Mfcttrf

[temp.constr.decl] says that we can constrain a template or a function with a constraint expression.

Declarators [dcl.decl] tells us that, for functions, we may add an optional trailing requires clause to constrain it, and the standard draft n4820 even gives these (seemingly pointless) examples:

void f1(int a) requires true;
auto f2(int a) -> bool requires true;

I understand that constraining a template or a concept is useful, but I fail to see how these constraints are useful for non-templated functions. What's the point of constraining a non-templated function?

One of the main points of constraining non-template functions is to be able to write constraints to non-template members of template classes. For example, you might have some type like this:

template<typename T>
class value

public:
 value(const T& t);
 value(T&& t);

private:
 T t_;
;

Now, you want value to be copyable/moveable from T. But really, you want it to be copyable/moveable from T only as far as T itself is copyable/moveable. So, how do you do it?

Pre-constraints, you would need to write a bunch of meta-programming hackery. Maybe you make these constructors templates, which require that the given type U is the same as T, in addition to the copy/move requirement. Or you might have to write a base class that you inherit from, which has different specializations based on the copy/moveability of T.

Post-constraints, you do this:

class value

public:
 value(const T& t) requires is_copy_constructible_v<T> : t_(t) 
 value(T&& t) requires is_move_constructible_v<T> : t_(std::move(t)) 

private:
 T t_;
;

No hackery. No applying templates to functions that don't need to be templates. It just works, and its easy for the user to understand what is going on.

This is especially important for functions which cannot be templates. In order for a constructor to be considered a copy or move constructor, it cannot be a template. Same goes for copy/move assignment operators. But such things can have constraints.

Just as a concept consider the following example

#include <iostream>

void f( long x ) requires ( sizeof( long ) == sizeof( int ) )

 std::cout << "Bye " << x << 'n';


void f( long long x ) requires ( sizeof( long ) == sizeof( long long ) )

 std::cout << "Hello " << x << 'n';


int main() 

 f( 0l );

If sizeof( long ) == sizeof( long long ) then the program output will be

Hello 0

Otherwise

Bye 0

For example you can use such an approach in a function that calculates the factorial to restrict the number of a loop iterations or to throw an exception.

Here is a demonstrative program.

#include <iostream>
#include <stdexcept>

unsigned long factorial( unsigned long n ) noexcept( false ) 
 requires ( sizeof( unsigned long ) == sizeof( unsigned int ) )

 const unsigned long MAX_STEPS = 12;

 if ( MAX_STEPS < n ) throw std::out_of_range( "Too big value." );

 unsigned long f = 1;

 for ( unsigned long i = 1; i < n; i++ ) f *= ( i + 1 );

 return f;


unsigned long long factorial( unsigned long long n ) noexcept( false ) 
 requires ( sizeof( unsigned long ) == sizeof( unsigned long long ) )

 const unsigned long long MAX_STEPS = 20;

 if ( MAX_STEPS < n ) throw std::out_of_range( "Too big value." );

 unsigned long f = 1;

 for ( unsigned long long i = 1; i < n; i++ ) f *= ( i + 1 );

 return f;


int main() 

 unsigned long n = 20;

 try
 
 std::cout << factorial( n ) << 'n';
 
 catch ( const std::out_of_range &ex )
 
 std::cout << ex.what() << 'n';
 

Its output might be either

2432902008176640000

or

Too big value.