Named Scopes Are Dead

Clarification: This post has been sitting unfinished on my disk for a couple of weeks but the update issues I ran into recently finally made me finish and publish it. I didn’t write this post to insult or bash Rails 3 or its contributors for the scope/arel bug I’m mentioning somewhere along the road. I’ve created a ticket in the Rails Lighthouse project that addresses the issue. If you have any insights or ideas with regards to the issue, please head over to the ticket and participate in the discussion.

The bug mentioned in this post has been fixed quite a while ago. Please read the updates.

About 2 years ago, Nick Kallen’s awesome has_finder plugin was brought to Rails in the form of named scopes. And I loved them dearly and used them a lot. In Rails 3, so far, I’ve found them to be obsolete – even harmful. Which is why I think they’re dead.

In this article, I’ll explain how named scopes were used in Rails 2, what was awesome about them and what wasn’t quite so awesome. Then I’ll take a look at what has changed in Rails 3 – mainly due to the advent of arel – and explain why you should think twice before using scopes in Rails 3.

What Named Scopes Used To Be

Named scopes’ awesomeness mainly came from the possibility to pass them around as objects, stack them etc. without them being evaluated. They were only evaluated when you tried to use them like any other ActiveRecord collection – like calling an iterator on them. Folks call this lazy loading. Here’s an example:

class Post < ActiveRecord::Base
  named_scope :published, :conditions => { :published => true }

class PostsController < ApplicationController
  def index
    @posts = Post.published # ... no query here ...
    # @posts = Post.published.all would trigger a query

# somewhere in the view:
@posts.each do |post| # here's where the query gets triggered
  # ...

One of the things it enabled you to do slightly more complex stuff in controllers without cluttering them up:

@posts = Post.published
@posts = @posts.by_author(@author) if @author
@posts = @posts.after(@start_date) if @start_date
# ...

Some people might argue that this stuff belongs into the model anyway (and I’m sure, we all remember having huge finder methods featuring the infamous with_scope method) but I say named scopes enabled a more balanced communication between the controller and its models.

But It’s Different Now

Different how? Well, thanks to arel, everything in Rails now is a scope of sorts anyway – a relation to be more exact. A relation only gets lazy loaded – just like a named scope used to. Take a look:

# in the controller
@posts = Post.where(["created_at >= ?", Time.current]) # no query yet ...
# again, @posts = Post.where(["created_at >= ?", Time.current]).all would trigger a query

# in the view
@posts.each do |post| # here's where it triggers the query
  # ...

This is somewhat similar to what you could do in Rails 2 by using the scoped method:

@posts = Post.scoped(:conditions => ["created_at >= ?", Time.current])

So effectively, every relation stays unevaluated until you hit it with a non-arel method like each or all. Read more in Pratik’s blog post on the new ActiveRecord Query API – no need for me to go into too much detail with him explaining it so well.

If you want to read more about relational algebra – the basic mathematical rules that arel is based on – there’s a pretty good wikipedia article on it.

Where’s The Harm?

You might argue that using named_scope (or, as it is in Rails 3, just scope) makes your models a bit more declarative and I tend to agree: It certainly looks good to expose most of your models’ APIs in class level statements like scope, has_many etc. and put them all on top. But let’s look at the downsides.

For starters, I’ve never quite liked the syntax to define dynamic scopes using lambdas:

scope :by_author, lambda { |author|
    :conditions => { :author_id => author.id }

There’s just too much punctuation flying around for my taste and you can’t use the do/end syntax unless you parenthesize the whole lambda, thanks to operator precedence. If the need for optional parameters arises, things can get really messy. Even more so, since Ruby 1.9 changes some things about lambdas – if you develop in both, 1.8 and 1.9, that can really wreck your brain.

But the real harm happens when you decide to ditch the hash syntax in favor of arel:

scope :published, where(:published => true)

Unknowingly, you might have just created an update blocker for your co-developers! Scopes – like associations – are evaluated when a class is loaded and arel checks that the underlying table (in this case posts) exists. Now imagine you check the new blog feature into your git repository, including the migration that actually creates the posts table. A co-developer tries to migrate and might just get the following:

~/tmp/scopes% rake db:migrate
(in /Users/clemens/tmp/scopes)
rake aborted!
Could not find table 'posts'

This happens if – by accident or intentionally – the post model is loaded somewhere as port the environment startup. This happens, for example, if you define an observer on the post model or extend some model in an initializer. To work around this, your co-developer would have to comment out the offending scopes, migrate and comment them in again. Seems tedious, huh? And what for? A little more expressiveness? And don’t forget that your continuous integration server needs to migrate every once in a while, too, and can’t just comment out a few lines …

Class Methods To The Rescue

Thanks to arel’s greatness, you can avoid the hassle and just go back to using good old class methods:

class << self
  def published
    where(:published => true)

Remember that relations stay unevaluated – and thus still chainable! – until they’re first used in a non-arel context.

This also helps to avoid the tedious lambda syntax that was needed for dynamic scopes:

class << self
  def by_author(author)
    where(:author_id => author.id)

No problems with optional parameters or differences between Ruby versions either – methods are just methods, after all. So after all, there is just one syntax for one thing again.

Should I Ditch Scopes Completely?

I say, for any new work you do, you definitely should avoid using scopes because of the update blocker it might become. Just use class methods instead as shown above.

If you’re porting a Rails 2 application to Rails 3, you’ll get a whole load of deprecation warnings because named_scope has been renamed to be just scope. If you want to avoid these deprecation warnings, you have to work through your code anyway, so you might as well just go ahead and turn scopes into class methods.

The only exception is if you want or need to use scope extensions. However, I’ve somehow managed to never have to use them so maybe you can avoid them, too.

If you want to keep some of scopes’ declarative nature, consider just putting them in a module named Scopes and extending the class like so:

class Post < ActiveRecord::Base
  module Scopes
    def by_author(author)
      where(:author_id => author.id)
  extend Scopes

This way, you keep scopes somehow separate from the rest of your models’ code.

If you feel that in any case you absolutely must use scopes, make sure you don’t convert use arel syntax for it and stick with the good old hash syntax for the time being. Keep in mind: Apparently, the hash syntax will be deprecated in Rails 3.1. But I guess Rails 3.1 won’t be out for some time so you should be fine for now.

In short: Named scopes are dead – long live arel!

Update (9/3/2010)

Iain Hecker correctly notes that using class methods was already possible in Rails 2 by using the scoped. Example:

class << self
  def by_author(author)
    scoped(:conditions => { :author_id => author.id })

You can, again, use this to circumvent the awkward lambda syntax but with Rails 2’s hash syntax you don’t do any harm in terms of update blockers.

Update 2 (21/4/2011)

Since some people have mentioned this post lately, I feel I should clarify that the bug that triggered this issue has been fixed quite a while ago. I’ve since mostly used scopes for “simple” rules and class methods for more complex rules because in my opinion it reads better.

Mainly, whether you use scopes or class methods has, again, become a matter of personal taste. My recommendation is to choose your own style or agree on one style with your fellow developers.