Loading 'Plugins' in Ruby

Question

I have written a plugin loader in Ruby, and I would like to know if it uses the best technique to handle classes, or if anyone has any recommendations on how to improve my code.

I have written a class called PluginManager, and an accompanying module named Plugin. The names of these are self explanatory, one handles the plugins (loading them), and the other is a module to be included by plugin files.

My Plugin module looks like the following, and currently just replaces the puts method, to ensure all output comes through the logger. I had some issues with making the @@logger variable definable, and I have solved it, but I'm not sure if there is a better way to solve it.

module Plugin
    @@logger = nil

    def puts(string)
        if @@logger.nil?
            super(string)
        else
            @@logger.info string
        end
    end

    # I tried the following, but it didn't work (still said "undefined method 'logger='")

    # class << self
    #   attr_accessor :logger
    # end

    def self.logger=(string)
        @@logger = string
    end
end

My PluginManager class is a bit more complex, but not by much. It simply takes an array of file names, performs some checks to ensure everything exists, then loads them into the loaded_plugins array.

class PluginManager

    def initialize(logger=nil)
        set_logger logger
        @loaded_plugins = []
    end

    def set_logger(logger)
        @logger = logger
        Plugin.logger = logger
    end

    def load(plugins, directory="./plugins/")

        # Iterate through plugins
        plugins.each do |plugin|

            # Work out plugin information
            class_name = plugin.split(".").first.downcase.capitalize
            full_path = directory + plugin
            @logger.info "Loading \"#{class_name}\" from \"#{full_path}\""

            # Ensure the plugin file exists
            if File.file? full_path

                # Load the plugin file and ensure the class exists
                require full_path
                unless eval("defined? #{class_name}")
                    @logger.error "Class \"#{class_name}\" not found!"
                    next
                end

                # Get the plugin's class and ensure it includes plugin
                plugin_class = eval class_name
                if plugin_class.ancestors.include? Plugin

                    # Ensure the initialize method takes no arguments
                    if plugin_class.allocate.method(:initialize).arity > 0
                        @logger.error "\"#{class_name}.initialize\" accepts more than 0 arguments"
                        next
                    end

                    # Load the plugin itself by pushing it into the loaded_plugins list
                    @logger.info "Plugin \"#{class_name}\" loaded!"
                    @loaded_plugins.push plugin_class.new
                    @logger.info "Plugin \"#{class_name}\" enabled!"
                else
                    @logger.error "The class \"#{class_name}\" does not include Plugin!"
                    next
                end
            else

                # Check if the file exists, or is a directory
                unless File.exist? full_path
                    @logger.error "Unable to find file \"#{full_path}\""
                else
                    @logger.error "\"#{full_path}\" is a directory!"
                end

                next
            end
        end
    end
end

I am wondering if there are any better ways to perform some of the checks, such as using exec to find if a class exists.

Answer 1

Formatting

Indenting

The standard ruby indent is 2 spaces. For example:

def puts(string)
  if @@logger.nil?
    super(string)
  else
    @@logger.info string
  end
end

Design

Class Variables

I would not use class variables to set the logging. As you discovered, they have surprising behavior.

Instead, I would just use a class instance variable that belonged just to the Plugin class and was not automatically shared with its descendants.

module Plugin
  class << self
    attr_accessor :logger
  end
end

If that's all you do, then the descendants will have to do this to access the logger:

Plugin.logger.info 'Logging some info'

That's awful, so we'll add an instance method that defers to the class method:

module Plugin

  class << self
    attr_accessor :logger
  end

  def logger
    self.class.logger
  end

end

Now instances of Plugin can log easily:

logger.info "Logging some info"

Ruby tricks

Use #inspect to quote strings

Instead of this,

@logger.info "Loading \"#{class_name}\" from \"#{full_path}\""

Use #inspect to do the quoting for you:

@logger.info "Loading #{class_name.inspect} from #{full_path.inspect}"

Remove eval

You don't need eval here:

unless eval("defined? #{class_name}")

instead:

unless defined?(class_name)

ancestors.include?

Instead of this:

if plugin_class.ancestors.include? Plugin

do this:

if plugin_class < Plugin

See this SO answer, and the rdoc for Module#<

instance_method

Instead of this:

if plugin_class.allocate.method(:initialize).arity > 0

this:

if plugin_class.instance_method(:initialize).arity > 0

This avoids the funny business of allocating one.

exceptions

The code wishes, when a plugin cannot be added, to log it and continue adding more plugins. That works well, but the way it's done can be simplified using exceptions.

Instead of this:

plugins.each do |plugin|
  ...
        unless eval("defined? #{class_name}")
            @logger.error "Class \"#{class_name}\" not found!"
            next
        end

  ...
end

consider this:

class PluginError < StandardError ; end


plugins.each do |plugin|
  begin
    ...
    unless eval("defined? #{class_name}")
      raise PluginError, "Class \"#{class_name}\" not found!"
    end
    ...
  rescue PluginError => e
    @logger.error e.to_s
  end
end

This gets rid of all of those "next" calls, simplifying the error handling.