I know if I pass a value, the thread will only work on the copy and has no effect after the thread returns.
No, that's not correct. The code should not silently make a copy and work on the copy, the standard says it must not even compile.
The standard requires that the arguments to the called function are copied (into storage managed by the C++ runtime) and then the copies are forwarded as rvalues. So in your example
f1 gets passed an rvalue of type
double and the parameter of type
double& cannot bind to that rvalue.
The reason the standard requires that is so there is no silent copying and loss of data: if the function requires a modifiable reference then it won't compile unless you pass a reference explicitly using a
The compiler error you get involves
result_of because that's how I made GCC's
std::thread check if the function can be called with the supplied arguments. I use
result_of<decltype(&f1)(double)> to check if the function pointer
&f1 (which is of type
void(*)(double&)) can be called with an rvalue of type
double. It can't be called with an argument of that type, so the nested type
result_of<decltype(&f1)(double)>::type is not defined, so the compiler says:
error: no type named 'type' in 'class std::result_of<void (*(double))(double&)>'
The error is a bit confusing because the C++ declarator rules mean that
decltype(&f1)(double) gets displayed as
That is not my concern. I want to know why it is giving error now, but it was not giving error to other posts on SO
Those posts were using an old pre-C++11 or non-conforming compiler which didn't meet the requirements of the C++11 standard and incorrectly compiled the code.