I have a class that holds some statically-sized containers:
template <typename Container>
struct Point {
Container container;
...
void bar();
}
where a Container
class might look like:
struct Container1 {
static constexpr size_t size = 5;
}
Now I want to specialize the bar
method based on the size of the container. I do not understand how to do that.
EDIT:
I would like a C++11 solution. C++14 might work, but the compilers we work with typically have spotty C++14 support.
EDIT:
Stack Danny suggested a solution which compiles with Clang, but not with GCC.
Instead of specialisation, use SFINAE
template <typename Container>
class Point {
Container container;
template<size_t S> std::enable_if_t<S==3>
bar_t() { std::cout << "specialisation for size=3\n"; }
template<size_t S> std::enable_if_t<S==5>
bar_t() { std::cout << "specialisation for size=5\n"; }
template<size_t S> std::enable_if_t<S==42>
bar_t() { std::cout << "specialisation for size=42\n"; }
public:
void bar()
{ bar_t<Container::size>(); }
};
std::enable_if_t
is C++14, but you can trivially declare it yourself:
#if __cplusplus < 201402L
template<bool C, typename T=void>
using enable_if_t = typename enable_if<C,T>::type;
#endif
Btw, your question smells like an XY problem: do you really need to specialize bar()
for the Container::size
? In the following example, a loop is unrolled for any size N
.
template<typename scalar, size_t N>
class point // a point in R^N
{
scalar array[N];
public:
// multiplication with scalar
point& operator*=(scalar x) noexcept
{
// unroll loop using template meta programming
loop([array,x](size_t i) { array[i] *= x; };);
/* alternatively: rely on the compiler to do it
for(size_t i=0; i!=N; ++i)
array[i] *= x;
*/
return *this;
}
private:
template<size_t I=0, typename Lambda>
static enable_if_t<(I<N)> loop(Lambda &&lambda)
{
lambda(I);
loop<I+1>(lambda);
}
template<size_t I=0, typename Lambda>
static enable_if_t<(I>=N)> loop(Lambda &&) {}
};
It is called template specialization and works as follows:
#include <cstdint>
#include <iostream>
struct Container1 {
static constexpr size_t size = 5;
};
struct Container2 {
static constexpr size_t size = 3;
};
template <typename Container>
struct Point {
Container container;
void bar() {
this->bar_templated<Container::size>();
}
private:
template<std::size_t size>
void bar_templated() {
std::cout << "default\n";
}
template<>
void bar_templated<3>() {
std::cout << "specialized <3>\n";
}
template<>
void bar_templated<5>() {
std::cout << "specialized <5>\n";
}
};
int main(){
Point<Container1> p1;
p1.bar();
Point<Container2> p2;
p2.bar();
}
output
specialized <5>
specialized <3>
due to bug 85282 in gcc making it impossible to compile an explicit specialization in non-namespace scope (thanks @songyuanyao), there are errors:
25:14: error: explicit specialization in non-namespace scope 'struct Point'
26:27: error: template-id 'bar_templated<3>' in declaration of primary template
...
30:10: error: 'void Point::bar_templated()' cannot be overloaded
But you can workaround this by moving the functions out of the class and still achieve specialization:
template<std::size_t size>
void bar_templated() {
std::cout << "default\n";
}
template<>
void bar_templated<3>() {
std::cout << "specialized 3\n";
}
template<>
void bar_templated<5>() {
std::cout << "specialized 5\n";
}
template <typename Container>
struct Point {
Container container;
void bar() {
bar_templated<Container::size>();
}
};
This way the functions are public, which may not be what you want, though. Well, if you're writing inside a header file you could define them inside an anonymous namespace.
Also: if constexpr - but that's only C++17 and forward. It reducing the code size by alot and keeping it's logical nature makes it the best approach here, for sure.
void bar() {
if constexpr (Container::size == 3) {
std::cout << "specialized <3>\n";
}
else if constexpr (Container::size == 5) {
std::cout << "specialized <5>\n";
}
}
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With