Polymorphic Function Pointers

[Ben Pope]

Hi all,

This is not something I've played around with much, but I'm designing
some factories and I want a function like this:

template<class T>
T* Creator() {
return new T;
}

So that I could store a pointer to an instance in a map along with a
string Id.

Now, the things I want to return would be polymorphic, and so I would
make use of covariant return types. The problem I ran into was
something like the following:

struct Base { virtual ~Base(){} };
struct Derived : Base {};

typedef Base* (*BaseFuncPtr)( );
typedef Derived* (*DerivedFuncPt r)();

Base* CreateBase() { return new Base; }
Derived* CreateDerived() { return new Derived; }

int main()
{
// Covariant return types
Base* b(CreateDerived ());

// Fine
BaseFuncPtr bfp(&CreateBase );

// Cannot convert?
BaseFuncPtr dfp(&CreateDeri ved);
}
Although the function pointers are to different types, the type being
returned is polymorphic and covariant. Can somebody just explain why
this doesn't work? Am I expecting too much?

Cheers for any insights,

Ben Pope
--
I'm not just a number. To many, I'm known as a string...

[Tomás]

I compiled your code with G++ and here's what I got:

g++ creator.cpp -ansi -pedantic -o creator.exe

creator.cpp: In function `int main()':
creator.cpp:19: warning: invalid conversion from `Derived*(*)()' to `Base*
(*)()
It compiled without errors but gave a warning.
Should it give a warning, should it give an error... seems like a quote from
the Standard is in order.
-Tomás

[Victor Bazarov]

Ben Pope wrote:

This is not something I've played around with much, but I'm designing
some factories and I want a function like this:

template<class T>
T* Creator() {
return new T;
}

So that I could store a pointer to an instance in a map along with a
string Id.

Now, the things I want to return would be polymorphic, and so I would
make use of covariant return types. The problem I ran into was
something like the following:

struct Base { virtual ~Base(){} };
struct Derived : Base {};

typedef Base* (*BaseFuncPtr)( );
typedef Derived* (*DerivedFuncPt r)();

Base* CreateBase() { return new Base; }
Derived* CreateDerived() { return new Derived; }

int main()
{
// Covariant return types
Base* b(CreateDerived ());

// Fine
BaseFuncPtr bfp(&CreateBase );

// Cannot convert?
BaseFuncPtr dfp(&CreateDeri ved);
}
Although the function pointers are to different types, the type being
returned is polymorphic and covariant. Can somebody just explain why
this doesn't work?
WHAT doesn't work? Your attempt to declare a pointer to a function that
takes no args and returns A and initialise it with an address of another
function that takes no args and returns B? They are of different types.
Unrelated.

B func(); // 'func' is a function that takes no args and returns B
A (*pf)() = func; // 'pf': ptr to a func, no args, returns A
// *cannot* be initialised with _anything_ except
// another func that takes no args, returns A

Covariance applies _only_ to virtual functions, and not pointers to
functions.
Am I expecting too much?

Absolutely.

V
--
Please remove capital As from my address when replying by mail

[Victor Bazarov]

Tomás wrote:

I compiled your code with G++ and here's what I got:

g++ creator.cpp -ansi -pedantic -o creator.exe

creator.cpp: In function `int main()':
creator.cpp:19: warning: invalid conversion from `Derived*(*)()' to `Base*
(*)()
It compiled without errors but gave a warning.
Should it give a warning, should it give an error... seems like a quote from
the Standard is in order.

From 4.3: "An lvalue of function type T can be converted to an rvalue
of type “pointer to T.” The result is a pointer to the function."

Since return value type is part of the function type, if functions have
different return value types, their respective types T1 and T2 are
different, hence a function of type T1 cannot be converted to a pointer
to [function type] T2.

V
--
Please remove capital As from my address when replying by mail

[Ben Pope]

Victor Bazarov wrote:

Covariance applies _only_ to virtual functions, and not pointers to
functions.

Yes, I have just realised that this is the difference between
overloading and overriding.

> Am I expecting too much?

Absolutely.

Of course.

Thanks Victor.

Ben Pope
--
I'm not just a number. To many, I'm known as a string...

[Alf P. Steinbach]

* Victor Bazarov:

Ben Pope wrote:

struct Base { virtual ~Base(){} };
struct Derived : Base {};

typedef Base* (*BaseFuncPtr)( );
typedef Derived* (*DerivedFuncPt r)();

Base* CreateBase() { return new Base; }
Derived* CreateDerived() { return new Derived; }

int main()
{
// Covariant return types
Base* b(CreateDerived ());

// Fine
BaseFuncPtr bfp(&CreateBase );

// Cannot convert?
BaseFuncPtr dfp(&CreateDeri ved);
}
[snip] Am I expecting too much?

Absolutely.

Unfortunately, yes. One would expect standard C++ to support this, and
there is no technical reason I'm aware why it couldn't. But it doesn't.

The support for covariance in C++ is half-baked, and there is no support
for contravariance -- we lack data flow direction designators such as
C# and Ada 'out'.

Although there is the general problem of argument forwarding, for a
given function of known signature something like the following can be a
practical solution:

struct Base { virtual ~Base(){} };
struct Derived : Base {};

typedef Base* (*BaseFuncPtr)( );
typedef Derived* (*DerivedFuncPt r)();

Base* CreateBase() { return new Base; }
Derived* CreateDerived() { return new Derived; }

template< typename R, R (*foo)() >
inline Base* asBaseFunction( ) { return foo(); }

int main()
{
// Covariant return types
Base* b(CreateDerived ());

// Fine
BaseFuncPtr bfp(&CreateBase );

// Cannot convert?
BaseFuncPtr dfp( &asBaseFunction <Derived*, &CreateDeriv ed> );
}
--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?