Skip to content

Latest commit

 

History

History
768 lines (581 loc) · 23.1 KB

code-organisation.md

File metadata and controls

768 lines (581 loc) · 23.1 KB

Code organisation + require in Ruby

This guide will cover:

  • How Ruby projects are organised
  • How to build gems from scratch
  • How to make one gem depend on another gem
  • How require works to load gems installed on your system

Requiring, relatively

In Ruby (and most other languages) you can choose to write your entire program inside one big file. Normally though, Ruby programs are split across multiple files. What we'll look at in this section is how one file can load code that is included in a separate file. This feature is a major underpinning of how Ruby projects are organised.

We'll start with a Ruby script that exists in a single file:

01-basic-project/project.rb

class Response
  attr_reader :answers

  def initialize(user:)
    @user = user
    @answers = []
  end
end

class Answer
  def initialize(rating:)
    @rating = rating
  end
end

class User
  def initialize(name:)
    @name = name
  end
end

user = User.new(name: "Ryan")
response = Response.new(user: user)

question_1 = "Do you feel like the work you do is meaningful?"

print "#{question_1} [1-5]"
rating = gets

response.answers << Answer.new(rating: rating.to_i)

question_2 = "Do you feel supported by your manager?"

print "#{question_2} [1-5]"
rating = gets

response.answers << Answer.new(rating: rating.to_i)

At about 40 lines long, this script fits very easily into most editor windows. It's just getting to the point where it's hard to see it all at once, but it's not quite there yet.

But imagine a situation where this code keeps growing, and more classes or features get added to this file. It would quickly become hard to work in it, because you would be moving up-and-down the file a lot, jumping to wherever you needed to be. For instance, if you wanted to add a method to the Answer class in this script, you would need to look for class Answer and then add the method inside that.

A better way to approach this problem would be to split the different classes and responsibilities of this script into separate files:

02-organised-project

.
├── answer.rb
├── cli.rb
├── response.rb
└── user.rb

Each of these files is then very short and it makes it easier to jump around the project. If we want to add a new method to the Answer class now, we know that the class is defined within answer.rb, and so we can jump there quickly using our editor's shortcuts.

In answer.rb, the code would be:

class Answer
  def initialize(rating:)
    @rating = rating
  end
end

In response.rb, the code would be:

class Response
  attr_reader :answers

  def initialize(user:)
    @user = user
    @answers = []
  end
end

And in user.rb, the code would be:

class User
  def initialize(name:)
    @name = name
  end
end

Then in cli.rb the code would use all of these classes:

user = User.new(name: "Ryan")
response = Response.new(user: user)

question_1 = "Do you feel like the work you do is meaningful?"

print "#{question_1} [1-5]"
rating = gets

response.answers << Answer.new(rating: rating.to_i)

question_2 = "Do you feel supported by your manager?"

print "#{question_2} [1-5]"
rating = gets

response.answers << Answer.new(rating: rating.to_i)

If we try to run cli.rb at the moment, it will fail because it doesn't know where to find the User class:

cli.rb:1:in `<main>': uninitialized constant User (NameError)

Before the code in cli.rb can run, we need to tell it where to find the User class, and probably also the Response and Answer classes. We know that these classes are in user.rb, response.rb and answer.rb respectively, but how can we tell cli.rb that? The answer is with require_relative:

require_relative 'answer'
require_relative 'response'
require_relative 'user'

user = User.new(name: "Ryan")
response = Response.new(user: user)

question = "Do you feel like the work you do is meaningful?"

puts "#{question} [1-5]"
rating = gets

response.answers << Answer.new(rating: rating.to_i)

The require_relative works here by looking at files located relative to the current file; files that are in the same directory.

Now when cli.rb runs, it will first load the answer.rb, evaluate that file, and then do the same with response.rb and user.rb. Evaluating these files will load in the Answer, Response and User classes that our cli.rb script needs to do its job.

Running ruby cli.rb will now run through our script with no problems:

Do you feel like the work you do is meaningful? [1-5] 5
Do you feel supported by your manager? [1-5] 5

Separating out the code into separate files makes it much easier to jump through the different files in our project. It's for this reason that it is best practice in Ruby projects to follow the one-class-per-file rule.

Introducing cross-project dependencies

Now let's say that we had another project that wanted to make use of these classes, and this project was called "comments". The comments project's code lives in another directory. Here's what the directory structure of the two projects might look like:

├── core
│   ├── answer.rb
│   ├── cli.rb
│   ├── response.rb
│   └── user.rb
└── comments
    └── comment.rb

In that comments/comment.rb file, we might want to make use of the User class also, to tie together a Comment and a User, for instance. An example of this would be code like this inside of comment.rb:

class Comment
  def initialize(text, author:)
    @text = text
    @author = User.new(name: author)
  end
end

(Ignoring the fact here that User.new should probably not be the responsibility of the Comment class, but rather the responsibility of whatever calls Comment.new...)

So in comment.rb, we could write this to require the user.rb file from the core project:

require_relative '../core/user'

The .. here means "in one directory 'up' from the current location". This will make require_relative start looking relative to this current file's location. It will look in the parent directory that core and comments reside in together, and then look in the core subdirectory to find the user.rb file.

This will work, but then this means that the comments and core projects must be located in the same directory, together. If either comments or core gets moved, then the whole thing falls apart. This won't do! There has to be a better way.

Depending on a gem from within another gem

Indeed, there is a better way. That better way is to separate these directories into gems, and then to very clearly define a dependency between the two gems. This is a common approach within the Ruby community. You have probably already a gem or two.

In the previous example, comments's code depends on features from the core codebase, and so there should be a dependency introduced between comments and core: comments depends on core.

Let's move our code into a gem structure now. We can generate two new gems by running the bundle gem command twice:

bundle gem jep_core
bundle gem jep_comments

Let's look at the structure of the jep_comments gem:

.
├── Gemfile
├── README.md
├── Rakefile
├── bin
│   ├── console
│   └── setup
├── jep_comments.gemspec
├── lib
│   ├── jep_comments
│   │   └── version.rb
│   └── jep_comments.rb
└── spec
    ├── jep_comments_spec.rb
    └── spec_helper.rb

The jep_core gem is almost identical, but just where it would say jep_comments in the above example it would say jep_core instead.

At the top-level of this gem, we have a file called jep_comments.gemspec. This file is the gem specification and this file's job is to contain data about what your gem is named, a description of its functionality, who wrote it, and most importantly: this gem's dependencies.

Near the bottom of this file, we can see jep_comments gem's dependencies listed:

spec.add_development_dependency "bundler", "~> 1.16"
spec.add_development_dependency "rake", "~> 10.0"
spec.add_development_dependency "rspec", "~> 3.0"

These three lines say that when we're developing this gem, we require three other gems to make our gem development life easy: the bundler, rake and rspec gems.

The second argument to add_development_dependency is the required version number of this package. The little ~> is called a "twiddle-wakka" or "tilde-wakka" and is explained here.

To make the jep_comments gem depend on the jep_core gem, we need to list it as a regular dependency, which we can do with this line, underneath the development dependencies:

spec.add_dependency "jep_core", "~> 0.1.0"

By adding a dependency here, we're now stating that in order for jep_comments to work at all, it needs jep_core. Otherwise, what's the point? The gemspec says that jep_comments has a dependency on jep_core, and so it probably wouldn't work without jep_core.

For us to be able to use the jep_comments gem, we'll need to first install the jep_core gem. Let's look at how we can install the jep_core gem, and then look at how we can install the jep_comments gem.

Installing the jep_core gem

Normally, we wouldn't have to manually install jep_core because the jep_core gem would be available on RubyGems, and by installing jep_comments it would automatically install jep_core too, because we list it as a dependency in jep_comments's gemspec. You might've noticed this happening when you install the rails gem: it doesn't just install the rails gem but it also installs all the dependencies needed for rails too.

Here's the beginning of the output of gem install rails as an example:

Fetching: concurrent-ruby-1.0.5.gem (100%)
Successfully installed concurrent-ruby-1.0.5
Fetching: i18n-0.9.1.gem (100%)
Successfully installed i18n-0.9.1
Fetching: thread_safe-0.3.6.gem (100%)
Successfully installed thread_safe-0.3.6
Fetching: tzinfo-1.2.4.gem (100%)
Successfully installed tzinfo-1.2.4
Fetching: activesupport-5.1.4.gem (100%)
...

We didn't ask for these gems to be installed, but they are anyway because they're listed as dependencies of the rails gem, or dependencies of those dependencies.

Our life isn't so easy for the jep_core and jep_comments gem, because these two gems do not exist on RubyGems.org yet. So we must install them manually.

To install the jep_core gem, we need to first build it, which we can do with the gem build command.

We'll need to be inside the jep_core directory to run this command:

gem build jep_core.gemspec

Unfortunately, we'll see an error the first time we run this command:

ERROR:  While executing gem ... (Gem::InvalidSpecificationException)
    "FIXME" or "TODO" is not a description

This error is happening because the description of our gem inside the gemspec is still the default description.

Before we can run gem build successfully, we will need to change a few lines in our jep_core.gemspec file:

spec.summary       = %q{TODO: Write a short summary, because RubyGems requires one.}
spec.description   = %q{TODO: Write a longer description or delete this line.}

Let's change these lines now to just the summary line:

spec.summary       = %q{The core part for the JEP}

This removes the word "TODO" from the description and summary, which should make that error go away. Let's try running gem build again:

gem build jep_core.gemspec

When we run this command, we'll see this output:

WARNING:  licenses is empty, but is recommended.  Use a license identifier from
http://spdx.org/licenses or 'Nonstandard' for a nonstandard license.
  Successfully built RubyGem
  Name: jep_core
  Version: 0.1.0
  File: jep_core-0.1.0.gem

The first line indicates a warning: the gem build command expects our gemspec to include license information, but it doesn't. We can make this warning go away with a small addition to our gemspec, under the summary:

spec.license = "MIT"

We can run gem build jep_core.gemspec again to verify that this warning has gone away.

Now to the more important part of the output of gem build:

Successfully built RubyGem
Name: jep_core
Version: 0.1.0
File: jep_core-0.1.0.gem

The gem build command has read our gemspec and has built a brand new gem for us. This file contains all the code necessary for your gem, namely everything but the spec directory. Users of the gem don't need the tests when using this gem; only the developers of the gem need it. What's included in the gem file is determined by the spec.files = ... code in the gem's gemspec.

Now that we've built the gem, we can try to install it with gem install:

gem install jep_core-0.1.0.gem

You might've installed a gem before by just using its name, using a command like gem install jep_core or gem install rails. This would make RubyGems look for that gem on rubygems.org. What we're doing here is different: we're specifying a local file to use for our gem install process. The gem install process will only use this local file to install our gem.

When we run this command, we'll see that it near-instantaneously installs our gem:

Successfully installed jep_core-0.1.0
1 gem installed

It's here that I should mention that there's a shortcut for gem build and gem install, and that is rake install. This will build a new gem, putting it inside the pkg directory, and then install it, all in one easy step! Let's try it now:

jep_core 0.1.0 built to pkg/jep_core-0.1.0.gem.
jep_core (0.1.0) installed.

Hooray! We've got one of our two gems installed. We'll use rake install in the future instead of gem build and gem install. Now let's look at installing the jep_comments gem.

Installing the jep_comments gem

Installing the jep_comments gem is a matter of following the same steps that we ran through for the jep_core gem:

  1. Fix up the summary in jep_comments.gemspec
  2. Add license information to jep_comments.gemspec.
  3. Run rake install

When we run that rake install command, we'll see this:

jep_comments 0.1.0 built to pkg/jep_comments-0.1.0.gem.
jep_comments (0.1.0) installed.

Hooray on this one too! We've now got both of our gems installed. However, these gems are both still just skeletons. We will need to move our code in from our old projects to make them actually do something useful.

A gem's file structure

Inside both the jep_core and jep_comments gems, there is a lib directory. This directory is where the code for our gems will live. Here's what the jep_core/lib directory looks like:

├── jep_core
│   └── version.rb
└── jep_core.rb

And here's what the jep_comments/lib directory looks like:

├── jep_comments
│   └── version.rb
└── jep_comment.rb

These are mostly the same, just the names are different. Let's take the files from our earlier examples and put them into our gems' lib directory. We'll put answer.rb response.rb and user.rb inside of jep_core/lib/jep_core, and we'll put comment.rb inside of jep_comments/lib/jep_comments.

jep_core/lib/jep_core/answer.rb

module JepCore
  class Answer
    def initialize(rating:)
      @rating = rating
    end
  end
end

jep_core/lib/jep_core/response.rb

class Response
  attr_reader :answers

  def initialize(user:)
    @user = user
    @answers = []
  end
end

jep_core/lib/jep_core/user.rb

module JepCore
  class User
    def initialize(name:)
      @name = name
    end
  end
end

jep_comments/lib/jep_comments/comment.rb

require_relative '../core/user'

module JepComment
  class Comment
    def initialize(text, author:)
      @text = text
      @author = JepCore::User.new(name: author)
    end
  end

There's a small thing we'll need to change in this comment.rb file, and that's the require_relative at the top. We'll need to change it to this:

require "jep_core/user"

This is because we now want to refer to the user.rb file from within the jep_core gem. But how does Ruby know where to find this file? The answer lies in how we've structured our gems, and how require works.

Requiring files and the load path

We've put these files at lib/jep_core and lib/jep_comments, rather than just under lib in both projects for a very good reason. This is so that the files do not conflict with each other. Why would the files conflict? I'm glad you asked!

When you're using gems in Ruby, their lib directories get added to a list of directories called the load path. The load path is where Ruby goes looking for files to require them. So in the case of our jep_core and jep_comments gems, both of their lib directories would be on the load path.

If we had a file in jep_core/lib that was called user.rb and one also in jep_comments/lib that was also called user.rb, and then in our project we tried doing this:

require "user"

How would Ruby know that we wanted the one from jep_core and not jep_comments or vice-versa? By putting these files inside a sub-directory of lib, we avoid this confusion:

# "give me the user.rb" file from jep_core"
require "jep_core/user"
# "give me the user.rb" file from jep_comments"
require "jep_comments/user"

How does Ruby know where to find these files in the first place after we've installed the gem? Well, an easy way to see this in action would be to try it out in irb. We've got both of our gems installed now, so this next part should be a cinch.

Let's open up irb and try working with these gems. It's not important where you open irb; it will work anywhere on your machine. This is because of when you install gems, they are made globally available by default.

We can require the jep_core and jep_comments gem like this:

irb(main):001:0> require "jep_core"
true
irb(main):002:0> require "jep_comments"
true

(NOTE: I have a technical explanation for how require finds the gem's file on my blog)

The true return value here tells us that Ruby has successfully found these files and is requiring them for the first time. If Ruby had required them before-hand, we would see false here.

But we can't use those gems' classes yet:

irb(main):003:0> JepCore::User.new(name: "Ryan") 
JepCore::User.new(name: "Ryan")
NameError: uninitialized constant JepCore::User
	from (irb):3
	from /Users/ryanbigg/.rubies/ruby-2.4.1/bin/irb:11:in `<main>'

Huh? Didn't we require the gem? Shouldn't it just know about our gem's classes?

Yes to the first question, no to the second. Requiring a gem doesn't magically load all that gem's classes. What that does is require the jep_core.rb file at the root of our project, which just has this content in it:

require "jep_core/version"

module JepCore
  # Your code goes here...
end

When we do require 'jep_core', it just loads / requires this one file. There is nothing in our code that tells it where to find the JepCore::User class at the moment, and so that class is uninitialized.

This jep_core.rb file should be considered as the "entrypoint" to this project. Whatever should be made available from this gem should be required here, in this file. This is so that we can simply do require "jep_core" and then the whole project is loaded.

So at the top of this file, let's put a few more require calls so that our User and Response classes are loaded:

require "jep_core/answer"
require "jep_core/response"
require "jep_core/user"
require "jep_core/version"

module JepCore
  # Your code goes here...
end

That's better! When this gem is loaded, it will now also load the jep_core/response.rb and jep_core/user.rb files too. We could use require_relative here too:

require_relative "jep_core/answer"
require_relative "jep_core/response"
require_relative "jep_core/user"
require_relative "jep_core/version"

This is because the jep_core.rb file lives inside lib, and jep_core/answer.rb and friends are located relative to the jep_core.rb file. But this is more typing than is necessary, so we'll stick with require.

require "jep_core/answer"
require "jep_core/response"
require "jep_core/user"
require "jep_core/version"

Going back to irb and running through the same steps as before will not get us any further:

irb(main):001:0> require 'jep_core'
irb(main):002:0> JepCore::User.new(name: "Ryan") 
NameError: uninitialized constant JepCore::User
	from (irb):2
	from /Users/ryanbigg/.rubies/ruby-2.4.1/bin/irb:11:in `<main>'

This is because Ruby is still loading the old version of our gem. We will need to install the new version of our gem to make Ruby load the new code. Let's do this now by going into the jep_core gem and re-running this command:

rake install

With the gem newly rebuilt + installed, our code should now work:

irb(main):001:0> require "jep_core"
=> true
irb(main):002:0> JepCore::User.new(name: "Ryan") 
=> #<JepCore::User:0x007fb3390d20c0 @name="Ryan">

This is a good start. We'll see the same kind of problem with our jep_comments gem if we try to use a class from there:

irb(main):003:0> JepComments::Comment.new("A new comment!", author: "Ryan")
NameError: uninitialized constant JepComments::Comment

Let's fix this now. We'll go into the jep_comments gem and open up the lib/jep_comments.rb file, and then we'll add a require for the jep_comments/comment.rb file:

require "jep_comments/comment"

Because we've made changes to the jep_comments gem, we will need to re-install it too. Let's run this command inside the jep_comments gem to do that:

rake install

We'll then go back into our irb prompt and try using the JepComments::Comment class again:

irb(main):001:0> require "jep_comments"
=> true
irb(main):002:0> JepComments::Comment.new("A new comment!", author: "me@ryanbigg.com")
=> #<JepComments::Comment:0x007ff95810b5b8 @text="A new comment!", @author=#<JepCore::User:0x007ff95810b540 @name="me@ryanbigg.com">>

This line creates a new instance of the JepComments::Comment class, which links to a JepCore::User class... but we didn't require jep_core in irb yet. What we're doing here is possible because of the require inside of jep_comments/lib/jep_comments/comment.rb:

require "jep_core/user"

This line tells Ruby that jep_comments/lib/jep_comments/comment.rb requires jep_core/user. Ruby will dutifully load that file, and then continue with executing the remainder of the file once jep_core/user has been loaded.

Conclusion

In this guide you've seen how to share code across different projects, by introducing dependencies between them, listing them in the gemspec and by cross-requiring files too. This guide has given you insight into how require works, and also how to build gems from scratch.

Homework

  • The cli.rb file was left out of the jep_core gem. Where is the right place to put this file? How could we make it so that someone who had the jep_core gem installed could run the command run_survey and it would prompt them for the questions?