Re: [boost] [safe_numerics] Review Part 2 (Implementation)

Subject: Re: [boost] [safe_numerics] Review Part 2 (Implementation)
From: Andrzej Krzemienski (akrzemi1_at_[hidden])
Date: 2017-03-07 19:33:03

2017-03-07 18:32 GMT+01:00 Robert Ramey via Boost <boost_at_[hidden]>:

247:
>>         using result_base_type = typename boost::mpl::if_c<
>>             std::numeric_limits<t_base_type>::is_signed
>>             || std::numeric_limits<u_base_type>::is_signed,
>>             std::intmax_t,
>>             std::uintmax_t
>>         >::type;
>> Why division doesn't use calculate_max_t like multiplication
>> deserves some explanation.  Division can overflow
>> too (numeric_limits<uintmax_t>::max() / 1).
>>
>
> Hmmm - I'm sure it's because I've presumed that the result
> can aways be contained in the same size type as the number
> being divide.  That t / x <= t - and divide by zero is
> checked separately. Am I wrong about this?
>

I am not sure if this is what Steve meant, but:

  int x = INT_MIN;
  int y = -1;
  int a = x / y; // overflows!

safe<int> does a check for this condition, but I find it quite surprising
that it throws a domain_error rather than overflow_error. How do you make a
call whether it is an overflow or a domain error? For mathematical ints
dividing -2147483648 by -1 is well defined.

Regards,
&rzej;