Bug #17218: Range#step sometimes behaves unexpectedly with Rational endpoints and increment

• Status: Closed
• Assignee: mrkn (Kenta Murata)
• Target version: -
• ruby -v: ruby 2.7.2p137 (2020-10-01 revision 5445e04352) [x86_64-darwin19]
• Backport: 2.5: DONTNEED, 2.6: REQUIRED, 2.7: DONE

[rorymolinari (Rory Molinari)]

I am using the latest stable version. The same behavior exists in 2.6.6.

11:28:25 $ ruby -v                                                                                                                                                                                                                    
ruby 2.7.2p137 (2020-10-01 revision 5445e04352) [x86_64-darwin19]

When creating an Enumerator::ArithmeticSequence with Rational endpoints and increment, sometimes inconsistent behavior results due to floating-point approximation in #last.

x = Rational(10997, 10000)
y = Rational(11, 10)
s = Rational(1, 10000)

puts "#{[x, y, s].map(&:to_f)}" # -> [1.0997, 1.1, 0.0001]

# intention: this contains exactly the precise Rational representations of 1.0997, 1.0998, 1.0999, 1.1
arith_seq = (x..y).step(s)

puts arith_seq.class # -> Enumerator::ArithmeticSequence

# Things look OK
puts arith_seq.first # -> 10997/10000
puts arith_seq.end  # -> 11/10
puts arith_seq.step  # -> 1/10000

# But the array that we get from #to_a is missing the last element, (11/10)
puts arith_seq.to_a.to_s # -> [(10997/10000), (5499/5000), (10999/10000)]

# This is apparently due to the value of #last
puts arith_seq.last # -> 1.0999999999999999

# The object itself is confused
puts arith_seq.size # -> 4
puts arith_seq.to_a.size # -> 3

The issue is in the Enumerator::ArithmeticSequence instance we get when we call Range#step without a block. The Range#step method passes the right things when it gets a block.

block_vals = []
(x..y).step(s) { |v| block_vals << v }
puts block_vals.to_s # -> [(10997/10000), (5499/5000), (10999/10000), (11/10)]

I would expect arith_seq.last to be the exact value Rational(11, 10). After all, arith_seq was created from a Range with Rational endpoints and given a rational step size.

[rorymolinari (Rory Molinari)]

• Subject changed from Range#step sometimes behaves unexpectedly with Rational endpoints and increment due to silent floating-point approximation to Range#step sometimes behaves unexpectedly with Rational endpoints and increment

[mame (Yusuke Endoh)]

• Status changed from Open to Assigned
• Assignee set to mrkn (Kenta Murata)

[jeremyevans0 (Jeremy Evans)]

This is caused by calling rb_int_plus/rb_int_mul/rb_int_minus directly instead of calling Ruby methods in arith_seq_last. I'm guessing the use of Rational begin/end/step was not anticipated. Calling Ruby methods seems to fix the issue. I've submitted a pull request with that approach: https://github.com/ruby/ruby/pull/3810

[mrkn (Kenta Murata)]

@jeremyevans0 (Jeremy Evans) Thank you for making a candidate patch. I have reviewed it, I noticed that it resolves only an issue of last method, but does not of to_a method.
I made another patch to resolve both issues: https://github.com/ruby/ruby/pull/3870

• Status changed from Assigned to Closed

Applied in changeset git|fad3023e94c45e7f03478732f7641b6f39ba9d12.

---

Fix ArithmeticSequence#last and ArithmeticSequence#each for non-integer sequences (#3870)

[Bug #17218]
[ruby-core:100312]

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN to 2.5: DONTNEED, 2.6: REQUIRED, 2.7: REQUIRED

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.5: DONTNEED, 2.6: REQUIRED, 2.7: REQUIRED to 2.5: DONTNEED, 2.6: REQUIRED, 2.7: DONE

ruby_2_7 82bce422ba9e131e62b528854dea69a6e8cc0c04 merged revision(s) 254bed302752a401b5fcc3b6c65a9c93711d91d6,fad3023e94c45e7f03478732f7641b6f39ba9d12,3156fb0f2c3ebf8229f392c8502c08fe165ab181.

[usa (Usaku NAKAMURA)]

backported into ruby_2_6 at r67936