Ways to Decompose Fat ActiveRecord Models | Hacker News Reader

Ways to Decompose Fat ActiveRecord Models

blog.codeclimate.com

85 pointslest20 comments

Ok, let me throw out a different perspective, why are you even using an ActiveRecord Model as your entities in the first place. That in itself violates the Single Responsibility Principle by attaching the persistance mechanism to the entity itself.

What if your entities were just objects that held data and did validation, but you let use case objects determine the behavior of your system beyond that? Data persistance at that point is literally persisting your entities to the DB.

Then you use the DB much more like you would a filesystem - to retrieve and save data. It doesn't determine your model, it just stores and retrieves your data.

So, you end up with 3 types of things in this system... entities, use cases, and data gateways.

Your data gateways can still use AR if you want, or something else, it doesn't matter.

This isn't my idea, Uncle Bob lays it out better than I can here: http://blog.8thlight.com/uncle-bob/2012/08/13/the-clean-arch...

Ok, time to embarrass myself due to lack of Ruby/programming knowledge. A few questions that perhaps someone can clear up as from my OO perspective this, to me, is all over the place. I must admit I'm only just getting started on Ruby.

Firstly, that doesn't look like the strategy pattern to me, isn't it back to front? And even if it were, what the hell are you doing? You do not pass the user to the authenticator, you'd pass the authenticator to the user constructor.

The way you've chosen is very brittle, it's a sure fire way to accidentally shoot yourself in the foot later on when someone accidentally deletes the authenticator or adds a new code path that doesn't contain one.

I also don't understand why you're creating a new class per object query? Ditto for the policy stuff. Why not just use a repository object if you don't want to clutter your main class. Like OrderRepository.GetByCompany(Id).

As for point 7, I don't understand why you're not completely extracting the facebook integration from the comment class. Does Ruby not have events? Why aren't you firing an event that the facebook integrator that initialized on the user object subscribed to? i.e. make a facebook integrator that registers itself on the user object creation.

Also, "View Model" not "View Object", that's what they're called, a lot of other frameworks already use them.

Good points/questions. I'll try to respond to each...

* RE: Strategy pattern. My understanding of the strategy pattern is it simply refers to "algorithms are encapsulate and can be selected at runtime". (http://en.wikipedia.org/wiki/Strategy_pattern)

* RE: New class per object query. Agreed that grouping these can make sense. Had to keep the example brief.

* RE: Events. That's another approach -- thanks for the suggestion.

* RE: "View Model" vs. "View". I had it as "View Model" in the original draft and got feedback from reviewers that they are usually called just "Views". :-) I've heard it both ways .

Thanks for the questions!

-Bryan

This is more of a theoretical question, but at what point does a project become large enough that concerns like this begin to truly matter?

When doing a small, limited-use project, it seems often that trying to follow "best practices" like avoiding fat models would be more trouble than it's worth. And yet I've worked on larger enterprise-scale projects that have most certainly benefited from following this and other practices.

Anyone know of any research or work on where the tipping point is for following increasingly complicated patterns and practices?

I don't have any research, just a rule of thumb: refactor when it hurts. In my experience, it's easy to go too far in the other direction, trying to craft perfectly generic code on the first pass. That doesn't work very well.

It's important to have a meta-awareness while programming - how firm is your grasp on the code? How confident are you that your changes will do what you expect? At some point, you feel that grasp slipping, and then it's time to refactor.

That's a great summary. It's easy to lose that meta-awareness, even if you're an experienced coder. You are holding a hammer, and the nail isn't going in, so your first instinct it "hammer harder". But usually it means you need a screwdriver.

I agree, great summary, thanks for the insight.

I try to follow a gradual approach to decomposition. When I start to get a lot of related methods I wrap them in a module, but still within the model definition(I just include the module immediately after definition). From there I might just move it to a separate file as is or perhaps I turn into a decorator and add an accessor for it on the model e.g.

  class Retailer
    ...
    def metrics
      RetailerMatrics.new(self)
    end
  end

  class RetailerMetrics < Struct.new(:retailer)
    def profit
      # devious stat twisting
    end
  end

The biggest danger with any of the patterns from the article is that you start using them before you actually need them. Then you just end up with a lot of complexity for no benefit.

Yes! Gradual is they key. Your architecture should scale up gradually to handle the app complexity.

I don't know anything about Ruby, but it's interesting to see typical "enterprisey" patterns appear for people using these scripting frameworks for web-app development. One thing I've always been curious about when evaluating Python frameworks, or looking at Ruby on Rails sample code in Github was how apps look when they start getting large. I usually see a mess, and it sort of turned me off. It's nice to know that some of the same patterns are being though of and applied in these worlds, too.

Also some good comments below on not just jumping into using all these patterns on smaller projects. Take the benefit of being light and nimble starting, then start breaking things apart as the app and teams grow. Best of both worlds.

I like Ruby on Rails, but "coming" from a Java background I find it odd to have lots of stuff in the models instead of using utility classes or services.

Coming from a java background to rails I found it entirely refreshing that everything wasn't in a utility class or service. Decomposition of a system using utility classes and services(in the n-tier style) is to make your system more procedural rather than more OO. This article is encouraging more OO patterns, closer in spirit to DDD. The article doesn't actually go into this but all of the techniques described can be used to compose a class so that there is still a single API for each model in your system i.e. the decomposition into policies, decorators, services(of the domain type) etc. can be hidden from the user behind the model api.

That's a good point. It sounds like what you're describing is essentially a Facade. That can be useful if you need coordinated objects to manage complexity, but want to keep it simple for clients to use that part of the application. I sometimes create a Facade for a Ruby module that is a package of fine grained classes.

you shouldn't have so many if at all utility classes in java as well services should be used to compose process behavior not domain behavior IMHO service -> enroll user user -> domain behavior eg. concrete distinct action in responsibility of that domain object I hate when people write code like this get attributes of object do something with attributes set attributes back

instead having operation on object doing this and clearly encapsulate behavior and hopefully data as well

Cool stuff. @brynary, how do you organize your app? Are all seven techniques dumped into app/models or do you keep each technique in its own directory?

I wrote about that bit in my previous post:

http://blog.codeclimate.com/blog/2012/02/07/what-code-goes-i...

But really, in short, don't worry too much about it. Plan on reorganizing once or twice as you find what works for you.

-Bryan

Perhaps it's my lack of training or the projects I work on, but I've never seen a Rails model so large/cumbersome as to justify even one of these techniques.

Is this a case of fixing the wrong problem?

Not all apps require these techniques to stay maintainable. My background is working on larger Rails apps, built up over multiple years by teams of 6+ people, so the post is shaped by those experiences.

I have a project which will benefit from Bryan's advice.

There are models which are overly large (the project started as a Rails 1.2 project now at 3.0). The project has moved through different hands and the cruft and bloat has built up.

I look forward to applying some of Bryan's points to these models.

Good article. Concise. Something I will come back and reference.

Reply on news.ycombinator.com