Project

General

Profile

Actions

Feature #15950

closed

Allow negative length in `Array#[]`, `Array#[]=`, `Array#slice`, `Array#slice!`, `String#[]`, `String#[]=`, `String#slice`, `String#slice!`

Added by sawa (Tsuyoshi Sawada) over 5 years ago. Updated over 5 years ago.

Status:
Rejected
Assignee:
-
Target version:
-
[ruby-core:93308]

Description

To take the first n characters of a string, using [] is straightforward:

"abcdefgh"[0, 3] # => "abc"

But to take the last n characters, we need to use n in two arguments: in the index (in negative form) in addition to the length:

"abcdefgh"[-3, 3] # => "fgh"

This is cumbersome.

I wish negative length to be allowed, and be interpreted as measuring leftward (while cycling the receiver if necessary).

"abcdefgh"[0, -3] # => "fgh"
"abcdefgh"[5, -3] # => "cde"

If there is not enough characters or elements, it should stop at the boundary.

"abcdefgh"[1, -3] # => "a"
Actions #1

Updated by sawa (Tsuyoshi Sawada) over 5 years ago

  • Subject changed from Allow negative length in `Array#[]`, `Array#[]=`, `String#[]`, `String[]=` to Allow negative length in `Array#[]`, `Array#[]=`, `String#[]`, `String#[]=`
Actions #2

Updated by sawa (Tsuyoshi Sawada) over 5 years ago

  • Description updated (diff)

Updated by nobu (Nobuyoshi Nakada) over 5 years ago

Negative offset is allowed, but negative length has never been allowed, IIRC.

Updated by shevegen (Robert A. Heiler) over 5 years ago

Hmm. I do not necessarily doubt that it may be useful (at the least for
some users), but that also actually surprised me. For example:

"abcdefgh"[0, -3] # => "fgh"

I guess my brain got confused here, since the 0 would put my brain to
assume to "start at the most left position". So that code confused me.

I am mostly neutral on this though, but have a slight preference to keep
the status quo without adding negative length. Either way, it may be a
good idea to ask matz because this may have been a old design decision
(at the least it seems plausible to me that this may have been in ruby
for a long time).

Updated by Eregon (Benoit Daloze) over 5 years ago

  • Status changed from Open to Rejected

sawa (Tsuyoshi Sawada) wrote:

But to take the last n characters, we need to use n in two arguments: in the index (in negative form) in addition to the length:

"abcdefgh"[-3, 3] # => "fgh"

Why not:

"abcdefgh"[-3..-1] # => "fgh"

I would think the Range form is more idiomatic for this usage.

Also, with endless ranges, we can just do:

"abcdefgh"[-3..] # => "fgh"

I find the negative length very confusing (I can't even understand the example above, rotation should not be part of indexing IMHO),
and I'm negative on making indexing semantics a lot more complicated.

So I think there are already good ways to achieve this, and I don't see a rationale to include this complicated feature, so I would like to reject this ticket.
Feel free to motivate why it's better than the existing 2 alternate syntax, though.

Updated by sawa (Tsuyoshi Sawada) over 5 years ago

shevegen (Robert A. Heiler) wrote:

"abcdefgh"[0, -3] # => "fgh"

I guess my brain got confused here, since the 0 would put my brain to
assume to "start at the most left position". So that code confused me.

It is the same idea as negative index. If you did not have the concept of cycling, then you would be surprised with negative index too. You would need to say "0 lets me assume to start at the left most position. If I had a negative index instead, then what can be even more left of that?" The key to this is cycling: to assume that the left end is attached to the right end so that you can continuously move beyond this border. I suggested to apply that notion to length as well. To give a metaphor, there is a video game called Pac Man, in which if you keep moving the character leftward past the left edge of the screen, you appear from the right egde, still moving leftward.

And needless to say, this proposal would be irrelevant to Ruby beginners who insist on needlessly creating a range object as in "abcdefg"[0...3] to take the first three characters of the string, when you can, and should, do "abcdefg"[0, 3], which is more straightforward and efficient.

Updated by Eregon (Benoit Daloze) over 5 years ago

sawa (Tsuyoshi Sawada) wrote:

And needless to say, this proposal would be irrelevant to Ruby beginners who insist on needlessly creating a range object as in "abcdefg"[0...3] to take the first three characters of the string, when you can, and should, do "abcdefg"[0, 3], which is more straightforward and efficient.

I am not a Ruby beginner and I would still recommend the Range notation, which I think is more idiomatic and easier to read. So whether people should use the Range notation or not is obviously subjective.

Moreover, you mention performance but it's not so relevant here.
For instance, TruffleRuby can escape analyze the Range, so it has no cost in this case.
And even on MRI in such a micro-benchmark there is only 10% of difference, which seems negligible.

require 'benchmark/ips'

S = "abcdefgh"
r = nil

Benchmark.ips do |x|
  x.report("s[0, 3]") do
    r = S[0, 3]
  end
  x.report("s[0...3]") do
    r = S[0...3]
  end
  x.compare!
end

Results:

ruby -v bench.rb
ruby 2.6.2p47 (2019-03-13 revision 67232) [x86_64-linux]
Warming up --------------------------------------
             s[0, 3]   411.458k i/100ms
            s[0...3]   369.962k i/100ms
Calculating -------------------------------------
             s[0, 3]      9.298M (± 2.3%) i/s -     46.495M in   5.003131s
            s[0...3]      8.471M (± 1.8%) i/s -     42.546M in   5.024158s

Comparison:
             s[0, 3]:  9298307.0 i/s
            s[0...3]:  8470923.4 i/s - 1.10x  slower
ruby -v bench.rb
truffleruby 19.0.0, like ruby 2.6.2, GraalVM CE Native [x86_64-linux]
Warming up --------------------------------------
             s[0, 3]     1.201M i/100ms
            s[0...3]     1.204M i/100ms
Calculating -------------------------------------
             s[0, 3]     27.141M (±31.9%) i/s -    114.054M in   5.016116s
            s[0...3]     26.553M (±29.1%) i/s -    115.561M in   5.044515s

Comparison:
             s[0, 3]: 27140752.2 i/s
            s[0...3]: 26553278.2 i/s - same-ish: difference falls within error
Actions #8

Updated by sawa (Tsuyoshi Sawada) over 5 years ago

  • Subject changed from Allow negative length in `Array#[]`, `Array#[]=`, `String#[]`, `String#[]=` to Allow negative length in `Array#[]`, `Array#slice`, `Array#[]=`, `String#[]`, `String#[]=`, `String#slice`
Actions #9

Updated by sawa (Tsuyoshi Sawada) over 5 years ago

  • Subject changed from Allow negative length in `Array#[]`, `Array#slice`, `Array#[]=`, `String#[]`, `String#[]=`, `String#slice` to Allow negative length in `Array#[]`, `Array#[]=`, `Array#slice`, `Array#slice!`, `String#[]`, `String#[]=`, `String#slice`, `String#slice!`

Updated by Eregon (Benoit Daloze) over 5 years ago

I'm negative on making indexing semantics a lot more complicated.

To clarify, I think adding more logic to indexing would slow down indexing in general, which is undesirable and goes against your goal.

Besides, this is really hard to read and understand, and is not intuitive, contrary to existing indexing ways.

Updated by matz (Yukihiro Matsumoto) over 5 years ago

I don't see any rational use of this proposal. Rejection confirmed.

Matz.

Actions

Also available in: Atom PDF

Like0
Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0Like0