Let’s talk about loops.

1
2
3

for i in (1..10) do
  puts "i= #{i}"
end

Three weeks ago, when I was new to Ruby, I focussed on this. This was a loop. I mean, I come from Python and that’s what loops look like.

I was wrong, wasn’t I?

1

10.times { |i| puts "i = #{i}" }

If you haven’t read my post on Ruby blocks, you should. And then come back to this.

Damn, Ruby. Why do you need to be scratching my brain in ways I didn’t think were possible all the time?

Where’s that meme?

Khaby Lazy Meme

What’s so great about this? Either you’ve never written Ruby before or you have your eyes closed.

That wasn’t a “loop” now was it? It looks like you’re simply saying “do this 10 times”, but I’m fresh off the heels of diving into blocks and I know that .times is a method call and the { |i| puts "i = #{i}" } is a block.

We can inspect the class of a Ruby object like this:

1
2

puts 10.class # Integer
puts 10.respond_to?(:times) # true

Those are method calls. But what is a method call?

I come from Python, and in Python, a method call looks like this:

1

object.method(argument)

I’m hand-waving a lot of the details away, but here, we’re calling a method on an object. The object has the method. The method does something.

In Ruby, that analogy isn’t exactly right. You’re not calling a method. You’re sending a message.

1
2

10.times { puts "hi" }
10.send(:times) { puts "hi" }

If you use .send you might understand this better. You are sending the message :times to the object 10.

The object 10 receives that message and then decides what to do with it. You could say this is a segue into philosophy.

How do you code? You write logic in a form that the computer can understand, and you use the tools you’ve been given by the language of your choice. It’s electronic telepathy, albeit an extremely broken form of it. The code represents what you want to do.

By saying:

1
2

for i in range(10)
  print(f"i = {i}")

You are not saying “print 5 times”, although that’s what your brain interprets it as.

You are saying:

  • Make an iterable object that yields integers from 0 to 9.
  • Iterate through this object, and assign the yielded value to i at each turn.
  • Upon each iteration, print "i = #{i}" to the screen.
  • Break out of the loop when you exhaust the iterator.

That’s a mouthful. It might indicate the same thing as:

1

10.times { |i| puts "i = #{i}" }

But it isn’t.

In the ruby block, you are saying:

  • Call the :times method on the Integer object of value 10;
  • For each iteration, pass the block { |i| puts "i = #{i}" }.

That… is lesser?

No, it’s different.

In Python, you indicate your purpose by controlling the loop. The code doesn’t attach the intent to the number 10 or to the range(10) iterator.

In Ruby, you are asking the object to do something. You are directly attaching the meaning of your code to the Integer of value 10.

.send is the inner mechanic that orchestrates this. If you’ve done any amount of Python metaprogramming, you might have seen something like

1
2

method = obj.getattr("method_name")
method.__call__(args)

Note the difference in names here. It’s subtle enough to blink and miss. Python says __call__. Ruby says __send__.

If you’ve sat on the Erlang fence for as long as I have (I’ll learn it yet one day!), your message passing alarms are going off, aren’t they?

But this isn’t from Erlang, in fact there’s nothing similar besides the metaphor, it’s from Smalltalk. And this style of programming philosophy is Protocol over Syntax.

Meet Smalltalk

I’ve had a bunch of people tell me that Ruby is inspired by Smalltalk. I’d never written or read a line of Smalltalk before today, but I thought I’d check it out to understand Ruby’s philosophy a little.

1

5 timesRepeat: [ Transcript show: 'Hello' ].

Okay, that’s familiar enough, isn’t it?

1

1 to: 10 do [:i | Transcript show: i].

Okay, that looks similar to:

1

1.upto(10) { |i| puts i }

How about iterating over a list?

1

#(1 2 3 4 5) do: [:each | Transcript show: each ].

1

[1, 2, 3, 4, 5].each { |each| puts each }

To understand this paradigm, consider a PseudoInteger class that behaves like an integer but overwrites the times method. For brevity I’m not going to implement any of the other methods.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30

class PseudoInteger
  def initialize(value)
    @value = value
  end

  def times(&block)
    puts "I'm going to run #{@value} times"

    @value.times do |y|
      block.call(y)
    end
    puts "Done with the block calls."
  end
end

i = PseudoInteger.new(10)
i.times { |x| puts x }

# I'm going to run 10 times
# 0
# 1
# 2
# 3
# 4
# 5
# 6
# 7
# 8
# 9
# Done with the block calls.

I’ve said this earlier, but here, the object PseudoInteger controls the iteration through .times. But what if the object chose to lie to you about what it’s doing?

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18

class LyingPseudoInteger
  def initialize(value)
    @value = value
  end

  def times(&block)
    puts "I'm going to run #{@value} times"
    block.call(0)
    puts "Done with the block calls."
  end
end

i = LyingPseudoInteger.new(10)
i.times { |x| puts x }

# I'm going to run 10 times
# 0
# Done with the block calls

The new class LyingPseudoInteger does not do what it says it does.

Leaving the handling of the loop to the method allows us to add behaviour to loops that are controlled by the object and not by the user. This is a nice way to add side-effects.

But what does this have to do with message-passing?

In most languages, methods are bound to objects at compile time (or at least, the languages pretend they are). In Smalltalk, and by association Ruby, methods are looked up at runtime in response to messages.

In Python, Java or C++, calling object.method() asks the compiler to find the method in the class and call it.

In Ruby, 10.times { puts 'hi' } is literally interpreted as 10.send(:times) { puts 'hi' }., which is equivalent to 10.public_send(:times) { puts 'hi' }.

The :times is a symbol. I haven’t written about Symbols yet, that’s a whole other rabbithole that I have yet to go down in.

But with this sort of implementation, we could (but shouldn’t), do this:

1
2
3
4
5

class Integer
  undef_method :times
end

10.times { puts 'hi' } # -> NoMethodError : undefined method 'times' for 10:Integer

The Ruby interpreter is saying “I have received a message :times but I don’t know how to respond to it.”

But I’m still not convinced, so let’s try this.

1
2
3
4
5
6
7
8
9

class Integer
  def method_missing(method_name, *args, &block)
    puts "I received a message I don't understand: #{method_name}"
    puts "I could do anything here - respond, forward it, ignore it..."
    puts "This is message passing - the message exists independently of the method"
  end
end

10.foobar # I received a message I don't understand: foobar...

In traditional method calling, this would be a compile-time error. This method doesn’t exist, so compilation fails. In message passing, the message got sent. The object receives it and then says “I do not know this message.” This is what we can catch.

But Python also looks up methods at runtime. Does that mean Python also does message passing?

Not quite.

In Python, you’d access methods like this:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

class MyClass:
    def my_method(self):
      return "Hello"

obj = MyClass()

# Methods are just attributes here.
print(obj.__dict__) # {}
print(MyClass.__dict__) # {... 'my_method': <function>...}

method = getattr(obj, "my_method")
method()

However, getattr is generic, it’s for all attributes. Not just methods. You could call getattr(obj, 'anything) and get a property called anything on the object.

In Python, “calling a method” is interpreted as “look up an attribute that implements __call__”.

Ruby does something different.

1
2
3
4
5
6
7
8

class MyClass

  def my_method
    "hello"
  end
end

obj = MyClass.new

You can explicitly SEND messages: obj.send(:my_method).

The language has built-in constructs for message handling.

1
2

obj.respond_to?(:my_method)     # -> Can you respond to this message?
obj.method(:my_method)          # -> Give me the method that corresponds to this message

And we can intercept unknown messages with .method_missing as shown above. Ruby treats “calling a method” as “sending a message to an object.”

While Python is also interpreted, any attempt to attach middleware to this sort of method resolution is wrapped in attribute access interception. You are always getting attributes.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

class Spy:
    def __getattribute__(self, name):
        print(f"Looking up attribute: {name}")
        return super().__getattribute__(name)

    def hello(self):
        return "hi"

spy = Spy()
spy.hello()
# Looking up attribute: hello
# "hi"

In Ruby, you’re intercepting messages.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

class Spy
  def method_missing(name, *args, &block)
    puts "Received message: #{name}"
    super
  end

  def hello
    "hi"
  end
end

spy = Spy.new
spy.hello
# "hi"

spy.foobar
# Received message: foobar

The difference is in the philosophy. Python gets the attribute method on obj and if it’s callable, it calls it. Ruby sends the :method message to obj. This emphasizes object autonomy: the object receives a message and chooses its response.

This is taken to the extreme when you realize that Ruby has primitives (Symbols) that you can use to treat message passing like the first class concept that it is in Ruby.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13

msg = :times
puts msg.class          # Symbol

# You can get the method that will respond
method = 10.method(msg) 
puts method.class       # Method

# You can call it later
method.call { |i| puts i }

# You can see what messages an object responds to:

puts 10.methods.grep(/time/) # [:times]

In Python, doing this is rather roundabout.

1
2
3
4
5

method = getattr(10, "times", None)

# method is None

# You can use `dir()` to get all the attributes in an object, but you'd have to filter them for methods yourself.

But what is a loop?

In Ruby, you can iterate through anything that includes Enumerable.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20

class Countdown
  include Enumerable
  def initialize(from)
    @from = from
  end

  def each
    current = @from
    while current > 0
      yield current
      current -= 1
    end
  end
end

c = Countdown.new(5)
c.each { |x| puts x }
c.map { |x| x * 2 }
c.select { |x| x.even? }
c.reduce(:*)

By including Enumerable and by implementing each, we get access to so many methods that we didn’t need to manually implement. This is the same as the Collection protocol from Smalltalk - implementing :do: gives us the methods we need.

Why For Feels Like It Doesn’t Belong

By now, you should look at for as an alien thing. It doesn’t feel like it belongs. But why is that?

for is syntax, and Ruby, by design, encourages message passing.

1
2
3
4
5
6
7

[1, 2, 3].each { |x| y = x }
puts y                        # NameError

for x in [1, 2, 3]
  y = x
end
puts y # 3

Using for pollutes the namespace, spilling variables over to the including scope. But that’s not the only thing. for is syntactical sugar. It’s not message passing.

In a sense, the familiarity of the loop construct deceives you. I reached towards for in my first days learning Ruby thinking that’s the way to loop. Yet, Rubyists encouraged me to look into .each and .times to write loops. I wanted to understand that choice better. It’s a different paradigm, because you could do something in Python that does the same thing, but it’s still not the same. You’d still shoehorn in the loop syntax and use that. And of course it also begs the question: why should you?

The Beauty of Protocol Over Syntax

In Smalltalk, there is no for loop. Everything is message passing.

To loop 5 times you send :timesRepeat to 5.

To iterate over a collection, send :do: to the collection.

To filter, send :select: to the collection.

Matz took this philosophy and ran with it.

1
2
3
4

5.times { puts 'Hello' }         # timesRepeat:
[1, 2, 3].each { |x| puts x }    # do:
[1, 2, 3].select { |x| x.even? } # select:
[1, 2, 3].map { |x| x * 2 }      # collect:

Understanding the message passing philosophy helps you map Ruby’s choices.

  • Integers have .times because they need respond to iteration messages.
  • Arrays have .each for the same reason.
  • Adding .countdown to an Integer teaches it to respond to a new message.
  • Adding include Enumerable to a class and implementing .each gives you the ability to send multiple types of iterative messages to objects of that class, because that implements the Collection Protocol.
  • for doesn’t send messages the same way.

Writing Ruby in Ruby and Python in Python is important. I’ve been a strong believer of learning how to write idiomatic code in a programming language. You cannot transplant features between languages without coding with an “accent”.

Asking an object to iterate over itself allows objects to develop interfaces that dictate how to iterate. This is at the heart of the message passing paradigm that Ruby and Smalltalk use.

Writing Ruby encourages you to embrace this, and that helps you build a sense of style that it comes with.

I’ve noticed this of Ruby. Python, a language I love, looks different depending on who’s writing it. Ruby coaxes you to develop this sense of style, whether you want to spend a weekend learning why it does this or not.

And now, when I see:

1

10.times { |i| puts "i = #{i}" }

I do not see a loop anymore. I see an object responding to a message. I see Smalltalk’s legacy. I see protocol over syntax.