I’m trying to write my own logging class and use it as a stream:
logger L;
L << "whatever" << std::endl;
This is the code I started with:
#include <iostream>
using namespace std;
class logger{
public:
template <typename T>
friend logger& operator <<(logger& log, const T& value);
};
template <typename T>
logger& operator <<(logger& log, T const & value) {
// Here I'd output the values to a file and stdout, etc.
cout << value;
return log;
}
int main(int argc, char *argv[])
{
logger L;
L << "hello" << '\n' ; // This works
L << "bye" << "alo" << endl; // This doesn't work
return 0;
}
But I was getting an error when trying to compile, saying that there was no definition for operator<< (when using std::endl):
pruebaLog.cpp:31: error: no match for ‘operator<<’ in ‘operator<< [with T = char [4]](((logger&)((logger*)operator<< [with T = char [4]](((logger&)(& L)), ((const char (&)[4])"bye")))), ((const char (&)[4])"alo")) << std::endl’
So, I’ve been trying to overload operator<< to accept this kind of streams, but it’s driving me mad. I don’t know how to do it. I’ve been loking at, for instance, the definition of std::endl at the ostream header file and written a function with this header:
logger& operator <<(logger& log, const basic_ostream<char,char_traits<char> >& (*s)(basic_ostream<char,char_traits<char> >&))
But no luck. I’ve tried the same using templates instead of directly using char, and also tried simply using “const ostream& os”, and nothing.
Another thing that bugs me is that, in the error output, the first argument for operator<< changes, sometimes it’s a reference to a pointer, sometimes looks like a double reference…
endlis a strange beast. It isn’t a constant value. It’s actually, of all things, a function. You need a special override to handle the application ofendl:This accepts insertion of a function that takes an
ostreamreference and returns anostreamreference. That’s whatendlis.Edit: In response to FranticPedantic’s interesting question of “why can’t the compiler deduce this automatically?”. The reason is that if you delve yet deeper,
endlis actually itself a template function. It’s defined as:That is, it can take any sort of
ostreamas its input and output. So the problem isn’t that the compiler can’t deduce thatT const &could be a function pointer, but that it can’t figure out whichendlyou meant to pass in. The templated version ofoperator<<presented in the question would accept a pointer to any function as its second argument, but at the same time, theendltemplate represents an infinite set of potential functions, so the compiler can’t do anything meaningful there.Providing the special overload of the
operator<<whose second argument matches a specific instantiation of theendltemplate allows the call to resolve.