model_plugins.rdoc

doc/model_plugins.rdoc
Last Update: 2020-08-31 08:10:12 -0700

Model Plugins

Sequel::Model (and Sequel in general) is designed around the idea of a small core, to which application-specific behavior can easily be added. Sequel::Model implements this design using a plugin system. Plugins are modules that include submodules for model class methods, model instance methods, and model dataset methods. All plugins can override the class, instance, and dataset methods added by earlier plugins, and call super to get the behavior before the plugin was added.

Default Plugins

The Sequel::Model class is completely empty by default, in that it has no class methods or instance methods. Sequel::Model is itself a plugin, and it is the first plugin loaded, and it is loaded into itself (meta!). So methods in Sequel::Model::ClassMethods become Sequel::Model class methods, methods in Sequel::Model::InstanceMethods become Sequel::Model instance methods, and methods in Sequel::Model::DatasetMethods become Sequel::Model dataset methods. The Sequel::Model plugin is often referred to as the base plugin.

By default, the Sequel::Model class also has the Sequel::Model::Associations plugin loaded by default, though it is possible to disable this.

Loading Plugins

Loading a plugin into a model class is generally as simple as calling the Sequel::Model.plugin method with the name of the plugin, for example:

Sequel::Model.plugin :subclasses

What is does is require the sequel/plugins/subclasses file, and then assumes that that file defines the Sequel::Plugins::Subclasses plugin module.

It’s possible to pass module instances to the plugin method to load plugins that are stored in arbitrary files or namespaces:

Sequel::Model.plugin MyApp::Plugins::Foo

In the examples shown above, the plugin is loaded into Sequel::Model, which means it is loaded into all subclasses that are created afterward. With many plugins, you are not going to want to add them to Sequel::Model, but to a specific subclass:

class Node < Sequel::Model
  plugin :tree
end

Doing this, only Node and future subclasses of Node will have the tree plugin loaded.

Plugin Arguments/Options

Some plugins require arguments and/or support options. For example, the single_table_inheritance plugin requires an argument containing the column that specifies the class to use, and options:

class Employee < Sequel::Model
  plugin :single_table_inheritance, :type_id, model_map: {1=>:Staff, 2=>:Manager}
end

You should read the documentation for the plugin to determine if it requires arguments and what if any options are supported.

Creating Plugins

The simplest possible plugin is an empty module in a file stored in sequel/plugins/plugin_name somewhere in ruby’s load path:

module Sequel
  module Plugins
    module PluginName
    end
  end
end

Well, technically, that’s not the simplest possible plugin, but it is the simplest one you can load by name. The absolute simplest plugin would be an empty module:

Sequel::Model.plugin Module.new

Example Formatting

In general, loading plugins by module instead of by name is not recommended, so this guide will assume that plugins are loaded by name. For simplicity, we’ll also use the following format for example plugin code (and assume a plugin named Foo stored in sequel/plugins/foo):

module Sequel::Plugins::Foo
end

This saves 4 lines per example. However, it’s recommended that you use the nested example displayed earlier for production code.

The examples also assume that the following model class exists:

class Bar < Sequel::Model
end

Adding Class Methods

If you want your plugin to add class methods to the model class it is loaded into, define a ClassMethods module under the plugin module:

module Sequel::Plugins::Foo
  module ClassMethods
    def a
      1
    end
  end
end

This allows a plugin user to do:

Bar.plugin :foo
Bar.a # => 1

Adding Instance Methods

If you want your plugin to add instance methods to the model class it is loaded into, define an InstanceMethods module under the plugin module:

module Sequel::Plugins::Foo
  module InstanceMethods
    def a
      1
    end
  end
end

This allows a plugin user to do:

Bar.plugin :foo
Bar.new.a # => 1

Adding Dataset Methods

If you want your plugin to add methods to the dataset of the model class it is loaded into, define a DatasetMethods module under the plugin module:

module Sequel::Plugins::Foo
  module DatasetMethods
    def a
      1
    end
  end
end

This allows a plugin user to do:

Bar.plugin :foo
Bar.dataset.a # => 1

Calling super to get Previous Behavior

No matter if you are dealing with class, instance, or dataset methods, you can call super inside the method to get the previous behavior. This makes it easy to hook into the method, add your own behavior, but still get the previous behavior:

module Sequel::Plugins::Foo
  module InstanceMethods
    def save
      if allow_saving?
        super
      else
        raise Sequel::Error, 'saving not allowed for this object'
      end
    end

    private

    def allow_saving?
      moon =~ /Waxing/
    end
  end
end

Running Code When the Plugin is Loaded

Some plugins require more than just adding methods. Any plugin that requires state is going to have to initialize that state and store it somewhere (generally in the model class itself). If you want to run code when a plugin is loaded (usually to initialize state, but possibly for other reasons), there are two methods you can define to do so. The first method is apply, and it is called only the first time the plugin is loaded into the class, before it is loaded into the class. This is generally only used if a plugin depends on another plugin or for initializing state. You define this method as a singleton method of the plugin module:

module Sequel::Plugins::Foo
  def self.apply(model)
    model.instance_eval do
      plugin :plugin_that_foo_depends_on
      @foo_states = {}
    end
  end
end

The other method is called configure, and it is called everytime the plugin is loaded into the class, after it is loaded into the class:

module Sequel::Plugins::Foo
  def self.configure(model)
    model.instance_eval do
      @foo_states[:initial] ||= :baz
    end
  end
end

Note that in the configure method, you know apply has already been called at least once (so @foo_state will definitely exist).

If you want your plugin to take arguments and/or support options, you handle that by making your apply and configure methods take arguments and/or an options hash. For example, if you want the user to be able to set the initial state via an option, you can do:

module Sequel::Plugins::Foo
  def self.apply(model, opts={})
    model.instance_eval do
      plugin :plugin_foo_depends_on
      @foo_states = {}
    end
  end

  def self.configure(model, opts={})
    model.instance_eval do
      @foo_states[:initial] = opts[:initial_state] || @foo_states[:initial] || :baz
    end
  end
end

This allows a user of the plugin to do either of the following

Bar.plugin :foo
Bar.plugin :foo, initial_state: :quux

If you want to require the initial state to be provided as an argument:

module Sequel::Plugins::Foo
  def self.apply(model, initial_state)
    model.instance_eval do
      plugin :plugin_foo_depends_on
      @foo_states = {}
    end
  end

  def self.configure(model, initial_state)
    model.instance_eval do
      @foo_states[:initial] = initial_state
    end
  end
end

This requires that the user of the plugin specify the argument:

Bar.plugin :foo, :quux

In general you should only require plugin arguments if you absolutely must have a value and there is no good default.

Handling Subclasses

Sequel::Model uses a copy-on-subclassing approach to model state. So instead of having a model subclass ask its superclass for a value if the subclass don’t have the value defined, the value should be copied from the parent class to the subclass when the subclass is created. While this can be implemented by overriding the inherited class method, there is an available shortcut that handles most cases:

module Sequel::Plugins::Foo
  module ClassMethods
    Sequel::Plugins.inherited_instance_variables(self, :@foo_states => :dup)
  end
end

Inside the ClassMethods submodule, you call the Sequel::Plugins.inherited_instance_variables method with the first argument being self. The second argument should be a hash describing how to copy the value from the parent class into the subclass. The keys of this hash are instance variable names, including the @ symbol (e.g. :@foo_state). The values of this hash describe how to copy it:

nil

Use the value directly.

:dup

Call dup on the value.

:hash_dup

Create a new hash with the same keys, but a dup of all the values.

Proc

An arbitrary proc that is called with the parent class value and should return the value to set into the subclass.

Handling Changes to the Model’s Dataset

In many plugins, if the model class changes the dataset, you need to change the state for the plugin. While you can do this by overriding the set_dataset class method, there is an available shortcut:

module Sequel::Plugins::Foo
  module ClassMethods
    Sequel::Plugins.after_set_dataset(self, :set_foo_table)

    private

    def set_foo_table
      @foo_states[:table] = table_name
    end
  end
end

With this code, any time the model’s dataset changes, the state of the plugin will be updated to set the correct table name. This is also called when creating a new model class with a dataset.

Making Dataset Methods Callable as Class Methods

In some cases, when dataset methods are added, you want to also create a model class method that will call the dataset method, so you can write:

Model.method

instead of:

Model.dataset.method

There is an available shortcut that automatically creates the class methods:

module Sequel::Plugins::Foo
  module ClassMethods
    Sequel::Plugins.def_dataset_methods(self, :quux)
  end

  module DatasetMethods
    def quux
      2
    end
  end
end