Is there any reason for not being able to have double as a parameter type with templates?
For example:
template<int N>//-< Legal
struct
{
};
template<double N>//-< Illegal
struct
{
};
Is there any update on this in C++11?
asked Nov 3, 2010 at 9:11
There is nothing we can do
24.9k32 gold badges114 silver badges203 bronze badges
-
Also see this related question.Prasoon Saurav– Prasoon Saurav2010-11-03 09:23:29 +00:00Commented Nov 3, 2010 at 9:23
Add a comment
|
3 Answers
This is to do with precision, floating point numbers cannot be precisely represented, and the likelyhood of you referring to the same type can depend on how the number is represented. Consider instead using an integer mantissa and exponent as the template parameters...
7 Comments
There is nothing we can do
@Nim but how would you do (your way) and represent 1.02?
Björn Pollex
@There: You could wrap the constant in a
struct and use that as a parameter for the template.
Nim
@Space, not sure that is that @There wants, X<double_struct> is significantly different to X<double N> - unless I'm missing something...
There is nothing we can do
@Space yes but, and I repeat myself, how would you go with 1.02?
Nim
@There, pretty much... internally this is then a trivial calculation,
double v = static_cast<double>(M) * ::exp10(static_cast<double>(E)); |
Given that floating point arithmetics only give fuzzy results (sqrt(2.0)*sqrt(2.0) might not be equal 2.0), how do you propose using double as template arguments could be useful? If you had a template
template<double D> // not allowed in C++
class X {};
and specialize that
template<>
class X<2.0> {};
then
X<1.0+1.0> x;
might not refer to the specialization.
I'd say that limits its usefulness so severely that I'd call it crippled.
2 Comments
Nim
I would hazard that X<1+1> would infact be of type X<int N>! ;) we know you mean X<1.0 + 1.0> x; ! ;)
Jon Purdy
1.0 + 1.0 is guaranteed to be 2.0 in any floating-point implementation. Something like 0.6 + 1.4 would not be, as those numbers are not representable as non-repeating fractions in binary.You can only use integral types for template parameters but you can cheat your way into using fractional values by using two integers, one as mantissa and one as exponent or one as numerator and one as denominator. With clever meta-programming technique you can possibly even reduce them to their lowest terms so if you use rationals then X<3,6> will work the same as X<1,2>
What you can do and what you should do are not necessarily the same. Consider whether this is really the right practical approach or just an exercise in meta-programming.
