Feature #3289: Division of negative numbers

• Status: Rejected
• Assignee: matz (Yukihiro Matsumoto)
• Target version: 3.0

[patrick_thomson (Patrick Thomson)]

=begin
The documentation for Numeric (http://www.ruby-doc.org/core/classes/Numeric.html#M000179) states that integer divmod() (and, by extension, /) rounds the quotient towards negative infinity. Python and Tcl behave similarly, while C, Java, bc, and gdb round the quotient towards zero, as is taught in standard arithmetic courses.

Is this a quirk of MRI's implementation, or is it desired Ruby behavior? If so, why?
It's counterintuitive that (-x/y) ≠ -(x/y), and even moreso when (-x/y) = -(x/y) if x or y is a non-integer.
=end

[shyouhei (Shyouhei Urabe)]

=begin
Japanese elementary schools tend to teach their children that "x / y is z remainder w" means

 x == y * z + w, where 0 <= w < y.

So we follow it.
=end

[mame (Yusuke Endoh)]

• Assignee set to matz (Yukihiro Matsumoto)
• Priority changed from Normal to 3

=begin
Hi,

2010/5/14 Patrick Thomson [e-mail]:

The documentation for Numeric (http://www.ruby-doc.org/core/classes/Numeric.html#M000179) states that integer divmod() (and, by extension, /) rounds the quotient towards negative infinity. Python and Tcl behave similarly, while C, Java, bc, and gdb round the quotient towards zero, as is taught in standard arithmetic courses.

It is very various. According to:

"Integer modulo operators in various programming languages"
http://en.wikipedia.org/wiki/Modulo_operation

J, Lua, Mathematica, Excel, Oberon, Perl, PL/I, Python, R, Ruby,
Smalltalk, Tcl and Turing seem to behave the same.

I bet Ruby's division is taken after Perl.
Anyway, it is too late to change.

--
Yusuke Endoh [e-mail]
=end

[shyouhei (Shyouhei Urabe)]

• Status changed from Open to Assigned

=begin

=end

[akr (Akira Tanaka)]

• Project changed from Ruby to Ruby master
• Category changed from core to core

[alexeymuranov (Alexey Muranov)]

I think that division with always rounding towards 0 is not logical. Rounding towards the closest integer, going towards 0 if there are 2 integers at equal distance, sounds better to me. This would be symmetric with respect to exchanging x and -x. The current behavior/definition of '/' is also an important operation. Maybe it can be renamed to 'div' and return both the quotient and the remainder in the "integral" sense (quotient being rounded towards -\infty)?

Update: Sorry, i have not read the documentation! So divmod does exactly what one would expect it to do. The division '/' for integers is indeed questionable. I would suggest that it raise an exception or return some special value if two integers do not divide evenly. Apart from being nice in mathematical sense, this can help to avoid unexpected bugs.

-Alexey.

[mame (Yusuke Endoh)]

• Description updated (diff)
• Target version set to 3.0

Given the context, Ruby choice seems right to me. But the real core of the problem is
the minus sign. My father says, that minus simply represents an unfinished operation,
intention to subtract. Citing anonymous Wikipedia contributor:

"Negative numbers appeared for the first time in history in the Nine Chapters on
the Mathematical Art, which in its present form dates from the period of the
Chinese Han Dynasty (202 BC. - AD 220), but may well contain much older material.
Indian mathematicians developed consistent and correct rules on the use of negative
numbers, which later spread to the Middle East and then into Europe. Prior to the
concept of negative numbers, negative solutions to problems were considered "false"
and equations requiring negative solutions were described as absurd (Diophantus's
Arithmetica cited)."

My father thinks, that Greek adherence to natural numbers, inherited by medieval Europe,
was not due to their backwardness (they eg. knew that π was irrational), but because
complicated objects catering to daily needs of bookkepers were bad elementary blocks
for general-purpose applied mathematics.

I would like to see basic math rebuilt without reaching for minus (and zero) too early.
Relevant Ruby proposal would be: have true natural numbers (class Natural), have morally
sound rationals (natural/ natural), and positive-only floats. Will I try to prototype
this myself one day? Regarding zero, which has similar issues as minus and is absent
from true natural numbers. I wonder how far would one get using nil (or null?) in its
stead, like ancient Romans did...

[mame (Yusuke Endoh)]

• Status changed from Assigned to Rejected

I believe this ticket is hopeless. Closing.