Way of converting between types of different overflow tags

[hbe72]

What is your question?

Is there a way of converting between datatypes with different overflow tags.

What are you trying to do?

One may need to change the overflow type within an algorithm.

TEST(math, convert_overflow_type)                                                                                            
{                                                                                                               
    auto a = cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::native_overflow_tag, int16_t>{0.5};     
    auto b = cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::native_overflow_tag, int16_t>{0.5};     
    auto c = a + b;                                                                                             
    cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::saturated_overflow_tag, int16_t> d = c;          
    std::cout << "d = " << d << std::endl;                                                                      
}  

Going through rep works but is quite cumbersome:

static constexpr auto a = cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::native_overflow_tag, int16_t>{0.5};
static constexpr cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::saturated_overflow_tag, int16_t> b = cnl::from_rep<cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::saturated_overflow_tag, int16_t>, int16_t> {}(static_cast<int16_t>(to_rep(a)));
static constexpr auto expected = cnl::static_number<16,-15, cnl::neg_inf_rounding_tag, cnl::saturated_overflow_tag, int16_t>{0.5};
static_assert(identical(expected, b));

[johnmcfarlane]

There's no such API currently.

I took a look at this yesterday. There is a set_rounding_t alias which does a near-identical thing for rounding_integer. But a quick fix to copy/paste this as set_overflow_t isn't as quite simple as it sounds and merely highlights the fact that there is a more general-purpose solution fighting to get out.

I think that adapting set_tag_t so that it works recursively, is probably the way to go. This will replace set_rounding_t and be used thus:

constexpr auto b{set_tag_t<decltype(a), saturated_overflow_t>{a}};