2011-12-30
It is practically impossible to teach OO design to students that have had a prior exposure to Rails: as potential programmers they are mentally mutilated beyond hope of regeneration.Edsger W. Dijkstra (paraphrased)
I love ActiveRecord. It was the first ORM I’d ever interacted with. My first technical employer had commissioned a DBA, and so of course, we wrote all our own queries. Which was fine; I know my way around a JOIN or two. The problem came when it was time to make a new class; time to write “SELECT * FROM ‘tableName’ WHERE ‘id’=‘%’”… for each class. “Just copy one of the other small classes, hack out everything, and change the table names,” were my instructions. Fine. Whatever. But I knew there had to be a better way…
Along comes Rails. Holy crap, ActiveRecord is awesome! It writes the exact SQL I would have written myself! I don’t need to do anything. This is the best thing since sliced bread. But years later, not everything is rainbows and sunshine. I’ve written a lot of crappy code, and most of it is due to following Rails ‘best practices.’ Which is totally fine! Rails is kinda getting old these days. It’s no longer the new and shiny. But while Rails developers have gotten really good at busting out little CRUD apps, we haven’t moved web application design forward in a really, really long time.
And that’s what Rails did, really. That fifteen minute blog video was shocking. I know several people that threw away two or three week old projects and re-built their stuff in Rails within moments of watching it. And Rails has continued to lead the way in improving how we build rich applications on the web; the (not so but kinda whatever nobody but me cares) RESTful routing was great. The asset pipeline, though it has some bugs, has been great. The obsession with DRY has been great. The test obsession has been great. But Rails has also not been great in aiding us in writing maintainable software. Many of Rails’ design decisions make it difficult to write great software. It helps us write good software, but that isn’t enough anymore.
The real problem is that to truly move forward, Rails will have to re-invent itself, and I’m not sure that it can.
I’m not going to talk about all of the problems Rails has today, but I’d like to show you the biggest, most central one: ActiveRecord.
ActiveRecord’s greatest strength is also its problem: Tying class names to table names. This means that it’s impossible to de-couple your persistence mechanism from your domain logic. You can manage it through a combination of set_table_name, making a bunch of Repository classes, and careful coding… but then you might as well be using DataMapper. Except the Ruby library of the same name doesn’t really implement the DataMapper pattern all that well either, having the same issue of tying it all together.
This has another interesting effect: it led directly to the ‘fat model’ recommendation. While ‘get your stuff out of the controller’ is correct, it’s lead Rails developers to build huge, monolithic models that are hard to test, and violate SRP. It took me two and a half years to realize that Ruby classes in the models folder don’t have to inherit from ActiveRecord::Base. That is a problem.
We’ve gotten into a situation with a local minimum: our quest for great software has lead us to DRY ourselves into a corner. Now we have code that’s incredibly tightly coupled. Our ‘single responsibility’ is ‘anything and everything that tangentially relates to a Post.’
ActionController relies on instance variables to pass information from the controller to the view. Have you ever seen a 200 line long controller method? I have. Good luck teasing out which instance variables actually get set over the course of all those nested ifs.
The whole idea is kinda crazy: Yeah, it looks nice, but we literally just say ‘increase the scope of variables to pass data around.’ If I wrote a post saying “Don’t pass arguments to methods, just promote your data to a global” I’d be crucified. Yet we do the same thing (albeit on a smaller scale) every time we write a Rails application.
The whole idea of logic in templates leads to all kinds of problems. They’re hard to test, they’re hard to read, and it’s not just a slippery slope, but a steep one. Things go downhill rapidly.
What I’d really like to see is Rails adopting a ‘ViewModel + templates’ system, with logic-less templates and presenter-esque models that represent the views. The differences between Django’s idea of ‘views’ and Rails’ idea of ‘views’ are interesting here.
If you’ll notice, I basically have said that models are a problem. Controllers are a problem. Views are a problem. MVC has served the web well, even if it isn’t the GUI style MVC that named the pattern. But I think we’re reaching its limits; the impedance mismatch between HTTP and MVC, for example, is pretty huge. There are other ways to build web applications; I’m particularly excited about WebMachine. I don’t have a constructive alternative to offer here, I just know there’s a problem. I’m still mulling this one over.
I love Rails. I build software with it daily. Even with its flaws, it’s been a massive success. But because I love Rails, I feel like I can give it straightforward criticism: it’s easier to trash something you love. The real issue is that changing these things would require some really serious changes. It’d involve re-architecting large portions of things that people classically identify with Rails, and I’m not sure that Rails wants to or can do that.
This post is light on examples. I want this to be the starting point of a discussion, not the end of it. Expect more, in detail, from me in the future. What do you think? Are these problems pain points for you? Are they worth fixing? Are they actually problems?