Sub classes do not inherit methods defined in models/*.js / models.json

[ritch]

/to @raymondfeng @bajtos

Consider this example:

// models.json
{
 MyBaseModel: {},
 SubModel: {
 options: { base: "MyBaseModel" }
 }
}

With methods defined in the base model like this:

// models/my-base-model.js
var MyBaseModel = require('../app').models.MyBaseModel;

MyBaseModel.myMethod = function() {}

...and finally somewhere else

// some-other-file.js
require('./app.js').models.SubModel.myMethod() // throws undefined is not a function

[ritch]

One potential solution is to create a sorted order using the settings.base property. Essentially create a dependency tree. Use the dependency tree to determine the sorted order to require() the models/*.js files in.

[raymondfeng]

I did something similar before. It's called topological sorting against a graph.

Sent from my iPhone 5

On Jun 10, 2014, at 6:00 PM, Ritchie Martori notifications@github.com wrote:

One potential solution is to create a sorted order using the settings.base property. Essentially create a dependency tree. Use the dependency tree to determine the sorted order to require() the models/*.js files in.

—
Reply to this email directly or view it on GitHub.

[bajtos]

This is already solved in the new project layout.

var loopback = require('loopback');

 // ensure the base class is defined
require('./my-base-model');

var SubModel = module.exports = loobpack.createModel({
 base: 'MyBaseModel'
});

One potential solution is to create a sorted order using the settings.base property. Essentially create a dependency tree. Use the dependency tree to determine the sorted order to require() the models/*.js files in.

I am not a fan of this idea, it makes index.js too dependent on the way how models are being defined in the .js files. To be more specific, index.js has to know the the base class of SubModel before it is defined. The only way how to find that is by loading submodel.json and expecting that submodel.js creates the model from the definition in submodel.json.

[ritch]

So the generated model file should include a require() to the base model? What is the purpose of the base: 'MyBaseModel' string then?

[bajtos]

So the generated model file should include a require() to the base model?

Either the base model, or the module providing the base model.

What is the purpose of the base: 'MyBaseModel' string then?

To allow yo loopback and Studio to edit/change the base model. Although now I see that changing the base property is not enough in this case :(

[ritch]

The requirement is that boot() should ensure a base model has been fully defined (any methods added) before a model that depends on it is defined.

[bajtos]

At the moment, boot is not defining any models, only configuring already defined models - see #304, #308 and strongloop/loopback-boot#9.

[ritch]

Sorry, by define I mean calling Model.extend(). Boot is actually calling that indirectly.

[bajtos]

Sorry, by define I mean calling Model.extend(). Boot is actually calling that indirectly.

loopback.boot and loopback-boot 1.x were calling Model.extend(), that's true. The current 2.0 of loopback-boot does not call Model.extend() at all.

I came up with a solution that should solve most if not all objections raised so far, I already started to work on it. Here is the overview:

• boot() loads all *.json files in modelSources directories and determines the order in which the models have to be defined using topological sort
• for each model to be defined, loopback.createModel(jsonConfig) is called. If there is a .js file with the same name as has the json file and it exports a function, this function is called to further customise the new model class.
• only models specified in models.json are defined - this should keep the size of the browserified bundle small.

Example

models/car.json

{
 "name": "car",
 "properties": {
 "id": {
 "type": "String",
 "required": true,
 "id": true
 },
 "make": {
 "type": "String",
 "required": true
 },
 "model": {
 "type": "String",
 "required": true
 }
 }
}

models/car.js

module.exports = function(Car, Base) {
 // Car is the model constructor
 // Base is the parent model (e.g. loopback.PersistedModel)

 Car.prototype.honk = function(duration, cb) {
 // make some noise for `duration` seconds
 cb();
 };
};

models.json

{
 "car": {
 "dataSource": "my-db"
 }
}

app.js

boot(app, {
 appRootDir: __dirname,
 // it is possible to specify multiple locations where to look for model files
 // paths are relative to appRootDir
 modelSources: ['./models']
});

@ritch @kraman @raymondfeng Thoughts?

[bajtos]

Here is the pull request implementing the change described above: strongloop/loopback-boot#10.

[bajtos]

For the record, this is the solution provided by loopback-boot#2.0 after strongloop/loopback-boot#11 was landed:

models/car.json

{
 "name": "car",
 "properties": {
 "id": {
 "type": "String",
 "required": true,
 "id": true
 },
 "make": {
 "type": "String",
 "required": true
 },
 "model": {
 "type": "String",
 "required": true
 }
 }
}

models/car.js

module.exports = function(Car, Base) {
 // Car is the model constructor
 // Base is the parent model (e.g. loopback.PersistedModel)

 Car.prototype.honk = function(duration, cb) {
 // make some noise for `duration` seconds
 cb();
 };
};

models.json

{
 "_meta": {
 "sources": ["./models"]
 },
 "car": {
 "dataSource": "my-db"
 }
}

app.js

boot(app, __dirname);

Note that _meta.sources is ["./models"] by default, so the _meta section can be omitted in simple applications. It is also possible to specify multiple locations where to look for model files.