Feature #6166: Enumerator::Lazy#pinch

• Status: Rejected
• Priority: Normal
• Assignee: Yukihiro Matsumoto

[Thomas Sawyer]

In previous issue #6158 it has been determined that Enumerator::Lazy#take should be lazy. But an eager form of #take would still be useful.

To this end I'll suggest Enumerator::Lazy#pinch. Examples of usage:

e.lazy.pinch 1

e.lazy.pinch 1..2

e.lazy.pinch 1,2

It is basically equivalent to calling to_a[index], but has the advantage of being a single invocation instead of two, and reads better.

The #pinch method would be strictly a Lazy method and have no counterpart in Enumerable.

[Thomas Sawyer]

Happy St. Patty's Day ;)

I don't seen any green on this site... pinch

[Shugo Maeda]

Hello,

trans (Thomas Sawyer) wrote:

In previous issue #6158 it has been determined that Enumerator::Lazy#take should be lazy. But an eager form of #take would still be useful.

To this end I'll suggest Enumerator::Lazy#pinch. Examples of usage:

e.lazy.pinch 1

e.lazy.pinch 1..2

e.lazy.pinch 1,2

It is basically equivalent to calling to_a[index], but has the advantage of being a single invocation instead of two, and reads better.

Enumerator::Lazy#pinch provides a random access feature for Enumerator::Lazy, but Enumerator::Lazy is not random accessible in general, so I prefer explicit conversion like to_a[1, 2] to pinch. How about others?

[Thomas Sawyer]

I like having a method myself b/c it reads better. I think "pinch" conveys the sort of "closing action" of the de-lazying.

Of course, #fetch would work as well, and that's a standard method, but it's interface only accepts an index, not a range, so it's too limited as is.

[Yusuke Endoh]

• Assignee set to Yukihiro Matsumoto
• Status changed from Open to Assigned

Hello,

If this kind of operations occur frequently, I think it is worth
to add a method. But I'm not sure because Lazy just entered trunk.

My current opinion. I like this style:

e.lazy.pinch(1, 2) == e.lazy.drop(1).first(2)

because it is:

• more explicit than pinch
• more efficient than to_a1, 2

--
Yusuke Endoh [e-mail]

[Yukihiro Matsumoto]

• Status changed from Assigned to Rejected

I like #first better.

Matz.

[Thomas Sawyer]

But #first can't give a range. e.g. (2..3) or (2,2). So there is no way to get such without de-lazying whole enumeration, which defeats purpose of lazy.

[Thomas Sawyer]

Also are you sure #first should be non-lazy? e.g.

max_records = 1000000

recs = records.first(max_records)

recs.pinch(@page_no, 25).each do |page_recs|
...
end

Hmm... maybe #page is better name then #pinch?

In any case, the point is I think every enumeration method that can be lazy should be lazy, and a special method that's not an enumerable method should allows us to extract subsets. That method (whether called #pinch or something else) would be the most flexible and optimized since it is designed to very task of de-lazying and extraction.

[Yukihiro Matsumoto]

I am sure #first not to be lazy. And you can combine it with #drop to take the value in the middle.
#pinch does not suggest the behavior you've proposed to me, a non native English speaker.

Matz.

[Thomas Sawyer]

I don't care about name "#pinch", but functionality. Using #first with #drop is not always optimal.

1) Given d = [index,length].

enum.drop(d.first).first(d.last)

2) Given a range (e.g. rng = 2...4):

if rng.exclude_end?
enum.drop(rng.begin).first(rng.end-1)
else
enum.drop(rng.begin).first(rng.end)
end

enum(*d) and enum(rng) is much better.

Also I take it #drop is lazy?

[Marc-Andre Lafortune]

Sorry to be late to the party.

If this method was called slice, would it be more acceptable?

It would be easy to remember, as enum.slice would be the same as enum.to_a.slice except it would stop the iteration as early as possible.

[Yukihiro Matsumoto]

What happens when you call #pinch (or whatever) twice on same lazy sequence?

e.g.

lz = (1..100).lazy
lz.pinch(0,2)
lz.pinch(0,2)

If second call to pinch gives [1,2], lazy sequence needs to keep all generated values inside (that makes lazy sequence very inefficient both in time and space-wise).
If it gives [3,4], I don't think it's the expected behavior for most of us.

#pinch implies offset which is not well fit with the concept of lazy sequence which is not always indexable.

Matz.

[Thomas Sawyer]

It should have no effect on lz either way. It does not act in place.

lz = (1..100).lazy
lz.pinch(0,2) #=> [1,2]
lz.pinch(0,2) #=> [1,2]

Marc-Andre's suggestion of #slice is better name. I forgot about that method, but it is exactly what I intended --so long as #slice is not going to be lazy itself.

Sequences are indexable in that #to_a can be called, the method follows the same procedure but stops when the end sentinel is reached.

Hmmm... I suppose another option would be to let #to_a take arguments.

lz = (1..100).lazy
lz.to_a(0,2) #=> [1,2]

But that may not be good idea b/c #to_a does not take arguments in other classes.

[Yukihiro Matsumoto]

I am not sure what you mean by "does not act in place". Lazy sequence has position. Retrieving value from sequence moves its position, even calling to_a. So to make #pinch work as if "not acting in place", sequence needs to keep the values inside, everytime it generates.

I don't want to do that since it hinders the best benefit from "lazy" sequence.

Matz.

[Thomas Sawyer]

I don't understand what you mean by "So to make #pinch work as if "not acting in place", sequence needs to keep the values inside, everytime it generates." Pinch is no different than #to_a except that it does not need to "resolve" all elements of the enumerable, only the ones up to requested sentinel.

In other words, you reject this b/c you say it "hinders the best benefit from "lazy" sequence." But the result is that end user must use #to_a instead (e.g. enum.to_a[1..2]), which makes lazy utterly pointless. Moreover, if pinch causes this issue, why doesn't drop(b).first(e) cause the same issue, since "pinch" is effectively just a convenience for the same?

I wonder if we are not thinking about the same idea. Maybe some code would clear things up. "Pinch", which as Marc-Andre points out is effectively #slice, can be implemented basically as:

def slice(b, e=nil)
  if e
    drop(b).first(e)
  else
    case b
    when Range
      if b.exclude_end?
        drop(b.begin).first(b.end-1)
      else
        drop(b.begin).first(b.end)
      end
    else
      first(b).last
    end
  end
end