[Rcpp-devel] int double multiplication

[Andreas Prescher]

Thanks a lot for the helpful answers!!

Andreas


Am 21.06.2016 um 22:19 schrieb Matt D.:
> Hi,
> 
> On 6/21/2016 16:33, Andreas Prescher wrote:
>> void f(double d) {
>>     int i = vector.size() * d;
>> }
> A quick note to add to the other answers: You don't want to be using
> `int` here.
> Especially since you mention a 64-bit platform -- but IMHO it's a bad
> idea in general.
> If you'd like a simple rule of thumb, here goes: Avoid using `int` for
> anything by default -- especially anything involving
> sizes/indexes/counting. Yes, there are exceptional lucky cases when you
> may be able to get away with an `int`, but taking these into account is
> brittle and anything but "simple."
> 
> Your `vector.size()`, assuming you're dealing with `vector` of type
> `std::vector<int>` (as in your later example), has type
> `std::vector<int>::size_type` -- which (on your 64-bit platform) will be
> 64-bits.
> See: http://en.cppreference.com/w/cpp/container/vector/size
> In contrast, type `int` is at most 32-bits:
> http://en.cppreference.com/w/cpp/language/types
> 
> If the value of the expression involving the vector size value is larger
> than what you can store in an `int`, then the size_type->double->int
> conversion you're doing when storing to `i` will cause undefined
> behavior due to overflow:
> http://blog.frama-c.com/index.php?post/2013/10/09/Overflow-float-integer
> 
> While you may (or may not) be in a particularly lucky special case when
> the size value just happens to fit in the 32-bits on one occasion, you
> may not be on another.
> Instead of spending time on deciding "is `int` luckily going to work
> today" (or performing an interval analysis involving making guesses
> about potential input sample sizes / the ranges of `vector.size()` and
> `d`) it's simpler to just use the proper type at all times.
> 
> Now, regarding the rounding and the "adding 0.5" trick, there are some
> subtleties here worth being aware of:
> http://blog.frama-c.com/index.php?post/2013/05/02/nearbyintf1
> 
> If you can use C++11 the right choice would be to use these:
> http://en.cppreference.com/w/cpp/numeric/math/round
> 
> Refs:
> https://randomascii.wordpress.com/2012/04/05/floating-point-complexities/
> https://randomascii.wordpress.com/2012/02/25/comparing-floating-point-numbers-2012-edition/
> 
> 
> Best,
> 
> Matt
> 
> _______________________________________________
> Rcpp-devel mailing list
> Rcpp-devel at lists.r-forge.r-project.org
> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
> 
> .
>