Motivation:
Almost for fun, I am trying to write a function overload that can tell apart whether the argument is a fixed-size array or a pointer.
double const d[] = {1.,2.,3.};
double a;
double const* p = &a;
f(d); // call array version
f(p); // call pointer version
I find this particularly difficult because of the well known fact that arrays decay to pointer sooner than later. A naive approach would be to write
void f(double const* c){...}
template<size_t N> void f(double const(&a)[N]){...}
Unfortunately this doesn't work. Because in the best case the compiler determines the array call f(d)
above to be ambiguous.
Partial solution:
I tried many things and the closest I could get was the following concrete code. Note also that, in this example code I use char
instead of double
, but it is very similar at the end.
First, I have to use SFINAE to disable conversions (from array ref to ptr) in the pointer version of the function. Second I had to overload for all possible arrays sizes (manually).
[compilable code]
#include<type_traits> // for enable_if (use boost enable_if in C++98)
#include<iostream>
template<class Char, typename = typename std::enable_if<std::is_same<Char, char>::value>::type>
void f(Char const* dptr){std::cout << "ptr" << std::endl;} // preferred it seems
void f(char const (&darr)[0] ){std::cout << "const arr" << std::endl;}
void f(char const (&darr)[1] ){std::cout << "const arr" << std::endl;}
void f(char const (&darr)[2] ){std::cout << "const arr" << std::endl;}
void f(char const (&darr)[3] ){std::cout << "const arr" << std::endl;}
void f(char const (&darr)[4] ){std::cout << "const arr" << std::endl;}
void f(char const (&darr)[5] ){std::cout << "const arr" << std::endl;}
void f(char const (&darr)[6] ){std::cout << "const arr" << std::endl;} // this is the one called in this particular example
// ad infinitum ...
int main(){
f("hello"); // print ptr, ok because this is the fixed size array
f(std::string("hello").c_str()); // print arr, ok because `c_str()` is a pointer
}
This works, but the problem is that I have to repeat the function for all possible values of N
and using template<size_t N>
gets me back to square zero, because with the template parameter the two calls get back to equal footing.
In other words, template<size_t N> void f(char const(&a)[N]){std::cout << "const arr" << std::endl;}
doesn't help.
Is there any way to generalize the second overload without falling back to an ambiguous call? or is there some other approach?
A C++ or C++1XYZ answer is also welcome.
Two details: 1) I used clang
for the experiments above, 2) the actual f
will end up being an operator<<
, I think know if that will matter for the solution.
Summary of solutions (based on other people's below) and adapted to a the concrete type char
of the example. Both seem to depend on making the char const*
pointer less obvious for the compiler:
template<class Char, typename = typename std::enable_if<std::is_same<Char, char>::value>::type>
void f(Char const* const& dptr){std::cout << "ptr" << std::endl;}
template<size_t N>
void f(char const (&darr)[N] ){std::cout << "const arr" << std::endl;}
template<class CharConstPtr, typename = typename std::enable_if<std::is_same<CharConstPtr, char const*>::value>::type>
void f(CharConstPtr dptr){std::cout << "ptr" << std::endl;}
template<size_t N>
void f(char const (&darr)[N] ){std::cout << "const arr" << std::endl;}
This seems to work for me
#include <iostream>
template<typename T>
std::enable_if_t<std::is_pointer<T>::value>
foo(T) { std::cout << "pointer\n"; }
template<typename T, std::size_t sz>
void foo(T(&)[sz]) { std::cout << "array\n"; }
int main()
{
char const* c;
foo(c);
foo("hello");
}
Bonus std::experimental::type_traits
:
using std::experimental::is_pointer_v;
std::enable_if_t<is_pointer_v<T>>
Your comment made me try something even simpler
template<typename T> void foo(T) { std::cout << "pointer\n"; }
template<typename T, unsigned sz> void foo(T(&)[sz]) { std::cout << "array\n"; }
Of course the problem here is that foo
is now callable for any type, depends on how lax you want your parameter checking to be.
One other way is to (ab)use rvalue references
void foo(char const*&) { std::cout << "pointer\n"; }
void foo(char const*&&) { std::cout << "array\n"; }
Obviously it's not foolproof.
You may use the following:
namespace detail
{
template <typename T> struct helper;
template <typename T> struct helper<T*> { void operator() () const {std::cout << "pointer\n";} };
template <typename T, std::size_t N> struct helper<T[N]> { void operator() ()const {std::cout << "array\n";} };
}
template <typename T>
void f(const T& )
{
detail::helper<T>{}();
}
Live example
I like using tag dispatching:
void foo(char const*, std::true_type /*is_pointer*/) {
std::cout << "is pointer\n";
}
template<class T, size_t N>
void foo( T(&)[N], std::false_type /*is_pointer*/) {
std::cout << "is array\n";
}
template<class X>
void foo( X&& x ) {
foo( std::forward<X>(x), std::is_pointer<std::remove_reference_t<X>>{} );
}
live example
Here is a simple solution that exploits the fact that in C the value of an array is equal to its address while this is generally not true for a pointer.
#include <iostream>
template <typename P>
bool is_array(const P & p) {
return &p == reinterpret_cast<const P*>(p);
}
int main() {
int a[] = {1,2,3};
int * p = a;
std::cout << "a is " << (is_array(a) ? "array" : "pointer") << "\n";
std::cout << "p is " << (is_array(p) ? "array" : "pointer") << "\n";
std::cout << "\"hello\" is " << (is_array("hello") ? "array" : "pointer");
}
Note that while a pointer normally points to a location different from itself, this is not necessarily guaranteed; indeed you can easily fool the code by doing something like this:
//weird nasty pointer that looks like an array:
int * z = reinterpret_cast<int*>(&z);
However, since you are coding for fun, this can be a fun, basic, first approach.
In my opinion it is as simple as this:
#include<iostream>
using namespace std;
template<typename T>
void foo(T const* t)
{
cout << "pointer" << endl;
}
template<typename T, size_t n>
void foo(T(&)[n])
{
cout << "array" << endl;
}
int main() {
int a[5] = {0};
int *p = a;
foo(a);
foo(p);
}
I don't get all the complication with std::enable_if
and std::true_type
and std::is_pointer
and magic. If there's anything wrong with my approach please tell me.
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