface) A function that computes the value and type of an attribute.

An attribute can be:

1. A reference to a record that exists in the identity map
2. A nested m3 record that exists as a child of the m3 parent record
3. A simple value, like a POJO or a string
4. A managedArray of references, nested records or simple values

If the attribute is a reference return: schemaInterface.reference({ id, type }) where the object properties are id The id of the referenced model (either m3 or @ember-data/model) type The type of the referenced model (either m3 or @ember-data/model) null is also a valid type in which case id will be looked up in a global cache.

Note that attribute references are all treated as synchronous. There is no ember-m3 analogue to @ember-data/model async relationships.

If you are returning a nested m3 model, return: schemaInterface.nested({ id, type, attributes })

If you are returning a managed array, return: schemaInterface.managedArray([schemaInterface.nested(obj), someOtherValue])

If you are returning the value you can return the raw value without passing it through the schemaInterface call

For example, if we have a book object:

```json
{
  id: 'book-id:1',
  type: 'com.example.library.book',
  mostSimiliarBook: 'book-id:2'
  bestChapter: {
    number: 7,
    title: 'My chapter'
    characterPOV: 'urn:character:2'
  }
}
```
We would want  `model.get('mostSimilarBook')` to return the book object and the

model.get('bestChapter.characterPOV') to return the character model with id 2 as an object. This requires us to interpret mostSimiliarBook as a reference and bestChapter a nested m3 model and not a simple object.

We would write the following method.


```js
computeAttribute(key, value, modelName, schemaInterface) {
  if (key === 'mostSimilarBook') {
    return schemaInterface.reference({
      type: 'com.example.library.book',
      id: value
    })
  } else if (key === 'bestChapter') {
    return schemaInterface.nested({
      type: 'chapter',
      attributes: value
    })
  }
}
```

• setAttribute(modelName, attrName, value, schemaInterface) A function that can be used to update the record-data with raw value instead of resolved value. schemaInterface.setAttr(key,value) should be invoked inside the function to set the value. If this function is not provided, m3 will set value as is.

  Example:

  setAttribute(modelName, attrName, value, schemaInterface) {
    // Check if the value is resolved as model
    // update attribute record-data with id information.
    if (value && value.constructor && value.constructor.isModel) {
      schemaInterface.setAttr(attrName, value.get('id'));
    }
  }

• isAttributeResolved(modelName, attrName, value, schemaInterface) A function that determines whether a value that is being set should be treated as resolved or not. Unresolved values that are set will be resoled when they are next accessed -- resolved values are cached upon being set.

  Example:

  isAttributeResolved(modelName, attrName, value, schemaInterface) {
    if (Array.isArray(value)) {
      // treat all arrays as unresolved without examining their contents
      return false;
    } else {
      return super.isAttributeResolved(...arguments);
    }
  }

• computeAttributes(keys, modelName) Compute the actual attribute names, default just return the array passed in. This is useful if you need to "decode/encode" your attribute names in a certain form, e.g., add a prefix when serializing.

• models an object containing type-specific information that cannot be inferred from the payload. The models property has the form:

  {
    models: {
      myModelName: {
        attributes: [],
        defaults: {
          attributeName: 'defaultValue',
        },
        aliases: {
          aliasName: 'attributeName',
        }
        transforms: {
          attributeName: transformFunctionn /* value */
        }
      }
    }
  }

  The keys to models are the types of your models, as they exist in your normalized payload.

  • attributes A list of whitelisted attributes. It is recommended to omit this unless you explicitly want to prevent unknown properties returned in the API payload from being read. If present, it is an array of strings that list whitelisted attributes. Reads of non-whitelisted properties will return undefined.

  • defaults An object whose key-value pairs map attribute names to default values. Reads of properties not included in the API will return the default value instead, if it is specified in the schema.

  • aliases Alternate names for payload attributes. Aliases are read-only, ie equivalent to Ember.computed.reads and not Ember.computed.alias

  • transforms An object whose key-value pairs map attribute names to functions that transform their values. This is useful to handle attributes that should be treated as Dates instead of strings, for instance.

    function dateTransform(value) {
      if (!value) { return; }
      return new Date(Date.parse());
    }
    
    {
      models: {
        'com.example.bookstore.book': {
          transforms: {
            publishDate: dateTransform,
          }
        }
      }
    }

• useUnderlyingErrorsValue(modelName) Helps the model.js determine whether the errors attribute should be read from the underlying data payload. The default return is false which creates an object compatible with how Ember Data treats errors property. Return true to read from the data payload for the model.

• useNativeProperties(modelName) If true is returned, removes the need to use .set and .get on m3 record of a given type. Instead of model.get('someAttribute') and model.set('someAttribute), you can do model.someAttribute and model.someAttribute = value. When set to false deprecates your current . access to aid in the migration. For migration and deprecation guide see the deprecations guide.

ember-m3 will use the -ember-m3 adapter to make queries via findRecord, queryRecord, queryURL &c. Responses will be normalized via the -ember-m3 serializer.

ember-m3 provides neither an adapter nor a serializer. If your app does not define an -ember-m3 adapter, the normal lookup rules are followed and your application adapter is used instead

It is perfectly fine to use your application adapter and serializer. However, if you have an app that uses both m3 models as well as DS.Models you may want to have different request headers, serialization or normalization for your m3 models. The -ember-m3 adapter and serializer are the appropriate places for this.

To learn how to debug m3 records, refer to the debugging documentation

For help with migrating deprecations refer to the deprecations guide

If your app customizes the store service, it will need to import and extend the store service provided by ember-m3 instead of the store provided by @ember-data/store. Example:

import M3Store from 'ember-m3/services/store';

export default class AppStore extends M3Store {}

If you are converting an application that uses DS.Models (perhaps because it has a very large number of them and ember-m3 can help with performance) you may have some patterns in your model classes beyond schema specification.

There are no particular requirements around refactoring these except that when you only have a single class for your models you won't be able to use typical object-oriented patterns.

The following are simply recommendations for common patterns.

Use the schema defaults feature to replace constant values in your DS.Model classes. For example:

// app/models/my-model.js
import Model from '@ember-data/model';

export Model.extend({
  myConstant: 24601,
});

// convert to

// app/initializers/schema-initializer.js
{
  models: {
    'my-model': {
      defaults: {
        myConstant: 24601,
      }
    }
  }
}

Use the schema aliases feature to replace use of Ember.computed.reads. You can likely do this also to replace the use of Ember.computed.alias as quite often they can be read only.

// app/models/my-model.js
import Model, { attr } from '@ember-data/model';

export Model.extend({
  name: attr(),
  aliasName: Ember.computed.reads('name'),
});

// convert to

// app/initializers/schema-initializer.js
{
  models: {
    'my-model': {
      aliases: {
        aliasName: 'name',
      }
    }
  }
}

Let's say you are converting the following Museum model:

// models/museum.js
import Model, { attr } from '@ember-data/model';

export Model.extend({
  name: attr(),
})

And that for Bad Reasons™ you discover that your team has been stashing a custom object on the museum describing some ad-hoc state (maybe for the ui?):

Ember.set(museum, 'retrofit', retrofitState);

Let's say this state has a formal class:

const RetrofitState = Ember.Object.extend({
  statusText: Ember.computed('statusCode', function () {
    let code = this.get('statusCode');

    switch (code) {
      case 0:
        return 'Not started';
      case 1:
        return 'In Progress';
      case 2:
        return 'Incomplete, on hold';
      case 3:
        return 'Completed';
      default:
        return 'Unknown';
    }
  }),
});

While you should not store local-state/ui-state (e.g. any state not part of the schema) on records, you can make this pattern temporarily work with M3 by doing a Bad Thing™ and giving the class constructor a static isModel flag:

RetrofitState.isModel = true; // THIS COMES WITH CONSEQUENCES

This is not without consequences. Setting this flag makes M3 treat this object as a resolvedValue, meaning that it will be included as an attribute when snapshot.eachAttribute is called by a serializer. This is very likely not what you want and very likely will cause "spooky action at a distance" bugs for others on your team (like suddenly sending serialized information about retrofits to the API).

Before saving these records, you would need to carefully scrub it by deleting this and any other local properties off of it, or you would need to ensure that the serializer did not serialize this attribute. This will be tedious, annoying and brittle, but that is the sacrifice paid for such Bad Things™.

Ultimately, you should refactor your application away from this Bad Practice™ to pass these separate objects alongside each other, for instance by wrapping them in an hash like the following:

let museumRetrofit = {
  museum,
  retrofit,
};

More involved computed properites can be converted to either utility functions (if used within JavaScript) or helper functions (if used in templates).

For properties used in both templates and elsewhere (eg components) a convenient pattern is to define a helper that exports both.

// app/models/my-model.js
import Model, { attr } from '@ember-data/model';

export Model.extend({
  name: attr('string'),
  sillyName: Ember.computed('name', function() {
    return `silly ${this.get('name')}`;
  }).readOnly(),
});

{{! some-template.hbs }}
{{model.sillyName}}
{{my-component name=model.sillyName}}

// app/routes/index.js
let sn = model.get('sillyName');

Coverted to

// app/helpers/silly-name.js
export function getSillyName(model) {
  if (!model) {
    return;
  }
  return `silly ${model.get('name')}`;
}

function sillyNameHelper(positionalArgs) {
  if (positionalArgs.length < 1) {
    return;
  }

  return getSillyName(positionalArgs[0]);
}

export default Ember.Helper.helper(sillyNameHelper);

{{! some-template.hbs }}
{{silly-name model}}
{{my-component name=(silly-name model)}}

// app/routes/index.js
import { getSillyName } from '../helpers/silly-name';

// ...
let sn = getSillyName(model);

ember-m3 does not impose any particular requirements with saving models. If your endpoints cannot reliably be determined via snapshot.modelName it is recommended to add support for adapterOptions.url in your adapter. For example:

// app/adapters/-ember-m3.js
import ApplicationAdapter from './application';
export default ApplicationAdapter.extend({
  findRecord(store, type, id, snapshot) {
    let adapterOptions = snapshot.adapterOptions || {};
    let url = adapterOptions.url;
    if (!url) {
      url = this.buildURL(snapshot.modelName, id, snapshot, 'findRecord');
    }

    return this.ajax(url, 'GET');
  },

  // &c.
});

// somewhere else, perhaps in a route
this.store.findRecord('com.example.bookstore.book', 1, { url: '/book/from/surprising/endpoint' });

• Ember >= 3.16.x < 4.x
• Ember Data >= 3.16.x < 4.x
• Node >= 14.x (i.e. an active version)

ember-m3 does not require all of EmberData to function properly, and if your app does not need all of EmberData either then you can choose to use ember-m3 with only the subset of EmberData packages ember-m3 currently requires.

As of 13 May 2020 this means:

• @ember-data/store >= 3.16
• @ember-data/model >= 3.16
• ember-inflector >= 3.0