sorbet-rails

Gem Version Build Status codecov

A set of tools to make the Sorbet typechecker work with Ruby on Rails seamlessly.

This gem adds a few Rake tasks to generate Ruby Interface (RBI) files for dynamic methods generated by Rails. It also includes signatures for related Rails classes. The RBI files are added to a sorbet/rails-rbi/ folder.

sorbet-rails supports Rails 5+ or later.

Notice: we no longer supports Rails 4.2. Version 0.5.6 is the last version supporting Rails 4.2.

Initial Setup

  1. Follow the steps here to set up the latest version of Sorbet, up to being able to run srb tc.

  2. Add sorbet-rails to your Gemfile and install them with Bundler.

# -- Gemfile --
gem 'sorbet-rails'

Add sorbet-rails to the :default group because of "Features Provided at Runtime" below.

❯ bundle install
  1. Generate RBI files for your routes, models, etc ```sh ❯ bundle exec rake rails_rbi:routes ❯ bundle exec rake rails_rbi:models ❯ bundle exec rake rails_rbi:helpers ❯ bundle exec rake rails_rbi:mailers ❯ bundle exec rake rails_rbi:jobs ❯ bundle exec rake rails_rbi:custom

or run them all at once

❯ bundle exec rake rails_rbi:all


4. Update hidden-definition files and automatically upgrade each file's typecheck level:
```sh
❯ srb rbi hidden-definitions
❯ srb rbi suggest-typed

Because we've generated RBI files for routes, models, and helpers, a lot more files should be typecheckable now. Many methods in hidden.rbi may be removed because they are now typed.

Static RBI Generation

Models

This Rake task generates RBI files for all models in the Rails application (all descendants of ActiveRecord::Base):

❯ bundle exec rake rails_rbi:models

You can also regenerate RBI files for specific models. To accommodate for STI, this will generate rbi for all the subclasses of the models included.

❯ bundle exec rake rails_rbi:models[ModelName,AnotherOne,...]

The generation task currently creates the following signatures:

  • Column getters & setters
  • Associations getters & setters
  • Enum values, checkers & scopes
  • Named scopes
  • Model Relation class

It is possible to add custom RBI generation logic for your custom module or gems via the plugin system. Check out the plugins section below if you are interested.

We also add following methods to make type-checking more easily:

pluck_to_tstruct instead of pluck

The pluck method in Rails is a performant way to query value without instantiating ActiveRecord objects. However, it doesn't have any type information: it doesn't have type information (or name) of the attribute plucked. Sorbet-rails provides pluck_to_tstruct method as a replacement that returns an array of T::Struct instead. The attributes plucked is based on props defined in the T::Struct

# -- API
Arel.pluck_to_tstruct(TA[ <TStructSubClass> ].new)

# -- example
class WizardStruct < T::Struct
  const :name, String
  const :house, T.nilable(String)
end

Wizard.pluck_to_tstruct(TA[WizardStruct].new)  # T::Array[WizardStruct]
Wizard.all.pluck_to_tstruct(TA[WizardStruct].new)  # T::Array[WizardStruct]

This method is based on pluck_to_hash gem.

Enums

If you use Rails enum, sorbet-rails will generate a corresponding T::Enum. It will also include getters, setters, and scope methods in the rbi file it generates.

ie.

  enum house: {
    Gryffindor: 0,
    Hufflepuff: 1,
    Ravenclaw: 2,
    Slytherin: 3,
  }

Will generate this enum:

class Wizard::House < T::Enum
  enums do
    Gryffindor = new('Gryffindor')
    Hufflepuff = new('Hufflepuff')
    Ravenclaw = new('Ravenclaw')
    Slytherin = new('Slytherin')
  end
end

And these methods:

  sig { returns(T.nilable(String)) }
  def house; end

  sig { params(value: T.nilable(T.any(Integer, String, Symbol))).void }
  def house=(value); end

  sig { returns(T.nilable(Wizard::House)) }
  def typed_house; end

  sig { params(value: T.nilable(Wizard::House)).void }
  def typed_house=(value); end

Alternatively, you can replace your call to enum with typed_enum. This will hide the Rails methods (house) from Sorbet static-check (they are still usable in un-checked files).

  typed_enum house: {
    Gryffindor: 0,
    Hufflepuff: 1,
    Ravenclaw: 2,
    Slytherin: 3,
  }

Generates only typed enum setter & getter:

  sig { returns(T.nilable(Wizard::House)) }
  def typed_house; end

  sig { params(value: T.nilable(Wizard::House)).void }
  def typed_house=(value); end

RelationType alias

There are several kinds of relations of a model: User::ActiveRecord_Relation, User::ActiveRecord_AssociationRelation and User::ActiveRecord_Associations_CollectionProxy. Usually the code may need just any relation type. We add a Model::RelationType type alias for every model to use it.

class User
  RelationType = T.type_alias do
    T.any(
      User::ActiveRecord_Relation,
      User::ActiveRecord_AssociationRelation,
      User::ActiveRecord_Associations_CollectionProxy
    )
  end
end

Controllers

❯ bundle exec  rake rails_rbi:custom

sorbet-rails adds TypedParams to extact typed controller parameters.

TypedParams takes a parameter definition, which is a subclass of T::Struct and coerces the params object into an instance of that subclass using sorbet-coerce:

class MyCoolController < ApplicationController
  class MyActionParams < T::Struct
    const :id, T.nilable(Integer)
    const :show, T.nilable(T::Boolean)
    const :wands, T::Array[Integer]
  end
  sig { void }
  def my_action
    typed_params = TypedParams[MyActionParams].new.extract!(params)
    # T.reveal_type(typed_params) => MyActionParams
    # T.reveal_type(typed_params.show) => T.nilable(T::Boolean)
  end
end

If it fails to coerce the params into the right type, an ActionController::BadRequest exception will be raised with the coercion context for debugging.

Note: The API TypedParams[...].new.extract! may seem verbose, but necessary to support this feature. Ideally, the API can be simply TypedParams.extract!(...). However, sorbet doesn't support defining a method that accept a type and return an instance of the type. If this feature is supported by sorbet in the future, it will be easy to codemod to be TypedParams.extract(...)! part from your code. Note: require_typed and fetch_typed are deprecated in favor of TypedParams.

Routes

This Rake task generates an RBI file defining _path and _url methods for all named routes in routes.rb:

❯ bundle exec rake rails_rbi:routes

Helpers

This Rake task generates a helpers.rbi file that includes a basic module definition which includes the Kernel module and ActionView::Helpers, to allow for some basic Ruby methods to be used in helpers without Sorbet complaining.

❯ bundle exec rake rails_rbi:helpers

If you have additional modules that are included in all your helpers and you want helpers.rbi to reflect this, you can configure it:

# -- config/initializers/sorbet_rails.rb
SorbetRails.configure do |config|
  config.extra_helper_includes = ['ApplicationHelper', 'Devise::Controllers::Helpers']
end

Mailers

This Rake task generates RBI files for all mailer classes in the Rails application (all descendants of ActionMailer::Base)

❯ bundle exec rake rails_rbi:mailers

Since mailing action methods are based on instance methods defined in a mailer class, the signature of a mailing action method will be dependent on the signature the instance method has

  • If there is a (sorbet) sig written for the instance method, it generates a matching sig for the mailing action method
  • If not, all the params in the mailing action method will be T.untyped.
  • For return type though, the mailing action method will return ActionMailer::MessageDelivery instead of the return type of the instance method.

Jobs

This Rake task generates RBI files for all jobs classes in the Rails application (all descendants of ActiveJob::Base).

❯ bundle exec rake rails_rbi:jobs

It will generate perform_later and perform_now methods for the job classes matching the signature of the perform method of the job. If there is a (sorbet) sig written for perform, it will use the same sig for perform_* methods.

Runtime Features

In addition to features provided by the static generator, sorbet-rails can provide additional features when required. This is why the installation instructions specify that sorbet-rails should be placed in the :default group of the Gemfile, not a specific environment group (eg. development only).

  • Model: The gem provides some helper method to a model to make type-checking easier:

    • find_n, first_n, last_n
    • pluck_to_tstruct
    • typed_enum
  • Model Relation:

    • Make relation classes public. By default, relation classes like User::ActiveRecord_Relation, User::ActiveRecord_AssociationRelation are private
    • Add type alias, eg Model::RelationType, to represents any type of relation of a model.
  • Controller: use TypedParams to convert controller parameters to a typed structure

In addition to requireing sorbet-rails, you must also run bundle exec rake rails_rbi:custom, which will produce the RBI for these runtime features.

Discussion: #211, #214

Tips & Tricks

Look for # typed: ignore files

Because Sorbet's initial setup tries to flag files at whichever typecheck level generates 0 errors, there may be files in your repository that are # typed: ignore. This is because sometimes Rails allows very dynamic code that Sorbet does not believe it can typecheck.

It is worth going through the list of files that is ignored and resolve them (and auto upgrade the types of other files; see initial setup above). Usually this will make many more files able to be typechecked.

Overriding generated signatures if needed

sorbet-rails relies on Rails reflection to generate signatures. There are features this gem doesn't support yet such as serialize and attribute custom types. The gem also doesn't know the signature of any methods you have overridden. However, it is possible to override the signatures that sorbet-rails generates.

For example, here is how to override the signature for a method in a model:

# -- app/models/model_name.rbi --

# typed: strong
class ModelName
  sig { returns(T::Hash[...]) }
  def field_name; end

  sig { params(obj: T::Hash[....]).void }
  def field_name=(obj); end
end

Replace Rails.application.routes.url_helpers

When using url helpers like _url or _path methods outside of a controller, we usually have to add include Rails.application.routes.url_helpers in the class. However, Sorbet does not allow code that includes dynamic module. Sorbet Rails provides a drop-in replacement module for the dynamic url_helpers module, called GeneratedUrlHelpers. Following code change should resolve the sorbet type-check error:

class MyClass
-  include Rails.application.routes.url_helpers
+  include GeneratedUrlHelpers
end

find, first and last

These 3 methods can either return a single nilable record or an array of records. Sorbet does not allow us to define multiple signatures for a function (except stdlib). It doesn't support defining one function signature that has varying returning value depending on the input parameter type. We opt to define the most commonly used signature for these methods, and monkey-patch new functions for the secondary use case.

In short:

  • Use find, first and last to fetch a single record.
  • Use find_n, first_n, last_n to fetch an array of records.

find_by_<attributes>, <attribute>_changed?, etc.

Rails supports dynamic methods based on attribute names, such as find_by_<attribute>, <attribute>_changed?, etc. They all have static counterparts. Instead of generating all possible dynamic methods that Rails support, we recommend to use of the static version of these methods instead (also recommended by RuboCop).

Following are the list of attribute dynamic methods and their static counterparts. The static methods have proper signatures:

  • find_by_<attributes> -> find_by(<attributes>)
  • find_by_<attributes>! -> find_by!(<attributes>)
  • <attribute>_changed? -> attribute_changed?(<attribute>)
  • <attribute>_was -> attribute_was(<attribute>)
  • saved_change_to_<attribute>? -> saved_change_to_attribute?(<attribute>)
  • <attribute>_before_type_cast -> read_attribute_before_type_cast(<attribute>)

after_commit and other callbacks

Consider converting after_commit callbacks to use instance method functions. Sorbet doesn't support binding an optional block with a different context. Because of this, when using a callback with a custom block, the block is evaluated in the wrong context (Class-level context). Refer to this page for a full list of callbacks available in Rails.

Before:

after_commit do ... end

After:

after_commit :after_commit
def after_commit
  ...
end

If you wanted to make these changes using Codemod, try these commands:

# from methods like after_commit do <...> end
❯ codemod -d app/models/ --extensions rb \
  '(\s*)(before|after)_(validation|save|create|commit|find|initialize|destroy) do' \
  '\1\2_\3 :\2_\3\n\1def \2_\3'

# from methods like after_commit { <...> }
❯ codemod -d app/models/ --extensions rb \
  '(\s*)(before|after)_(validation|save|create|commit|find|initialize|destroy) \{ (.*) \}' \
  '\1\2_\3 :\2_\3\n\1def \2_\3\n\1\1\4\n\1end'

Note that Codemod's preview may show that the indentation is off, but it works.

unscoped with a block

The unscoped method returns a Relation when no block is provided. When a block is provided, unscoped calls the block and returns its result, which could be any type.

sorbet-rails chooses to define unscoped as returning a Relation because it's more common and more useful. If you want to use a block, either override the unscoped definition, or replace:

Model.unscoped do  end

with:

Model.unscoped.scoping do  end

select with a block

The select method in Rails has two modes: it can be given a list of symbols, in which case rails will only return the given columns from the database, or it can be given a block, in which case it acts like Enumerable.select and returns an array. We have chosen to support the first use case. If you want to pass a block to select, you can simply call to_a before you do. Note that this would be done within the select call anyway, so the performance penalty will be minimal.

flatten an array of relation

When you call flatten on an array of ActiveRecord::Relation, sorbet doesn't recognize that it will flatten the relation and return an array of model. The work around is to call to_a on the relation first.

# doesn't work
arr = [Model.recent, Model.old].flatten # T::Array[Model::ActiveRecord_Relation]

# work around
arr = [Model.recent, Model.old].map(&:to_a).flatten # T::Array[Model]

Avoid and_call_original in rspecs

If you run into the following issue when running rspec, it's likely because you're using expect(:method_name).and_call_original to mock a method in RSpec. We've found the double mock doesn't interact well with Sorbet's sig wrapper and caused flaky spec. The spec should be rewritten to expect the outcome of the method instead. (It still works with expect(:method_name) and expect(:method_name).and_return(...)

     RuntimeError:
       `sig` not present for method `:method_name` but you're trying to run it anyways. This should
       only be executed if you used `alias_method` to grab a handle to a method after `sig`ing it, but
       that clearly isn't what you are doing. Maybe look to see if an exception was thrown in your `sig`
       lambda or somehow else your `sig` wasn't actually applied to the method. Contact #dev-productivity
       if you're really stuck.

Extending Model Generation Task with Custom Plugins

sorbet-rails support a customizable plugin system that you can use to generate additional RBI for each model. This will be useful to generate RBI for methods dynamically added by gems or private concerns. If you write plugins for public gems, please feel free to contribute it to this repo.

Defining a Custom ModelPlugin

A custom plugin should be a subclass of SorbetRails::ModelPlugins::Base. Each plugin would implement a generate(root) method that generate additional rbi for the model.

At a high level, here is the structure of a plugin:

# -- lib/my_custom_plugin.rb
class MyCustomPlugin < SorbetRails::ModelPlugins::Base
  sig { override.params(root: Parlour::RbiGenerator::Namespace).void }
  def generate(root)
    # TODO: implement the generation logic
    # You can use @model_class and @available_classes here
  end
end

During the generation phase, the system will create a new instance of the plugin, with the model_class to be generated and a set of all available_classes (available models) detected in the Rails App.

We use Parlour gem to generate the RBI for a model. Please check out Parlour wiki for guide to add RBI, eg create a new module, class, or method in the generated file.

At a high level, you'd usually want to create a model-scoped module for your methods, and include or extend it in the base model class. The generation logic usually looks like this:

  def generate(root)
    # Make sure this is only run for relevant class
    return unless @model_class.include?(CustomModule)

    custom_module_name = self.model_module_name("CustomModule")
    custom_module_rbi = root.create_module(custom_module_name)

    # here we re-create the model class!
    model_class_rbi = root.create_class(self.model_class_name)
    model_class_rbi.create_extend(custom_module_name)

    # then create custom methods, constants, etc. for this module.
    custom_module_rbi.create_method(...)

    # this is allowed but not recommended, because it limit the ability to override the method.
    model_class_rbi.create_method(...)
  end

Notice that we re-create model_class_rbi here. Parlour's ConflictResolver will merge the classes or modules with the same name together to generate 1 beautiful RBI file. It'll also flag and skip if any method is created multiple times with conflict signatures. Check-out useful predefined module names & helper methods in model_utils.

It is also allowed to put methods into a model class directly. However, it is not recommended because it'll be harder to override the method. sorbet will enforce that the overriding method match the signature generated. It also makes the generated RBI file less modularized.

However, sometimes this is required to make sorbet recognize the signature. This is the case for class methods added by ActiveRecord::Concerns. Because ActiveSupport::Concern class methods will be inserted to the class directly, you need to also put the sig in the model class rbi directly.

It is also recommended to check if the generated methods are detected by sorbet as a gem method (in sorbet/rbi/gem/gem_name.rbi) or hidden methods (in sorbet/rbi/hidden-definitions/hidden.rbi). If so, you may need to re-run srb rbi hidden-definition or put method in the model class directly.

Check out the plugins written for sorbet-rails's own model RBI generation logic for examples.

Registering new plugins

You can register your plugins in an initializer:

# -- config/initializers/sorbet_rails.rb
SorbetRails::ModelRbiFormatter.register_plugin(MyCustomPlugin)

Enabling built-in plugins

sorbet-rails comes with a handful of gem plugins that can be enabled in an initializer. You can pass enabled gem plugins to config.enabled_gem_plugins, like so:

# -- config/initializers/sorbet_rails.rb
SorbetRails.configure do |config|
  config.enabled_gem_plugins = [
    :kaminari
  ]
end

These are the currently-supported gems and their symbolized names:

Gem Symbol
ElasticSearch :elastic_search
FriendlyId :friendly_id
Kaminari :kaminari
PgSearch :pg_search
Shrine :shrine
active_flag :active_flag
Paperclip :paperclip

You can also configure the core model plugins if needed. The default plugins are defined in the config. For the full list of plugin symbols, check out here.

Contributing

Contributions and ideas are welcome! Please see our contributing guide and don't hesitate to open an issue or send a pull request to improve the functionality of this gem.

This project adheres to the Contributor Covenant code of conduct. By participating, you are expected to uphold this code. Please report unacceptable behavior to [email protected].

License

MIT