pipe_operator

Elixir/Unix style pipe operations in Ruby - PROOF OF CONCEPT

"https://api.github.com/repos/ruby/ruby".pipe do
  URI.parse
  Net::HTTP.get
  JSON.parse.fetch("stargazers_count")
  yield_self { |n| "Ruby has #{n} stars" }
  Kernel.puts
end
#=> Ruby has 15120 stars

-9.pipe { abs; Math.sqrt; to_i } #=> 3

# Method chaining is supported:
-9.pipe { abs; Math.sqrt.to_i.to_s } #=> "3"

sqrt = Math.pipe.sqrt #=> #<PipeOperator::Closure:[email protected]_operator/closure.rb:18>
sqrt.call(9)          #=> 3.0
sqrt.call(64)         #=> 8.0

[9, 64].map(&Math.pipe.sqrt)           #=> [3.0, 8.0]
[9, 64].map(&Math.pipe.sqrt.to_i.to_s) #=> ["3", "8"]

Why?

There's been some recent activity related to Method and Proc composition in Ruby:

• #6284 - Add composition for procs
• #13581 - Syntax sugar for method reference
• #12125 - Shorthand operator for Object#method

This gem was created to propose an alternative syntax for this kind of behavior.

Matz on Ruby

Source: ruby-lang.org/en/about

Ruby is a language of careful balance of both functional and imperative programming.

Matz has often said that he is trying to make Ruby natural, not simple, in a way that mirrors life.

Building on this, he adds: Ruby is simple in appearance, but is very complex inside, just like our human body.

Concept

The general idea is to pass the result of one expression as an argument to another expression - similar to Unix pipelines:

echo "testing" | sed "s/ing//" | rev
#=> tset

The Elixir pipe operator documentation has some other examples but basically it allows expressions like:

JSON.parse(Net::HTTP.get(URI.parse(url)))

To be inverted and rewritten as left to right or top to bottom which is more natural to read in English:

# left to right
url.pipe { URI.parse; Net::HTTP.get; JSON.parse }

# or top to bottom for clarity
url.pipe do
  URI.parse
  Net::HTTP.get
  JSON.parse
end

The differences become a bit clearer when other arguments are involved:

loans = Loan.preapproved.submitted(Date.current).where(broker: Current.user)
data = loans.map { |loan| LoanPresenter.new(loan).as_json }
json = JSON.pretty_generate(data, allow_nan: false)

Using pipes removes the verbosity of maps and temporary variables:

json = Loan.pipe do
  preapproved
  submitted(Date.current)
  where(broker: Current.user)
  map(&LoanPresenter.new.as_json)
  JSON.pretty_generate(allow_nan: false)
end

While the ability to perform a job correctly and efficiently is certainly important - the true beauty of a program lies in its clarity and conciseness:

"https://api.github.com/repos/ruby/ruby".pipe do
  URI.parse
  Net::HTTP.get
  JSON.parse.fetch("stargazers_count")
  yield_self { |n| "Ruby has #{n} stars" }
  Kernel.puts
end
#=> Ruby has 15115 stars

There's nothing really special here - it's just a block of expressions like any other Ruby DSL and pipe operations have been around for decades!

Ruby.is.so(elegant, &:expressive).that(you can) do
  pretty_much ANYTHING if it.compiles!
end

This concept of pipe operations could be a great fit like it has been for many other languages:

• Caml composition operators
• Closure threading macros
• Elixir pipe operator
• Elm operators
• F# function composition and pipelining
• Hack pipe operator
• Haskell pipes
• JavaScript pipeline operator proposals
• LiveScript piping
• Unix pipelines

Usage

WARNING - EXPERIMENTAL PROOF OF CONCEPT

This has only been tested in isolation with RSpec and Ruby 2.5.3!

# First `gem install pipe_operator`
require "pipe_operator"

Implementation

The PipeOperator module has a method named __pipe__ which is aliased as pipe for convenience:

module PipeOperator
  def __pipe__(*args, &block)
    Pipe.new(self, *args, &block)
  end
end

BasicObject.send(:include, PipeOperator)
Kernel.alias_method(:pipe, :__pipe__)

When no arguments are passed to __pipe__ then a PipeOperator::Pipe object is returned:

Math.pipe #=> #<PipeOperator::Pipe:Math>

Any methods invoked on this object returns a PipeOperator::Closure which calls the method on the object later:

sqrt = Math.pipe.sqrt       #=> #<PipeOperator::Closure:[email protected]_operator/closure.rb:18>
sqrt.call(16)               #=> 4.0

missing = Math.pipe.missing #=> #<PipeOperator::Closure:[email protected]_operator/closure.rb:18>
missing.call                #=> NoMethodError: undefined method 'missing' for Math:Module

Math.method(:missing)       #=> NameError: undefined method 'missing' for class '#<Class:Math>'

When __pipe__ is called with arguments but without a block then it behaves similar to __send__:

sqrt = Math.pipe(:sqrt) #=> #<PipeOperator::Closure:[email protected]_operator/closure.rb:18>
sqrt.call(16)           #=> 4.0

sqrt = Math.pipe(:sqrt, 16) #=> #<PipeOperator::Closure:[email protected]_operator/closure.rb:18>
sqrt.call                   #=> 4.0
sqrt.call(16)               #=> ArgumentError: wrong number of arguments (given 2, expected 1)

These PipeOperator::Closure objects can be bound as block arguments just like any other Proc:

[16, 256].map(&Math.pipe.sqrt) #=> [4.0, 16.0]

Simple closure composition is supported via method chaining:

[16, 256].map(&Math.pipe.sqrt.to_i.to_s) #=> ["4", "16"]

The block form of __pipe__ behaves similar to instance_exec but can also call methods on other objects:

"abc".pipe { reverse }        #=> "cba"
"abc".pipe { reverse.upcase } #=> "CBA"

"abc".pipe { Marshal.dump }                  #=> "\x04\bI\"\babc\x06:\x06ET"
"abc".pipe { Marshal.dump; Base64.encode64 } #=> "BAhJIghhYmMGOgZFVA==\n"

Outside the context of a __pipe__ block things behave like normal:

Math.sqrt     #=> ArgumentError: wrong number of arguments (given 0, expected 1)
Math.sqrt(16) #=> 4.0

But within a __pipe__ block the Math.sqrt expression returns a PipeOperator::Closure instead:

16.pipe { Math.sqrt }     #=> 4.0
16.pipe { Math.sqrt(16) } #=> ArgumentError: wrong number of arguments (given 2, expected 1)

The piped object is passed as the first argument by default but can be customized by specifying self:

class String
  def self.join(*args, with: "")
    args.map(&:to_s).join(with)
  end
end

"test".pipe { String.join("123", with: "-") }       #=> "test-123"

"test".pipe { String.join("123", self, with: "-") } #=> "123-test"

Instance methods like reverse below do not receive the piped object as an argument since it's available as self:

Base64.encode64(Marshal.dump("abc").reverse)          #=> "VEUGOgZjYmEIIkkIBA==\n"

"abc".pipe { Marshal.dump; reverse; Base64.encode64 } #=> "VEUGOgZjYmEIIkkIBA==\n"

"abc".pipe { Marshal.dump.reverse; Base64.encode64 }  #=> "VEUGOgZjYmEIIkkIBA==\n"

Pipes also support multi-line blocks for clarity:

"abc".pipe do
  Marshal.dump.reverse
  Base64.encode64
end

The closures created by these pipe expressions are evaluated via reduce:

pipeline = [
  -> object { Marshal.dump(object) },
  -> object { object.reverse },
  -> object { Base64.encode64(object) },
]

pipeline.reduce("abc") do |object, pipe|
  pipe.call(object)
end

Intercepting methods within pipes requires prepending a PipeOperator::Proxy module infront of ::Object and all nested constants:

define_method(method) do |*args, &block|
  if Pipe.open
    Pipe.new(self).__send__(method, *args, &block)
  else
    super(*args, &block)
  end
end

These proxy modules are prepended everywhere!

It's certainly something that could be way more efficient as a core part of Ruby.

Maybe somewhere lower level where methods are dispatched? Possibly somewhere in this vm_eval.c switch?

again:
  switch (cc->me->def->type) {
    case VM_METHOD_TYPE_ISEQ
    case VM_METHOD_TYPE_NOTIMPLEMENTED
    case VM_METHOD_TYPE_CFUNC
    case VM_METHOD_TYPE_ATTRSET
    case VM_METHOD_TYPE_IVAR
    case VM_METHOD_TYPE_BMETHOD
    case VM_METHOD_TYPE_ZSUPER
    case VM_METHOD_TYPE_REFINED
    case VM_METHOD_TYPE_ALIAS
    case VM_METHOD_TYPE_MISSING
    case VM_METHOD_TYPE_OPTIMIZED
    case OPTIMIZED_METHOD_TYPE_SEND
    case OPTIMIZED_METHOD_TYPE_CALL
    case VM_METHOD_TYPE_UNDEF
  }

Then we'd only need Ruby C API ports for PipeOperator::Pipe and PipeOperator::Closure!

All other objects in this proof of concept are related to method interception and would no longer be necessary.

Bugs

This test case doesn't work yet - seems like the object is not proxied for some reason:

class Markdown
  def format(string)
    string.upcase
  end
end

"test".pipe(Markdown.new, &:format) # expected "TEST"
#=> ArgumentError: wrong number of arguments (given 0, expected 1)

Caveats

• PIPE_OPERATOR_AUTOLOAD
  • Constants flagged for autoload are NOT proxied by default (for performance)
  • Set ENV["PIPE_OPERATOR_AUTOLOAD"] = 1 to enable this behavior
• PIPE_OPERATOR_FROZEN
  • Objects flagged as frozen are NOT proxied by default
  • Set ENV["PIPE_OPERATOR_FROZEN"] = 1 to enable this behavior (via Fiddle)
• PIPE_OPERATOR_REBIND
  • Object and its recursively nested constants are only proxied ONCE by default (for performance)
  • Constants defined after __pipe__ is called for the first time are NOT proxied
  • Set ENV["PIPE_OPERATOR_REBIND"] = 1 to enable this behavior
• PIPE_OPERATOR_RESERVED
  • The following methods are reserved on PipeOperator::Closure objects:
    • ==
    • []
    • __chain__
    • __send__
    • __shift__
    • call
    • class
    • kind_of?
  • The following methods are reserved on PipeOperator::Pipe objects:
    • !
    • !=
    • ==
    • __call__
    • __id__
    • __pop__
    • __push__
    • __send__
    • instance_exec
    • method_missing
  • These methods can be piped via send as a workaround:
    • 9.pipe { Math.sqrt.to_s.send(:[], 0) }
    • example.pipe { send(:__call__, 1, 2, 3) }
    • example.pipe { send(:instance_exec) { } }

Testing

bundle exec rspec

Inspiration

• https://github.com/hopsoft/pipe_envy
• https://github.com/akitaonrails/chainable_methods
• https://github.com/kek/pipelining
• https://github.com/k-motoyan/shelike-pipe
• https://github.com/nwtgck/ruby_pipe_chain
• https://github.com/teamsnap/pipe-ruby
• https://github.com/danielpclark/elixirize
• https://github.com/tiagopog/piped_ruby
• https://github.com/jicksta/methodphitamine
• https://github.com/jicksta/superators
• https://github.com/baweaver/xf

Contributing

• Fork the project.
• Make your feature addition or bug fix.
• Add tests for it. This is important so we don't break it in a future version unintentionally.
• Commit, do not mess with the version or history.
• Open a pull request. Bonus points for topic branches.

Authors

• Sean Huber

License

MIT - Copyright © 2018 LendingHome