Printing floating points

[Afgdr]

On Saturday, 30 January 2021 at 11:02:54 UTC, Afgdr wrote:
> On Thursday, 28 January 2021 at 23:27:28 UTC, Ola Fosheim 
> Grøstad wrote:
>> I do however refute the idea that an exhaustive test is 
>> needed, since the problem involves regular patterns.
>
> how about a full coverage of a half float type, assuming the 
> function tested works the same with float and double and anay 
> arbitrary lond FP type, e.g template param for mantissa and 
> exp. Sof if a smaller FP type fully passes the for loop test, 
> e.g all permutations we can assume bigger FP types will be 100% 
> correctly converted.

Well I forgot that it may use LUTs so maybe not in this case, but 
otherwise you should get the idea.