What's the Best Way to Return an Enumerator::Lazy When Your Class Doesn't Define #Each

What's the best way to return an Enumerator::Lazy when your class doesn't define #each?

I think you should return a normal Enumerator using to_enum:

class Calendar
# ...
def each_from(first)
return to_enum(:each_from, first) unless block_given?
loop do
yield first if include?(first)
first += step
end
end
end

This is what most rubyists would expect. Even though it's an infinite Enumerable, it is still usable, for example:

Calendar.new.each_from(1.year.from_now).first(10) # => [...first ten dates...]

If they actually need a lazy enumerator, they can call lazy themselves:

Calendar.new.each_from(1.year.from_now)
.lazy
.map{...}
.take_while{...}

If you really want to return a lazy enumerator, you can call lazy from you method:

  # ...
def each_from(first)
return to_enum(:each_from, first).lazy unless block_given?
#...

I would not recommend it though, since it would be unexpected (IMO), could be an overkill and will be less performant.

Finally, there are a couple of misconceptions in your question:

  • All methods of Enumerable assume an each, not just lazy.

  • You can define an each method that requires a parameter if you like and include Enumerable. Most methods of Enumerable won't work, but each_with_index and a couple of others will forward arguments so these would be usable immediately.

  • The Enumerator.new without a block is gone because to_enum is what one should use. Note that the block form remains. There's also a constructor for Lazy, but it's meant to start from an existing Enumerable.

  • You state that to_enum never creates a lazy enumerator, but that's not entirely true. Enumerator::Lazy#to_enum is specialized to return a lazy enumerator. Any user method on Enumerable that calls to_enum will keep a lazy enumerator lazy.

Returning Enumerator in own each method

Your first issue is that you're not explicitly returning anything from your method. In the case that a block is passed, the convention for each is to return self, as Array does.

In the case that a block is not given, the standard way to create an Enumerator is with the Kernel#to_enum method. So, for example, you can use the following as the first line of your method, as recommended in What's the best way to return an Enumerator::Lazy when your class doesn't define #each?

return to_enum(:each) unless block_given?

Enumerator returning each line of a file, but chomped

You can create the enumerator as follows. First, let's create a file for demonstration.

str =<<_
Now is the
time for all
Rubiests to
come to the
aid of their
bowling team.
_

fname = "temp"
File.write(fname, str)
#=> 75

IO#foreach without a block returns an enumerator:

efe = File.foreach(fname)
#=> #<Enumerator: File:foreach("temp")>

so we need to merely embed this enumerator in another that chomps the newline character from a line:

echomp = Enumerator.new do |y|
loop do
y << efe.next.chomp
end
end
#=> #<Enumerator: #<Enumerator::Generator:0x007fbe128837b8>:each>

Let's try it:

echomp.next
#=> "Now is the"
echomp.next
#=> "time for all"
echomp.next
#=> "Rubiests to"
echomp.next
#=> "come to the"
echomp.next
#=> "aid of their"
echomp.next
#=> "bowling team."
echomp.next
#=> StopIteration: iteration reached an end

You could of course wrap this in a method:

def foreach_with_chomp(fname)
efe = File.foreach(fname)
Enumerator.new do |y|
loop do
y << efe.next.chomp
end
end
end

foreach_with_chomp(fname).each { |s| print "#{s} " }
Now is the time for all Rubiests to come to the aid of their bowling team.

Why return an enumerator?

This completely in accordance with the spirit of 1.9: to return enumerators whenever possible. String#bytes, String#lines, String#codepoints, but also methods like Array#permutation all return an enumerator.

In ruby 1.8 String#to_a resulted in an array of lines, but the method is gone in 1.9.

How to use an enumerator

The main distinction between an Enumerator and most other data structures in the Ruby core library (Array, Hash) and standard library (Set, SortedSet) is that an Enumerator can be infinite. You cannot have an Array of all even numbers or a stream of zeroes or all prime numbers, but you can definitely have such an Enumerator:

evens = Enumerator.new do |y|
i = -2
y << i += 2 while true
end

evens.take(10)
# => [0, 2, 4, 6, 8, 10, 12, 14, 16, 18]

zeroes = [0].cycle

zeroes.take(10)
# => [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

So, what can you do with such an Enumerator? Well, three things, basically.

  1. Enumerator mixes in Enumerable. Therefore, you can use all Enumerable methods such as map, inject, all?, any?, none?, select, reject and so forth. Just be aware that an Enumerator may be infinite whereas map returns an Array, so trying to map an infinite Enumerator may create an infinitely large Array and take an infinite amount of time.

  2. There are wrapping methods which somehow "enrich" an Enumerator and return a new Enumerator. For example, Enumerator#with_index adds a "loop counter" to the block and Enumerator#with_object adds a memo object.

  3. You can use an Enumerator just like you would use it in other languages for external iteration by using the Enumerator#next method which will give you either the next value (and move the Enumerator forward) or raise a StopIteration exception if the Enumerator is finite and you have reached the end.

Eg., an infinite range: (1..1.0/0)

Is there a way to pass a size to `enum_for`/`to_enum`?

You can pass a block to enum_for that calculates the enum's size:

enum_for(:each) { width * height }

Get slice of an Enumerator effectively

An Enumerator is very general interface, it makes only very simple assumptions about the "collection" it is traversing. In particular, it really only supports two operations: get the current element and iterate to the next element.

Given those two operations, if you want to get the 10 millionth element, there is only one thing you can do: iterate 10 million times. Which takes time.

There is no such thing as "slicing" an Enumerator. An Enumerator enumerates. That's it.

Now, as you discovered, there is another problem: Ruby's collection operations are not type-preserving. No matter what type of collection you call map or select or take or whatever on, it will always return the same type: a fully realized, concrete, strict Array. That's how most collection frameworks in most languages work, e.g. in .NET all collection operations return IEnumerable. That's because most of these methods have only a single common implementation in the Enumerable mixin.

Smalltalk is an exception, but there is another problem: the collection operations are duplicated for every single collection type. Every collection type has its own nearly-indetical practically copy&paste implementation of collect:, select: etc. This code duplication is hard to maintain and places a big burden on anyone who wants to integrate their own collection into the framework. In Ruby, it's easy: implement each, mixin Enumerable and you're done.

Note: as of Ruby 1.9, there is actually some of that duplication: Hash implements its own version of select which does actually return a Hash and not an Array. So, now, not only is there code duplication but also an asymmetry in the interface: all implementations of select return Arrays except for the one in Hash.

The Scala 2.8 collection framework is the first time ever that someone has figured out how to provide type-preserving collection operations without code duplication. But Ruby's collection framework was designed 15 years before Scala 2.8, so it cannot take advantage of that knowledge.

In Ruby 2.0, there are lazy Enumerators, where all collection operations return another lazy Enumerator. But that won't help you here: the only difference is that the lazy Enumerator will delay the 10 million iterations until you actually print the values. It still has to perform those 10 million iterations because there is simply no way to do otherwise.

If you want slicing, you need a sliceable data structure, such as an Array.



Related Topics



Leave a reply



Submit