Ruby constant lookup: The good, the bad and the ugly

Updated . Posted . Visible to the public. Deprecated.

Since Ruby 2.5 top-level constant lookup ("the bad") has been removed.

Since Rails 6 autoloading ("the ugly") is done using Zeitwerk, which fixes most edge cases.

In Ruby, classes and modules are called constants. This card explains how Ruby resolves the meaning of a constant.

The good

E. g. in the following example, Array could mean either Foo::Array or simply Array:

class Foo
  def list
    Array.new
  end
end

What Ruby does here is to see if the name Array makes sense inside of Foo::, and if that fails, resolves it to ::Array (without a namespace).

The bad

This is relevant for old Ruby versions. Ruby 2.5+ removes top-level constant lookup which means that Array::String will not resolve to String and raise an error instead.

You might be surprised that these are all valid ways to reference Ruby's String class:

String
Array::String
Array::Hash::String

When you see Array::String, do not think "a class String inside the Array namespace". Rather think: "What does String resolve to from the viewpoint of the Array class?".

When you do this, Ruby will print a warning:

warning: toplevel constant String referenced by Array::String

However, that warning is usually lost in a sea of log messages. And here is where it gets ugly (see below).

The ugly

This part is a bit lengthy, but it allows you to debug strange bugs with Rails autoloading the wrong constants. Due to the constant lookup rules above, Rails sometimes guesses wrong and loads the wrong file. This happens when you heavily namespace your models. The mechanism how files are loaded differs in development and production, but the problem exist for both environment.

Development

During development, Rails unloads all classes after every request. This way code changes take effect immediately, without requiring you to restart the server. If you have been working with Rails for a while, you might even have forgotten that this is a feature of Rails. In the regular world of Ruby, classes don't refresh themselves automatically after a change.

Rails also autoloads classes on demand during development. This means every request begins with none of your models, controllers, loaded. Whenever Ruby encounters a constant name it doesn't know yet, it guesses the correct .rb file and requires the file. Again, this is a feature activated by Rails. If you write a regular Ruby script without Rails, you need to require all the files you want to use. Classes don't magically load themselves from disk.

Production

All files are loaded in a deterministic order once the server boots.


Take this example with some constants referencing each other:

# app/models/contract/document/uploader.rb
class Contract::Document::Uploader < CarrierWave::Uploader::Base
  ...
end

# app/models/document/uploader.rb
class Document::Uploader < CarrierWave::Uploader::Base
  ...
end


# app/models/document.rb
class Document < ActiveRecord::Base
  mount_uploader Document::Uploader
end


# app/models/contract.rb
class Contract < ActiveRecord::Base
  mount_uploader Contract::Document::Uploader
end

In the example above it can happen that the Contract class actually mounts the uploader Document::Uploader instead of the requested Contract::Document::Uploader.

  • In development this can happen randomly, as you never know in which order the files are autoloaded. If the autoload mechanism has already loaded app/models/document/uploader.rb, a constant lookup for Contract::Document::Uploader will never fail as it is actually a valid way to reference Document::Uploader. The file app/models/contract/document/uploader.rb is never autoloaded.

  • In production you either have the issue or not. The order app/models/document/uploader.rb > app/models/contract.rb > ... will have unexpected references, whereas the order app/models/contract/document/uploader.rb > app/models/document/uploader.rb > app/models/contract.rb > ... will reference the classes as expected.

Conclusion

  • You can't fix it with require
  • Never make a class name for which an existing class name is a suffix. E.g. don't create a class Api::User when you also have a ::User.
  • A workaround when you have control over every involved class name:
    • prefix the root level class with 'Generic' e.g. GenericDocument
    • then the names Contract::Document and Deal::Document are available
  • You can write require_dependency if the fix above is not possible
  • If you want to know more about constant lookup in ruby read "Everything you ever wanted to know about constant lookup in Ruby"
Henning Koch
Last edit
Emanuel
License
Source code in this card is licensed under the MIT License.
Posted by Henning Koch to makandra dev (2013-12-17 11:49)