Mocktail 🍸

Mocktail is a test double library for Ruby that provides a terse and robust API for creating mocks, getting them in the hands of the code you're testing, stub & verify behavior, and even safely override class methods.

If you'd prefer a voice & video introduction to Mocktail aside from this README, you might enjoy this ⚡️Lightning Talk⚡️ from RailsConf 2022.

An aperitif

Before getting into the details, let's demonstrate what Mocktail's API looks like. Suppose you want to test a Bartender class:

class Bartender
  def initialize
    @shaker = Shaker.new
    @glass = Glass.new
    @bar = Bar.new
  end

  def make_drink(name, customer:)
    if name == :negroni
      drink = @shaker.combine(:gin, :campari, :sweet_vermouth)
      @glass.pour!(drink)
      @bar.pass(@glass, to: customer)
    end
  end
end

You could write an isolated unit test with Mocktail like this:

shaker = Mocktail.of_next(Shaker)
glass = Mocktail.of_next(Glass)
bar = Mocktail.of_next(Bar)
subject = Bartender.new
stubs { shaker.combine(:gin, :campari, :sweet_vermouth) }.with { :a_drink }
stubs { bar.pass(glass, to: "Eileen") }.with { "🎉" }

result = subject.make_drink(:negroni, customer: "Eileen")

assert_equal "🎉", result
# Oh yeah, and make sure the drink got poured! Silly side effects!
verify { glass.pour!(:a_drink) }

Why Mocktail?

Besides helping you avoid a hangover, Mocktail offers several advantages over Ruby's other mocking libraries:

• Simpler test recipes: Mocktail.of_next(type) both creates your mock and supplies to your subject under test in a single one-liner. No more forcing dependency injection for the sake of your tests
• WYSIWYG API: Want to know how to stub a call to phone.dial(911)? You just demonstrate the call with stubs { phone.dial(911) }.with { :operator }. Because stubbing & verifying looks just like the actual call, your tests becomes a sounding board for your APIs as you invent them
• Argument validation: Ever see a test pass after a change to a mocked method should have broken it? Not with Mocktail, you haven't
• Mocked class methods: Singleton methods on modules and classes can be faked out using Mocktail.replace(type) without sacrificing thread safety
• Super-duper detailed error messages A good mocking library should make coding feel like paint-by-number, thoughtfully guiding you from one step to the next. Calling a method that doesn't exist will print a sample definition you can copy-paste. Verification failures will print every call that did occur. And Mocktail.explain() provides even more introspection
• Expressive: Built-in argument matchers and a simple API for adding custom matchers allow you to tune your stubbing configuration and call verification to match exactly what your test intends
• Powerful: Argument captors for assertions of very complex arguments, as well as advanced configuration options for stubbing & verification

Ready to order?

Install the gem

The main ingredient to add to your Gemfile:

gem "mocktail", group: :test

Sprinkle in the DSL

Then, in each of your tests or in a test helper, you'll probably want to include Mocktail's DSL. (This is optional, however, as every method in the DSL is also available as a singleton method on Mocktail.)

In Minitest, you might add the DSL with:

class Minitest::Test
  include Mocktail::DSL
end

Or, in RSpec:

RSpec.configure do |config|
  config.include Mocktail::DSL
end

Clean up when you're done

To reset Mocktail's internal state between tests and avoid test pollution, you should also call Mocktail.reset after each test:

In Minitest:

class Minitest::Test
  # Or, if in a Rails test, in a `teardown do…end` block
  def teardown
    Mocktail.reset
  end
end

And RSpec:

RSpec.configure do |config|
  config.after(:each) do
    Mocktail.reset
  end
end

API

The entire public API is listed in the top-level module's source. Below is a longer menu to explain what goes into each feature.

Mocktail.of

Mocktail.of(module_or_class) takes a module or class and returns an instance of an object with fake methods in place of all its instance methods which can then be stubbed or verified.

class Clothes; end;
class Shoe < Clothes
  def tie(laces)
  end
end

shoe = Mocktail.of(Shoe)
shoe.instance_of?(Shoe) # => true
shoe.is_a?(Clothes) # => true
shoe.class == Shoe # => false!
shoe.to_s # => #<Mocktail of Shoe:0x00000001343b57b0>"

Mocktail.of_next

Mocktail.of_next(klass, [count: 1]) takes a class and returns one mock (the default) or an array of multiple mocks. It also effectively overrides the behavior of that class's constructor to return those mock(s) in order and finally restoring its previous behavior.

For example, if you wanted to test the Notifier class below:

class Notifier
  def initialize
    @mailer = Mailer.new
  end

  def notify(name)
    @mailer.deliver!("Hello, #{name}")
  end
end

You could write a test like this:

def test_notifier
  mailer = Mocktail.of_next(Mailer)
  subject = Notifier.new

  subject.notify("Pants")

  verify { mailer.deliver!("Hello, Pants") }
end

There's nothing wrong with creating mocks using Mocktail.of and passing them to your subject some other way, but this approach allows you to write very terse isolation tests without foisting additional indirection or dependency injection in for your tests' sake.

Mocktail.stubs

Configuring a fake method to take a certain action or return a particular value is called "stubbing". To stub a call with a value, you can call Mocktail.stubs (or just stubs if you've included Mocktail::DSL) and then specify an effect that will be invoked whenever that call configuration is satisfied using with.

The API is very simple in the simple case:

class UserRepository
  def find(id, debug: false); end

  def transaction(&blk); end
end

You could stub responses to a mock of the UserRepository like this:

user_repository = Mocktail.of(UserRepository)

stubs { user_repository.find(42) }.with { :a_user }
user_repository.find(42) # => :a_user
user_repository.find(43) # => nil
user_repository.find # => ArgumentError: wrong number of arguments (given 0, expected 1)

The block passed to stubs is called the "demonstration", because it represents an example of the kind of calls that Mocktail should match.

If you want to get fancy, you can use matchers to make your demonstration more dynamic. For example, you could match any number with:

stubs { |m| user_repository.find(m.numeric) }.with { :another_user }
user_repository.find(41) # => :another_user
user_repository.find(42) # => :another_user
user_repository.find(43) # => :another_user

Stubbings are last-in-wins, which is why the stubbing above would have overridden the earlier-but-more-specific stubbing of find(42).

A stubbing's effect can also be changed dynamically based on the actual call that satisfied the demonstration by looking at the call block argument:

stubs { |m| user_repository.find(m.is_a(Integer)) }.with { |call|
  {id: call.args.first}
}
user_repository.find(41) # => {id: 41}
# Since 42.5 is a Float, the earlier stubbing will win here:
user_repository.find(42.5) # => :another_user
user_repository.find(43) # => {id: 43}

It's certainly more complex to think through, but if your stubbed method takes a block, your demonstration can pass a block of its own and inspect or invoke it:

stubs {
  user_repository.transaction { |block| block.call == {id: 41} }
}.with { :successful_transaction }

user_repository.transaction {
  user_repository.find(41)
} # => :successful_transaction
user_repository.transaction {
  user_repository.find(40)
} # => nil

There are also several advanced options you can pass to stubs to control its behavior.

times will limit the number of times a satisfied stubbing can have its effect:

stubs { |m| user_repository.find(m.any) }.with { :not_found }
stubs(times: 2) { |m| user_repository.find(1) }.with { :someone }

user_repository.find(1) # => :someone
user_repository.find(1) # => :someone
user_repository.find(1) # => :not_found

ignore_extra_args will allow a demonstration to be considered satisfied even if it fails to specify arguments and keyword arguments made by the actual call:

stubs { user_repository.find(4) }.with { :a_person }
user_repository.find(4, debug: true) # => nil

stubs(ignore_extra_args: true) { user_repository.find(4) }.with { :b_person }
user_repository.find(4, debug: true) # => :b_person

And ignore_block will similarly allow a demonstration to not concern itself with whether an actual call passed the method a block—it's satisfied either way:

stubs { user_repository.transaction }.with { :transaction }
user_repository.transaction {} # => nil

stubs(ignore_block: true) { user_repository.transaction }.with { :transaction }
user_repository.transaction {} # => :transaction

Mocktail.verify

In practice, we've found that we stub far more responses than we explicitly verify a particular call took place. That's because our code normally returns some observable value that is influenced by our dependencies' behavior, so adding additional assertions that they be called would be redundant. That said, for cases where a dependency doesn't return a value but just has a necessary side effect, the verify method exists (and like stubs is included in Mocktail::DSL).

Once you've gotten the hang of stubbing, you'll find that the verify method is intentionally very similar. They almost rhyme.

For this example, consider an Auditor class that our code might need to call to record that certain actions took place.

class Auditor
  def record!(message, user_id:, action: nil); end
end

Once you've created a mock of the Auditor, you can start verifying basic calls:

auditor = Mocktail.of(Auditor)

verify { auditor.record!("hello", user_id: 42) }
# => raised Mocktail::VerificationError
# Expected mocktail of Auditor#record! to be called like:
#
#   record!("hello", user_id: 42)
#
# But it was never called.

Wups! Verify will blow up whenever a matching call hasn't occurred, so it should be called after you've invoked your subject under test along with any other assertions you have.

If we make a call that satisfies the verify call's demonstration, however, you won't see that error:

auditor.record!("hello", user_id: 42)

verify { auditor.record!("hello", user_id: 42) } # => nil

There, nothing happened! Just like any other assertion library, you only hear from verify when verification fails.

Just like with stubs, you can any built-in or custom matchers can serve as garnishes for your demonstration:

auditor.record!("hello", user_id: 42)

verify { |m| auditor.record!(m.is_a(String), user_id: m.numeric) } # => nil
# But this will raise a VerificationError:
verify { |m| auditor.record!(m.is_a(String), user_id: m.that { |arg| arg > 50}) }

When you pass a block to your demonstration, it will be invoked with any block that was passed to the actual call to the mock. Truthy responses will satisfy the verification and falsey ones will fail:

auditor.record!("ok", user_id: 1) { Time.new }

verify { |m| auditor.record!("ok", user_id: 1) { |block| block.call.is_a?(Time) } } # => nil
# But this will raise a VerificationError:
verify { |m| auditor.record!("ok", user_id: 1) { |block| block.call.is_a?(Date) } }

verify supports the same options as stubs:

• times will require the demonstrated call happened exactly times times (by default, the call has to happen 1 or more times)
• ignore_extra_args will allow the demonstration to forego specifying optional arguments while still being considered satisfied
• ignore_block will similarly allow the demonstration to forego specifying a block, even if the actual call receives one

Mocktail.matchers

You'll probably never need to call Mocktail.matchers directly, because it's the object that is passed to every demonstration block passed to stubs and verify. By default, a stubbing (e.g. stubs { email.send("text") }) is only considered satisfied if every argument passed to an actual call was passed an == check. Matchers allow us to relax or change that constraint for both regular arguments and keyword arguments so that our demonstrations can match more kinds of method invocations.

Matchers allow you to specify stubbings and verifications that look like this:

stubs { |m| email.send(m.is_a(String)) }.with { "I'm an email" }

Built-in matchers

These matchers come out of the box:

• any - Will match any value (even nil) in the given argument position or keyword
• is_a(type) - Will match when its type passes an is_a? check against the actual argument
• includes(thing, [**more_things]) - Will match when all of its arguments are contained by the corresponding argument—be it a string, array, hash, or anything that responds to includes?
• matches(pattern) - Will match when the provided string or pattern passes a match? test on the corresponding argument; usually used to match strings that contain a particular substring or pattern, but will work with any argument that responds to match?
• not(thing) - Will only match when its argument does not equal (via !=) the actual argument
• numeric - Will match when the actual argument is an instance of Integer, Float, or (if loaded) BigDecimal
• that { |arg| … } - Takes a block that will receive the actual argument. If the block returns truthy, it's considered a match; otherwise, it's not a match.

Custom matchers

If you want to write your own matchers, check out the source for examples. Once you've implemented a class, just pass it to Mocktail.register_matcher in your test helper.

class MyAwesomeMatcher < Mocktail::Matchers::Base
  def self.matcher_name
    :awesome
  end

  def match?(actual)
    "#{@expected}✨" == actual
  end
end

Mocktail.register_matcher(MyAwesomeMatcher)

Then, a stubbing like this:

stubs { |m| user_repository.find(m.awesome(11)) }.with { :awesome_user }

user_repository.find("11")) # => nil
user_repository.find("11✨")) # => :awesome_user

Mocktail.captor

An argument captor is a special kind of matcher… really, it's a matcher factory. Suppose you have a verify call for which one of the expected arguments is really complicated. Since verify tends to be paired with fire-and-forget APIs that are being invoked for the side effect, this is a pretty common case. You want to be able to effectively snag that value and then run any number of specific assertions against it.

That's what Mocktail.captor is for. It's easiest to make sense of this by example. Given this BigApi class that's presumably being called by your subject at the end of a lot of other work building up a payload:

class BigApi
  def send(payload); end
end

You could capture the value of that payload as part of the verification of the call:

big_api = Mocktail.of(BigApi)

big_api.send({imagine: "that", this: "is", a: "huge", object: "!"})

payload_captor = Mocktail.captor
verify { big_api.send(payload_captor.capture) } # => nil!

The verify above will pass because a call did happen, but we haven't asserted anything beyond that yet. What really happened is that payload_captor.capture actually returned a matcher that will return true for any argument while also sneakily storing a copy of the argument value.

That's why we instantiated payload_captor with Mocktail.captor outside the demonstration block, so we can inspect its value after the verify call:

payload_captor = Mocktail.captor
verify { big_api.send(payload_captor.capture) } # => nil!

payload = payload_captor.value # {:imagine=>"that", :this=>"is", :a=>"huge", :object=>"!"}
assert_equal "huge", payload[:a]

Mocktail.replace

Mocktail was written to support isolated test-driven development, which usually results in a lot of boring classes and instance methods. But sometimes you need to mock singleton methods on classes or modules, and we support that too.

When you call Mocktail.replace(type), all of the singleton methods on the provided type are replaced with fake methods available for stubbing and verification. It's really that simple.

For example, if our Bartender class has a class method:

class Bartender
  def self.cliche_greeting
    ["It's 5 o'clock somewhere!", "Norm!"].sample
  end
end

We can replace the behavior of the overall class, and then stub how we'd like it to respond, in our test:

Mocktail.replace(Bartender)
stubs { Bartender.cliche_greeting }.with { "Norm!" }

[Obligatory warning: Mocktail does its best to ensure that other threads won't be affected when you replace the singleton methods on a type, but your mileage may very! Singleton methods are global and code that introspects or invokes a replaced method in a peculiar-enough way could lead to hard-to-track down bugs. (If this concerns you, then the fact that class methods are effectively global state may be a great reason not to rely too heavily on them!)]

Mocktail.explain

Test debugging is hard enough when there aren't fake objects flying every which way, so Mocktail tries to make it a little easier on you. In addition to returning useful messages throughout the API, the gem also includes an introspection method Mocktail.explain(thing), which returns a human-readable message and a reference object with useful attributes (that vary depending on the type of fake thing you pass in. Below are some things explain() can do.

Fake instances created by Mocktail

Any instances created by Mocktail.of or Mocktail.of_next can be passed to Mocktail.explain, and they will list out all the calls and stubbings made for each of their fake methods.

Suppose these interactions have occurred:

ice_tray = Mocktail.of(IceTray)

Mocktail.stubs { ice_tray.fill(:tap_water, 30) }.with { :some_ice }

ice_tray.fill(:tap_water, 50)

You can interrogate what's going on with the fake instance by passing it to explain:

explanation = Mocktail.explain(ice_tray)

explanation.reference.type      #=> IceTray
explanation.reference.double    #=> The ice_tray instance
explanation.reference.calls     #=> details on each invocation of each method
explanation.reference.stubbings #=> all stubbings configured for each method

Calling explanation.message will return:

This is a fake `IceTray' instance.

It has these mocked methods:
  - fill

`IceTray#fill' stubbings:

  fill(:tap_water, 30)

`IceTray#fill' calls:

  fill(:tap_water, 50)

Modules and classes with singleton methods replaced

If you've called Mocktail.replace() on a class or module, it can also be passed to Mocktail.explain() for a summary of all the stubbing configurations and calls made against its faked singleton methods for the currently running thread.

Imagine a Shop class with self.open! and self.close! singleton methods:

Mocktail.replace(Shop)

stubs { |m| Shop.open!(m.numeric) }.with { :a_bar }

Shop.open!(42)

Shop.close!(42)

explanation = Mocktail.explain(Shop)

explanation.reference.type      #=> Shop
explanation.reference.replaced_method_names #=> [:close!, :open!]
explanation.reference.calls     #=> details on each invocation of each method
explanation.reference.stubbings #=> all stubbings configured for each method

And explanation.message will return:

`Shop' is a class that has had its singleton methods faked.

It has these mocked methods:
  - close!
  - open!

`Shop.close!' has no stubbings.

`Shop.close!' calls:

  close!(42)

  close!(42)

`Shop.open!' stubbings:

  open!(numeric)

  open!(numeric)

`Shop.open!' calls:

  open!(42)

  open!(42)

Methods on faked instances and replaced types

In addition to passing the test double, you can also pass a reference to any fake method created by Mocktail to Mocktail.explain:

ice_tray = Mocktail.of(IceTray)

ice_tray.fill(:chilled, 50)

explanation = Mocktail.explain(ice_tray.method(:fill))

explanation.reference.receiver  #=> a reference to the `ice_tray` instance
explanation.reference.calls     #=> details on each invocation of the method
explanation.reference.stubbings #=> all stubbings configured for the method

The above may be handy in cases where you want to assert the number of calls of a method outside the Mocktail.verify API:

assert_equal 1, explanation.reference.calls.size

The explanation will also contain a message like this:

`IceTray#fill' has no stubbings.

`IceTray#fill' calls:

  fill(:chilled, 50)

Replaced singleton methods can also be passed to explain(), so something like Mocktail.explain(Shop.method(:open!)) from the earlier example would also work (with Shop being the receiver on the explanation's reference).

Undefined methods

There's no API for this one, but Mocktail also offers explanations for methods that don't exist yet. You'll see this error message whenever you try to call a method that doesn't exist on a test double. The message is designed to facilitate "paint-by-numbers" TDD, by including a sample definition of the method you had attempted to call that can be copy-pasted into a source listing:

class IceTray
end

ice_tray = Mocktail.of(IceTray)

ice_tray.fill(:water_type, 30)
# => No method `IceTray#fill' exists for call: (NoMethodError)
#
#      fill(:water_type, 30)
#
#    Need to define the method? Here's a sample definition:
#
#      def fill(water_type, arg)
#      end

From there, you can just copy-paste the provided method stub as a starting point for your new method:

class IceTray
  def fill(water_type, amount)
  end
end

Mocktail.explain_nils

Is a faked method returning nil and you don't understand why?

By default, methods faked by Mocktail will return nil when no stubbing is satisfied. A frequent frustration, therefore, is when the way stubs {}.with {} is configured does not satisfy a call the way you expected. To try to make debugging this a little bit easier, the gem provides a top-level Mocktail.explain_nils method that will return an array of summaries of every call to a faked method that failed to satisfy any stubbings.

For example, suppose you stub this fill method like so:

ice_tray = Mocktail.of(IceTray)

stubs { ice_tray.fill(:tap_water, 30) }.with { :normal_ice }

But then you find that your subject under test is just getting nil back and you don't understand why:

def prep
  ice = ice_tray.fill(:tap_water, 50)
  glass.add(ice) # => why is `ice` nil?!
end

Whenever you're confused by a nil, you can call Mocktail.explain_nils for an array containing UnsatisfyingCallExplanation objects (one for each call to a faked method that did not satisfy any configured stubbings).

The returned explanation objects will include both a reference object to explore as well a summary message:

def prep
  ice = ice_tray.fill(:tap_water, 50)
  puts Mocktail.explain_nils.first.message
  glass.add(ice)
end

Which will print:

This `nil' was returned by a mocked `IceTray#fill' method
because none of its configured stubbings were satisfied.

The actual call:

  fill(:tap_water, 50)

The call site:

  /path/to/your/code.rb:42:in `prep'

Stubbings configured prior to this call but not satisfied by it:

  fill(:tap_water, 30)

The reference object will have details of the call itself, an array of other_stubbings defined on the faked method, and a backtrace to determine which call site produced the unexpected nil value.

Mocktail.reset

This one's simple: you probably want to call Mocktail.reset after each test, but you definitely want to call it if you're using Mocktail.replace or Mocktail.of_next anywhere, since those will affect state that is shared across tests.

Calling reset in a teardown or after(:each) hook will also improve the usefulness of messages returned by Mocktail.explain and Mocktail.explain_nils.

References

Mocktail is designed following a somewhat academic understanding of what mocking is and how it should be used. Below are several references on this topic.

Blog Posts and Papers:

• [Endo-Testing: Unit Testing with Mock Objects](https://www2.ccs.neu.edu/research/demeter/related-work/extreme-programming/MockObjectsFinal.PDF by Tim Mackinnon, Steve Freeman, and Philip Craig, the paper that introduced mocking presented by the creators of mocking.
• Michael Feathers' The Flawed Theory Behind Unit Testing

Books:

• Growing Object-Oriented Software, Guided by Tests by Steve Freeman and Nat Price

Talks:

• Please don’t mock me by Justin Searls

Acknowledgements

Mocktail is created & maintained by the software agency Test Double. If you've ever come across our eponymously-named testdouble.js, you might find Mocktail's API to be quite similar. The term "test double" was originally coined by Gerard Meszaros in his book xUnit Test Patterns.

The name is inspired by the innovative Java mocking library Mockito. Mocktail also the spiritual successor to gimme, which offers a similar API but which fell victim to the limitations of Ruby 1.8.7 (and @searls's Ruby chops). Gimme was also one of the final projects we collaborated with Jim Weirich on, so this approach to isolated unit testing holds a special significance to us.

Code of Conduct

This project follows Test Double's code of conduct for all community interactions, including (but not limited to) one-on-one communications, public posts/comments, code reviews, pull requests, and GitHub issues. If violations occur, Test Double will take any action they deem appropriate for the infraction, up to and including blocking a user from the organization's repositories.