How to avoid non-standard behavior in taking the address of a standard library function

Reading about Designated addressable functions, I realized that taking the address of a standard library function is undefined behavior.

Now, I do not know how to rewrite my code that always take the address of standard functions.

#include <iostream>
#include<vector>
#include<cmath>
bool divisible_3(double n){
	return int(n)%3==0;
}
void applyFunction(std::vector<double>& vector, bool(* filterf)(double), double(*calc)(double)){
	for(int index=0;index<int(vector.size());index++){
		if(filterf(vector[index])){
			vector[index]=calc(vector[index]);
		}
	}
}
int main(){
	
	std::vector<double> vect = {1, 2, 3, 5, 9, 16, 21, 24};
	
	applyFunction(vect,divisible_3,std::sqrt);
	
	for(auto&& i: vect){
		std::cout << i << std::endl;
	}
}

There are a few choices. The first is the most obvious: write your own delegator function. std::sqrt might be a non-addressable function, but you know what is addressable?

double totally_not_std_sqrt(double input) {
  return std::sqrt(input);
}

This function is entirely addressable, so it can be passed in as an argument in completely standard C++.

The other, more compact option is to use a lambda.

applyFunction(vect, divisible_3, [](double x) { return std::sqrt(x); });

Now you might be thinking "lambdas are convertible to std::function, but I take a function pointer". That's true. But lambdas that don't capture are convertible to function pointers, so even though your function takes a function pointer, it can still take non-capturing lambdas. (You might consider having your function take std::function though, unless you're planning on interfacing with C, but that's another issue)