From: Tanaka Akira <akr@...>
Date: 2011-09-17T09:07:03+09:00
Subject: [ruby-core:39595] Re: [Ruby 1.9 - Bug #4576] Range#step miss the last value, if end-exclusive and has float number

2011/9/17 Tanaka Akira <akr@fsij.org>:
>
> I don't think the patch is a appropriate fix for this problem.
>
> The test, "n*unit+beg < end", is fragile.

I made a sample script to show the fragileness on x86_64 to explain
this is not the x86 80bit float issue.
(x86_64 doesn't use 80bit float.)

% ./ruby -v
ruby 1.9.4dev (2011-09-15 trunk 33274) [x86_64-linux]
% ./ruby -e '
h = Hash.new(0)
1000.times {
  a = rand
  b = a+rand*10000
  s = (b - a) / 10
  l = (a...b).step(s).to_a.length
  h[l] += 1
}
p h
'
{10=>940, 11=>60}

In this script, the result length vary.

Some people may think this is not a bug because
floating point calculation may have errors.

But in the following script, which changes "a...b" to "a..b" from the above
script, the result length doesn't vary.

% ./ruby -e '
h = Hash.new(0)
1000.times {
  a = rand
  b = a+rand*10000
  s = (b - a) / 10
  l = (a..b).step(s).to_a.length
  h[l] += 1
}
p h
'
{11=>1000}

This is because we tried to consider float errors
in Numeric#step for Ruby 1.8.  [ruby-dev:20163]

The implementation is extracted and reused for Range#step for Ruby 1.9.
[ruby-dev:37691] r21298

The exclude_end support in the implementation is new
since Numeric#step don't have correspondence to exclude_end.

So my understanding of this problem is that no one implemented
proper exclude_end support with well considered float errors, yet.
-- 
Tanaka Akira