[Rd] Modulus Bug (PR#7409)

[Thomas Lumley]

On Thu, 9 Dec 2004 ripley at stats.ox.ac.uk wrote:

> This is nothing to do with integers: 1e18 and 11 are doubles here.
>
> It is a result of rounding error: 1e18/11 is not representable accurately,
> and this should have been a warning to you that your calculations were
> unreasonable.

It was -- the poster originally asked about large integer representations 
and said he was planning to use other software for large integers.

At the time I said that the fact %% didn't give some sort of error or 
warning was a bug, and was planning to return either a warning or NA or 
NaN if
   abs(x1)>1/.Machine$double.eps

 	-thomas

> The C code is
>
> double myfmod(double x1, double x2)
> {
>     double q = x1 / x2;
>     return x1 - floor(q) * x2;
> }
>
> We can improve the answer, but what you are doing is fundamentally flawed
> and it is hard to detect whether rounding error has affected this. A
> warning rather than an error seems appropriate.
>
> If you really want to do things like this, try the gmp package (which
> seems to give the wrong answer here) or a more appropriate calculator.
>
>
> On Thu, 9 Dec 2004 Robert.McGehee at geodecapital.com wrote:
>
>> R Developers,
>>
>> 1000000000000000000 %% 11
>> [1] -32
>>
>> I now understand that integers cannot be larger than
>> .Machine$integer.max, but because the above produces a result than is
>> patently wrong instead of an error, I'm reporting this as a bug.
>
> -- 
> Brian D. Ripley,                  ripley at stats.ox.ac.uk
> Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
> University of Oxford,             Tel:  +44 1865 272861 (self)
> 1 South Parks Road,                     +44 1865 272866 (PA)
> Oxford OX1 3TG, UK                Fax:  +44 1865 272595
>
> ______________________________________________
> R-devel at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>