What's the standard say about this code?

[Daniel T.]

#include <cassert>

class Foo {
public:
virtual void fnA() = 0;
virtual void fnB() = 0;
};

int main() {
assert( &Foo::fnB );
assert( &Foo::fnA );
}

What does the standard say about the above code? In the compiler I'm
using now, the first assert will not fire, but the second one will. I
expected that neither assert would fire...

 

[Maxim Yegorushkin]

#include <cassert>

class Foo {
public:
        virtual void fnA() = 0;
        virtual void fnB() = 0;

};

int main() {
        assert( &Foo::fnB );
        assert( &Foo::fnA );

}

What does the standard say about the above code? In the compiler I'm
using now, the first assert will not fire, but the second one will. I
expected that neither assert would fire...

Are you sure that the second assert is failing?

Both &Foo::fnB and &Foo::fnA should yield a non-NULL member function
pointer, which gets converted to bool(true) when passed into assert().

 

[Daniel T.]

Are you sure that the second assert is failing?

Both &Foo::fnB and &Foo::fnA should yield a non-NULL member function
pointer, which gets converted to bool(true) when passed into assert().

Yes, I am sure that the second assert is failing. If this is a
compiler bug, then I will submit it to the vendor, but I want to make
sure it actually *is* a compiler bug first.

 

[James Kanze]

#include <cassert>

class Foo {
public:
virtual void fnA() = 0;
virtual void fnB() = 0;
};

int main() {
assert( &Foo::fnB );
assert( &Foo::fnA );
}

What does the standard say about the above code?

There should be no problem with it.

In the compiler I'm using now, the first assert will not fire,
but the second one will. I expected that neither assert would
fire...

It's guaranteed by the standard. It works with the three
compilers I have access to (Sun CC, g++ and VC++), at least when
you compile in standard conformant mode. (Note that by default,
pointers to member functions do not work in VC++. You must use
the option /vmg. Not that I think that their non-conformity
otherwise would play a role here.)

 

[Triple-DES]

Can I get chapter and verse from the standard on this? It sounds like I
need to submit a bug report to the compiler vender.

I believe it is disallowed by 4.11/1 [conv.mem]:

"A null pointer constant (4.10) can be converted to a pointer to
member type; the result is the null member pointer value of that type
and is distinguishable from any pointer to member not created from a
null pointer constant.(...)"

Therefore, an rvalue obtained by using address-of on a member shall
not yield a null pointer constant.

 

[Maxim Yegorushkin]

Can I get chapter and verse from the standard on this? It sounds like I
need to submit a bug report to the compiler vender.

I believe it is disallowed by 4.11/1 [conv.mem]:

"A null pointer constant (4.10) can be converted to a pointer to
member type; the result is the null member pointer value of that type
and is distinguishable from any pointer to member not created from a
null pointer constant.(...)"

Therefore, an rvalue obtained by using address-of on a member shall
not yield a null pointer constant.

In the original question a member function pointer gets converted to
bool. Is 4.11/1 applicable here?

 

[Erik Wikström]

#include <cassert>

class Foo {
public:
virtual void fnA() = 0;
virtual void fnB() = 0;
};

int main() {
assert( &Foo::fnB );
assert( &Foo::fnA );
}

What does the standard say about the above code?

There should be no problem with it.

In the compiler I'm using now, the first assert will not fire,
but the second one will. I expected that neither assert would
fire...

It's guaranteed by the standard. It works with the three
compilers I have access to (Sun CC, g++ and VC++), at least when
you compile in standard conformant mode. (Note that by default,
pointers to member functions do not work in VC++. You must use
the option /vmg. Not that I think that their non-conformity
otherwise would play a role here.)

Can I get chapter and verse from the standard on this? It sounds like I
need to submit a bug report to the compiler vender.

I believe it is disallowed by 4.11/1 [conv.mem]:

"A null pointer constant (4.10) can be converted to a pointer to
member type; the result is the null member pointer value of that type
and is distinguishable from any pointer to member not created from a
null pointer constant.(...)"

Therefore, an rvalue obtained by using address-of on a member shall
not yield a null pointer constant.

In the original question a member function pointer gets converted to
bool. Is 4.11/1 applicable here?

No, 4.12 is probably the correct section:

"An rvalue of arithmetic, enumeration, pointer, or pointer to member
type can be converted to an rvalue of type bool. A zero value, null
pointer value, or null member pointer value is converted to false; any
other value is converted to true."

 

[Triple-DES]

#include <cassert>
class Foo {
public:
        virtual void fnA() = 0;
        virtual void fnB() = 0;
};
int main() {
        assert( &Foo::fnB );
        assert( &Foo::fnA );
}
What does the standard say about the above code?
There should be no problem with it.
In the compiler I'm using now, the first assert will not fire,
but the second one will. I expected that neither assert would
fire...
It's guaranteed by the standard.  It works with the three
compilers I have access to (Sun CC, g++ and VC++), at least when
you compile in standard conformant mode.  (Note that by default,
pointers to member functions do not work in VC++.  You must use
the option /vmg.  Not that I think that their non-conformity
otherwise would play a role here.)
Can I get chapter and verse from the standard on this? It sounds like I
need to submit a bug report to the compiler vender.
I believe it is disallowed by 4.11/1 [conv.mem]:
"A null pointer constant (4.10) can be converted to a pointer to
member type; the result is the null member pointer value of that type
and is distinguishable from any pointer to member not created from a
null pointer constant.(...)"
Therefore, an rvalue obtained by using address-of on a member shall
not yield a null pointer constant.

In the original question a member function pointer gets converted to
bool. Is 4.11/1 applicable here?

No, 4.12 is probably the correct section:

"An rvalue of arithmetic, enumeration, pointer, or pointer to member
type can be converted to an rvalue of type bool. A zero value, null
pointer value, or null member pointer value is converted to false; any
other value is converted to true."

Of course, but the problem is that taking the address of a function
yields a null member pointer value, not the fact that it is
subsequently converted to false. The OP might as well have written:

assert( &Foo::fnB != 0);
assert( &Foo::fnA != 0);

And 4.12 would be completely irrelevant, but the result of the second
assert would still violate 4.11/1

DP