Add --modelsRequire option. Permits to control the models loading order

[jacargentina]

My case is: model A, depends on model B, but keystone.import seems not to know that, so when loading that way from folder ./models i got the error list B not found

With this option, i can load the models in the needed order, as done in my keystone website itself, where i load the models by doing require('./models'); and have the ./models/index.js with

require('./b.js');
require('./a.js');

[creynders]

Not saying I won't merge this, but IMO having models depend on each other is a bad code smell. That said, maybe I don't need the keystone.import at all, come to think of it... I.e. maybe it should be reqmod( opts.models, process.cwd() ); by default?

[jacargentina]

An Article model depends on the User model (its author).
So, you must first load the User, and THEN the Article.

What is the bad code smell there ?

[creynders]

The case with Articles and Users is best solved through a relationship, i.e. there's no need for a direct dependency there.

Edit: just realised I misunderstood your use case. But to make sure, you mean model A uses keystone.list('B'), right?

[creynders]

The code smell is that the models are too tightly coupled and probably have fuzzied responsibilities and/or business logic. Possible solutions would be: merge them to one model (if applicable) or extract the dependency to a separate file.

For example if you have a model which is only valid if there are enough documents of another collection, the appropriate way would be to extract that business rule into a separate file like this:

//file: hooks/FooQux.js

const keystone = require('keystone');

keystone.list('Foo').schema.pre('save', function(done){
  keystone.list('Qux').model.count(function(n){
    if(n<=10){
        return done(new Error("Too few Qux documents!"));
    }
    done();
  })
});

Then in my keystone.js I have

keystone.import('./models');
keystone.import('./hooks');

[waltonryan]

One of our use cases is using foreign relationships in the pre and post save methods as you described. Using the hooks methodology is a bit overkill as it adds another layer of complexity to the code.

I think the ideal solution would be to load the models and their imports irregardless of their order in the filesystem.

The solution we use now is creating folders inside /models to handle dependent vs. independent models.

Update: @creynders starting to come around on your approach.