Awesome

Kleisli

An idiomatic, clean implementation of a few common useful monads in Ruby, written by Ryan Levick and me.

It aims to be idiomatic Ruby to use in Enter-Prise production apps, not a proof of concept.

In your Gemfile:

gem 'kleisli'

We would like to thank Curry and Howard for their correspondence.

Notation

For all its monads, Kleisli implements return (we call it lift instead, as return is a reserved keyword in Ruby) with convenience global methods (see which for each monad below).

Kleisli uses a clever Ruby syntax trick to implement the bind operator, which looks like this: >-> when used with a block. We will probably burn in hell for this. You can also use > or >> if you're going to pass in a proc or lambda object.

Maybe and Either are applicative functors with the apply operator *. Read further to see how it works.

Function composition

You can use Haskell-like function composition with F and the familiar .. This is such a perversion of Ruby syntax that Matz would probably condemn this:

Think of F as the identity function. Although it's just a hack to make it work in Ruby.

# Reminder that (f . g) x= f(g(x))
f = F . first . last
f.call [[1,2], [3,4]]
# => 3

f = F . capitalize . reverse
f.call "hello"
# => "Olleh"

Functions and methods are interchangeable:

foo = lambda { |s| s.reverse }

f = F . capitalize . fn(&foo)
f.call "hello"
# => "Olleh"

All functions and methods are partially applicable:

# Partially applied method:
f = F . split(":") . strip
f.call "  localhost:9092     "
# => ["localhost", "9092"]

# Partially applied lambda:
my_split = lambda { |str, *args| str.split(*args) }
f = F . fn(":", &my_split) . strip
f.call "  localhost:9092     "
# => ["localhost", "9092"]

Finally, for convenience, F is the identity function:

F.call(1) # => 1

Maybe monad

The Maybe monad is useful to express a pipeline of computations that might return nil at any point. user.address.street anyone?

>-> (bind)

require "kleisli"

maybe_user = Maybe(user) >-> user {
  Maybe(user.address) } >-> address {
    Maybe(address.street) }

# If user exists
# => Some("Monad Street")
# If user is nil
# => None()

# You can also use Some and None as type constructors yourself.
x = Some(10)
y = None()

As usual (with Maybe and Either), using point-free style is much cleaner:

Maybe(user) >> F . fn(&Maybe) . address >> F . fn(&Maybe) . street

fmap

require "kleisli"

# If we know that a user always has an address with a street
Maybe(user).fmap(&:address).fmap(&:street)

# If the user exists
# => Some("Monad Street")
# If the user is nil
# => None()

* (applicative functor's apply)

require "kleisli"

add = -> x, y { x + y }
Some(add) * Some(10) * Some(2)
# => Some(12)
Some(add) * None() * Some(2)
# => None

Try

The Try monad is useful to express a pipeline of computations that might throw an exception at any point.

>-> (bind)

require "kleisli"

json_string = get_json_from_somewhere

result = Try { JSON.parse(json_string) } >-> json {
  Try { json["dividend"].to_i / json["divisor"].to_i }
}

# If no exception was thrown:

result       # => #<Try::Success @value=123>
result.value # => 123

# If there was a ZeroDivisionError exception for example:

result           # => #<Try::Failure @exception=#<ZeroDivisionError ...>>
result.exception # => #<ZeroDivisionError ...>

fmap

require "kleisli"

Try { JSON.parse(json_string) }.fmap(&:symbolize_keys).value

# If everything went well:
# => { :my => "json", :with => "symbolized keys" }
# If an exception was thrown:
# => nil

to_maybe

Sometimes it's useful to interleave both Try and Maybe. To convert a Try into a Maybe you can use to_maybe:

require "kleisli"

Try { JSON.parse(json_string) }.fmap(&:symbolize_keys).to_maybe

# If everything went well:
# => Some({ :my => "json", :with => "symbolized keys" })
# If an exception was thrown:
# => None()

to_either

Sometimes it's useful to interleave both Try and Either. To convert a Try into a Either you can use to_either:

require "kleisli"

Try { JSON.parse(json_string) }.fmap(&:symbolize_keys).to_either

# If everything went well:
# => Right({ :my => "json", :with => "symbolized keys" })
# If an exception was thrown:
# => Left(#<JSON::ParserError: 757: unexpected token at 'json'>)

Either

The Either monad is useful to express a pipeline of computations that might return an error object with some information.

It has two type constructors: Right and Left. As a useful mnemonic, Right is for when everything went "right" and Left is used for errors.

Think of it as exceptions without messing with the call stack.

>-> (bind)

require "kleisli"

result = Right(3) >-> value {
  if value > 1
    Right(value + 3)
  else
    Left("value was less or equal than 1")
  end
} >-> value {
  if value % 2 == 0
    Right(value * 2)
  else
    Left("value was not even")
  end
}

# If everything went well
result # => Right(12)
result.value # => 12

# If it failed in the first block
result # => Left("value was less or equal than 1")
result.value # => "value was less or equal than 1"

# If it failed in the second block
result # => Left("value was not even")
result.value # => "value was not even"

# Point-free style bind!
result = Right(3) >> F . fn(&Right) . *(2)
result # => Right(6)
result.value # => 6

fmap

require "kleisli"

result = if foo > bar
  Right(10)
else
  Left("wrong")
end.fmap { |x| x * 2 }

# If everything went well
result # => Right(20)
# If it didn't
result # => Left("wrong")

* (applicative functor's apply)

require "kleisli"

add = -> x, y { x + y }
Right(add) * Right(10) * Right(2)
# => Right(12)
Right(add) * Left("error") * Right(2)
# => Left("error")

or

or does pretty much what would you expect:

require 'kleisli'

Right(10).or(Right(999)) # => Right(10)
Left("error").or(Left("new error")) # => Left("new error")
Left("error").or { |err| Left("new #{err}") } # => Left("new error")

to_maybe

Sometimes it's useful to turn an Either into a Maybe. You can use to_maybe for that:

require "kleisli"

result = if foo > bar
  Right(10)
else
  Left("wrong")
end.to_maybe

# If everything went well:
result # => Some(10)
# If it didn't
result # => None()

Future

The Future monad models a pipeline of computations that will happen in the future, as soon as the value needed for each step is available. It is useful to model, for example, a sequential chain of HTTP calls.

There's a catch unfortunately -- values passed to the functions are wrapped in lambdas, so you need to call .call on them. See the examples below.

>-> (bind)

require "kleisli"

f = Future("myendpoint.com") >-> url {
  Future { HTTP.get(url.call) }
} >-> response {
  Future {
    other_url = JSON.parse(response.call.body)[:other_url]
    HTTP.get(other_url)
  }
} >-> other_response {
  Future { JSON.parse(other_response.call.body) }
}

# Do some other stuff...

f.await # => block until the whole pipeline is realized
# => { "my" => "response body" }

fmap

require "kleisli"

Future { expensive_operation }.fmap { |x| x * 2 }.await
# => result of expensive_operation * 2

Who's this

This was made by Josep M. Bach (Txus) and Ryan Levick under the MIT license. We are @txustice and @itchyankles on twitter (where you should probably follow us!).