Ruby Gem Configuration Patterns

This is a great approach for setting config variables. I also like to enable the user to add custom functionality to the gem by accepting a block that is called with any relevant variables for the user to access.

For example, in Exceptionally[1], the user can add an initializer that looks like:

    Exceptionally::Handler.before_render do |message, status, error, params|
      # put any code here
    end

In the gem, this translates to:

    def self.before_render(&block)
      &&callback = Proc.new(&block)
    end

The gem can then call the user's block and give it access to the same variables the rest of the gem does. This approach is a lot simpler than setting up an entire Middleware stack, as suggested in the blog, and makes more sense for simpler gems.

1: https://github.com/neilgupta/exceptionally

Re: Ruby Gem Configuration Patterns

#5

Post by feca » Wed, Mar 19, 2014, 3:08 PM UTC

This is actually a very complicated way of achieving the goal. It works, but it's not good from an economic point of view as it is more expensive in computational terms than a simpler OO alternative, which is to pass the size when you instantiate the MegaLotto::Drawing object.

For example, compare the proposed solution:

    MegaLotto.configure do |config|
      config.drawing_count = 10
    end

    MegaLotto::Drawing.new.draw

With the alternative:

    MegaLotto::Drawing.new(10).draw

If you want to make it extensible, you can use keyword arguments:

    MegaLotto::Drawing.new(size: 10).draw

The interface and the implementation are simpler, but also the performance is better because there are less method calls. If you look at the code of both implementations, you will find the simpler one easier to understand. As a side effect, you will also get simpler stack traces if anything goes wrong.

Re: Ruby Gem Configuration Patterns

#6

Post by danso » Wed, Mar 19, 2014, 3:17 PM UTC

feca wrote: post #5

This is actually a very complicated way of achieving the goal. It works, but it's not good from an economic point of view as it is more expensive in computational terms than a simpler OO alternative, which is to pass the size when you instantiate the MegaLotto::Drawing object. For example, compare the proposed solution: MegaLotto.configure do |config| config.drawing_count = 10 end MegaLotto::Drawing.new.draw With the…

Right...but even in this trivial gem, there may be the need to configure more than one parameter. And by having Configuration be it's own object, you can encode some validation logic at the configuration stage.

I do agree that the constructor should have the option of passing in a Hash, which is then passed directly to the Configuration option.

Re: Ruby Gem Configuration Patterns

#7

Post by route66 » Wed, Mar 19, 2014, 3:25 PM UTC

feca wrote: post #5

This is actually a very complicated way of achieving the goal. It works, but it's not good from an economic point of view as it is more expensive in computational terms than a simpler OO alternative, which is to pass the size when you instantiate the MegaLotto::Drawing object. For example, compare the proposed solution: MegaLotto.configure do |config| config.drawing_count = 10 end MegaLotto::Drawing.new.draw With the…

In this case you are probably right. The solution with separate configuration shines though, when you have many values to configure (notwithstanding keyword args). But, and that is my main takeaway, TDD does apparently (and not surprisingly) not show the way to one or the other solution. On the contrary: the solution was already given in advance and the TDD part was more or less toying around with the moving parts of the solution.

Re: Ruby Gem Configuration Patterns

#8

Post by feca » Wed, Mar 19, 2014, 3:33 PM UTC

danso wrote: post #6

feca wrote: post #5

This is actually a very complicated way of achieving the goal. It works, but it's not good from an economic point of view as it is more expensive in computational terms than a simpler OO alternative, which is to pass the size when you instantiate the MegaLotto::Drawing object. For example, compare the proposed solution: MegaLotto.configure do |config| config.drawing_count = 10 end MegaLotto::Drawing.new.draw With the…

Right...but even in this trivial gem, there may be the need to configure more than one parameter. And by having Configuration be it's own object, you can encode some validation logic at the configuration stage. I do agree that the constructor should have the option of passing in a Hash, which is then passed directly to the Configuration option.

Another problem with the approach from the blog post is locality. If you need to draw 10 numbers in all places, then configuring that value in the module will work. But localy, when you look at the code, it won't tell you how many numbers you are drawing.

If you have this in many different places:

    MegaLotto::Drawing.new.draw

You don't have the information of how many numbers you are getting back. That's not a big deal, but adds to the cognitive load (or requires some comments). Also, if you need to draw different numbers in several different places, you will have to change the configuration many times:

    # First use, we need 10 numbers
    MegaLotto.configure do |config|
      config.drawing_count = 10
    end

    MegaLotto::Drawing.new.draw

    # Second use, we need 6 numbers
    MegaLotto.configure do |config|
      config.drawing_count = 6
    end

    MegaLotto::Drawing.new.draw

And as soon as you do that, you may need to take multi-threading into account, because you are mutating the class.

Extrapolating, it is like defining the size of an array:

    Array.new(4)

If instead you configure Array.new to have a given size for all instantiations, you also lose locality and you may run into thread safety issues.

In the case of MegaLotto::Drawing.new needing multiple configuration options, you can use keyword arguments. If you need too many arguments, maybe the abstraction is wrong. Even if you want to move forward with too many arguments, you can add getters/setters to the newly created instance:

   # Another approach which modifies the instance
   drawing = MegaLotto::Drawing.new
   drawing.size = 10
   drawing.draw #=> returns ten numbers

But this is not optimal design given the elements we have.

Re: Ruby Gem Configuration Patterns

#9

Post by riffraff » Wed, Mar 19, 2014, 3:59 PM UTC

I wish people would stop doing this.

It generally means down the road I find myself wanting to use two different configurations and everything explodes in my face because the author of the gem thought there would always be exactly one configuration.

What is wrong with

     MegaLotto.new do |lotto|
       lotto.drawing_count = 10
     end

and just wrapping it in a "default_megalotto" factory method?

Re: Ruby Gem Configuration Patterns

#10

Post by danso » Wed, Mar 19, 2014, 4:23 PM UTC

feca wrote: post #8

danso wrote: post #6

Earlier quoted context omitted.

Right...but even in this trivial gem, there may be the need to configure more than one parameter. And by having Configuration be it's own object, you can encode some validation logic at the configuration stage. I do agree that the constructor should have the option of passing in a Hash, which is then passed directly to the Configuration option.

Another problem with the approach from the blog post is locality. If you need to draw 10 numbers in all places, then configuring that value in the module will work. But localy, when you look at the code, it won't tell you how many numbers you are drawing. If you have this in many different places: MegaLotto::Drawing.new.draw You don't have the information of how many numbers you are getting back. That's not a big dea…

Ah yes, mutating the class is most definitely a concern...which is why I give you an additional +1 for advocating for a Hash to be passed into the constructor.

This seems to be what the Twitter gem did too, in its newest versions. AFAIK, the Twitter configuration was a mutation to the class via config object, and now it's been revised to be thread safe:

https://github.com/sferik/twitter#configuration

    client = Twitter::REST::Client.new do |config|
      config.consumer_key        = "YOUR_CONSUMER_KEY"
      config.consumer_secret     = "YOUR_CONSUMER_SECRET"
      config.access_token        = "YOUR_ACCESS_TOKEN"
      config.access_token_secret = "YOUR_ACCESS_SECRET"
    end