Bug #16518: Should we rationalize Rational's numerator automatically?

• Status: Closed
• Assignee: mrkn (Kenta Murata)
• Target version: -
• ruby -v: 2.6.5
• Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN

[st0012 (Stan Lo)]

In this Rails issue we found that using BigDecimal as a Rational's numerator can have inconsistent behaviors when the denominator gets bigger. For example:

> 1 == BigDecimal(1)
=> true
> Rational(1, 1) == Rational(BigDecimal(1), 1)
=> true
> Rational(1, 6) == Rational(BigDecimal(1), 6)
=> true
> Rational(1, 60) == Rational(BigDecimal(1), 60)
=> false

So my question is, is this behavior expected?
If it's not expected, do we have a plan to fix it?
If it is, does it make sense to manually rationalize the numerator before passing it into the Rational call, in order to get a consistent result?

[Eregon (Benoit Daloze)]

• Assignee set to mrkn (Kenta Murata)

This seems due to the fact Rational(BigDecimal, Integer) does not return a Rational but a BigDecimal!

That seems highly surprising to me, any Foo() "constructor" should return a Foo, isn't it?

Also a Float as numerator correctly produces a Rational, so the bug seems to be with BigDecimal as a numerator:

> Rational(BigDecimal(1), 2).class
=> BigDecimal
> Rational(BigDecimal(1).to_f, 2).class
=> Rational

[st0012 (Stan Lo)]

Thanks for the quick response!

I also found that it returns BigDecimal if any of its arguments is a BigDecimal:

irb(main):005:0> Rational(BigDecimal(1), 1).class
=> BigDecimal
irb(main):006:0> Rational(BigDecimal(1), BigDecimal(1)).class
=> BigDecimal
irb(main):007:0> Rational(0, BigDecimal(1)).class
=> BigDecimal
irb(main):008:0>

[elct9620 (ZhengXian Qiu)]

After a quick trace of the source code, the problem is BigDecimal is a subclass of Numeric but defined in Ruby.

The nurat_conver method check for argument type and convert to Rational, but the Numeric doesn't.
Ref: https://github.com/ruby/ruby/blob/5275d8bf4c43db9f057d24a26cf33ecd69f8b345/rational.c#L2632

When it calls the f_div method, it calculates it directly without converting them to Rational.

We can simply force convert the result, but I didn't think it is the best way:

return to_rational(f_div(a1, a2));

[mrkn (Kenta Murata)]

• Status changed from Open to Assigned

I'm working on it at https://github.com/ruby/ruby/pull/3702

• Status changed from Assigned to Closed

Applied in changeset git|69837229d74813c807e9d079aaee33aec2c7a4d0.

---

rational.c: convert a numerator to rational before calling fdiv in Kernel.Rational() (#3702)

This makes Rational(BigDecimal(1), 60) == Rational(1, 60).
[Bug #16518]