C / C++ / MFC

3-Oct-03 10:24

I have an abstract base class CFoo that contains a pure virtual method void setDefault() as well as a public static method void doSomething(), like so:

class CFoo {
  public:
    // Construction/destruction
    CFoo()
      { setDefault(); }
    virtual ~CFoo()
      {}

    // Subclasses must implement this
    virtual void setDefault() = 0;

    // A public static method
    static void doSomething() {
       int x = 0;
       x = 32;
    }
}

One of CFoo's clients does this:

CFoo::doSomething();

which causes VC6's linker to complain that CFoo::setDefault() is not implemented. It's as if the compiler wants to construct a CFoo object (which makes no sense since it's an abstract base class).

Is it me or VC6? (It's probably me, but I can't figure out what I'm doing wrong). Frown | :(

/ravi

Let's put "civil" back in "civilization"
Home | Articles | Freeware | Music
ravib@ravib.com

Orhun Birsoy3-Oct-03 10:37

Orhun Birsoy

3-Oct-03 10:37

I think it is VC6. Both VC7.1 and Comeau compiled it just fine.

The results I get from Comeau,

MODE:strict errors C++

"ComeauTest.c", line 7: warning: call of pure virtual function
setDefault();
^

"ComeauTest.c", line 16: warning: variable "x" was set but never used
int x = 0;
^

In strict mode, with -tused, Compile succeeded (but remember, the Comeau online compiler does not link).

http://www.comeaucomputing.com/tryitout/[^]

Ravi Bhavnani3-Oct-03 10:42

3-Oct-03 10:42

Thanks.

What's weird is, I went ahead and implemented CFoo::setDefault() keeping the pure virtual specifier, and guess what - VC6 didn't complain! Eeek! OMG | :OMG:

(Nor did the problem go away.)

/ravi

Let's put "civil" back in "civilization"
Home | Articles | Freeware | Music
ravib@ravib.com

Peter Weyzen3-Oct-03 11:26

Peter Weyzen

3-Oct-03 11:26

If no one actually attempts to instantiate a CFoo object, then it won't complain.

If you just do:

CFoo fred;

it will.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~<br />
Peter Weyzen<br />
Staff Engineer<br />
<A HREF="http://www.santacruznetworks.com">Santa Cruz Networks</A>

Ravi Bhavnani3-Oct-03 12:18

3-Oct-03 12:18

Actually, you can never instantiate a CFoo, since it's an abstract class.

The problem has to do with the fact that CFoo's static (factory) method was constructing a derived class which caused the linker to want to resolve CFoo::setDefault(). This is sorta hokey, but what I was doing was wrong, since the derived class's vtable has not been initialized in the base class' constructor.

/ravi

Let's put "civil" back in "civilization"
Home | Articles | Freeware | Music
ravib@ravib.com

Michael Dunn3-Oct-03 13:06

Michael Dunn

3-Oct-03 13:06

It's giving an error because you're calling CFoo::setDefault() directly, and that method is not implemented. Remember the virtual only kicks in if you call the method through a pointer. You need to call plain setDefault() to get the polymorphism that I think you are expecting.

--Mike--
Ericahist | Homepage | RightClick-Encrypt | 1ClickPicGrabber
CP SearchBar v2.0.2 released

Ravi Bhavnani3-Oct-03 13:17

3-Oct-03 13:17

Actually I'm calling CFoo::setDefault() indirectly - I neglected to mention Smile | :)

that doSomething() is actually a factory method that serves up a specialization of a CFoo, which causes CFoo's constructor to be called, which causes a call to a method in a derived class, which is just plain wrong!

But I still think the compiler should have caught the error.

/ravi

Let's put "civil" back in "civilization"
Home | Articles | Freeware | Music
ravib@ravib.com

Michael Dunn3-Oct-03 14:26

Michael Dunn

3-Oct-03 14:26

Ravi Bhavnani wrote:
which causes CFoo's constructor to be called, which causes a call to a method in a derived class

er, you aren't calling a virtual function in the CFoo ctor, are you? The vtbl isn't initialized yet so that will probably crash at runtime with a pure virtual function call.

--Mike--
Ericahist | Homepage | RightClick-Encrypt | 1ClickPicGrabber
CP SearchBar v2.0.2 released

Ravi Bhavnani4-Oct-03 6:11

4-Oct-03 6:11

Quite right, Mike. VC6 catches that link time, presumably because the vtable entry for setDefault() is NULL.

/ravi

Let's put "civil" back in "civilization"
Home | Articles | Freeware | Music
ravib@ravib.com

peterchen3-Oct-03 20:32

peterchen

3-Oct-03 20:32

Michael is right (almost) - you don't have "virtuality" in the constructor. Your doSomething should Create the object, and then call setDefault explicitely.

(almost: IIRC the VTable is initialized to the one ofthe base class, and you have a virtual call not only through a pointer but also in all member functions and the destructor)

"Vierteile den, der sie Hure schimpft mit einem türkischen Säbel."

sighist | Agile Programming | doxygen

Ravi Bhavnani4-Oct-03 6:13

4-Oct-03 6:13

peterchen wrote:
Your doSomething should Create the object, and then call setDefault explicitely.

Yes, that's exactly the tack I took. Thanks for your reply!

/ravi

Let's put "civil" back in "civilization"
Home | Articles | Freeware | Music
ravib@ravib.com

Weird VC++ behaviour

DaFrawg3-Oct-03 9:45

3-Oct-03 9:45

While writing some code for a class, wich also contains oveloads for the assignment operator ( = ), VC++ keeps giving C2801 errors. According to MSDN, C2801 means that assignment, class member access, subscripting or function call operators must be a non-static class member.

This is my code:

class Foo<br />
{<br />
private:<br />
   CString m_String;<br />
public:<br />
   Foo();<br />
   Foo(CString);<br />
   ~Foo();<br />
   friend Foo& operator= (const class Bar);<br />
}<br />
<br />
Foo::Foo (CString string)<br />
{<br />
   m_String = string;<br />
}<br />
<br />
Foo& operator= (const class Bar BarInstance)<br />
{<br />
   return Foo (BarInstance.m_String);<br />
}<br />
<br />
class Bar<br />
{<br />
private:<br />
   CString m_String;<br />
public:<br />
   Bar();<br />
   ~Bar();<br />
}

But that's not the whole story; the code

char* yourName = "DaFrawg";<br />
MessageBox("Hello there, %s", yourName);

will generate a messagebox with the prompt "Hello there, %s" and the caption "DaFrawg". What I want is, of course, to have the message "Hello there, DaFrawg" displayed.

Can somebody help me, please?

//Please don't mind my bad English

Michael P Butler3-Oct-03 10:00

Michael P Butler

3-Oct-03 10:00

DaFrawg wrote:
But that's not the whole story; the code
char* yourName = "DaFrawg";
MessageBox("Hello there, %s", yourName);
will generate a messagebox with the prompt "Hello there, %s" and the caption "DaFrawg". What I want is, of course, to have the message "Hello there, DaFrawg" displayed.

Can somebody help me, please?

CString strMessage;<br />
<br />
strMessage.Format("Hello there, %s", yourName);<br />
MessageBox(strMessage);

Michael

'Logic, my dear Zoe, merely enables one to be wrong with authority.' - The Doctor: The Wheel in Space

DaFrawg10-Oct-03 8:42

10-Oct-03 8:42

Then why doesn't TRACE1("Hello %s", name) fail?

---QUOTE---

"ERROR: Keyboard not present - Press F1 to continue" - Most BOIS chips

Michael P Butler10-Oct-03 9:03

Michael P Butler

10-Oct-03 9:03

DaFrawg wrote:
Then why doesn't TRACE1("Hello %s", name) fail?

Because the TRACE1 function supports variable arguments. MessageBox does not.

Michael

'Logic, my dear Zoe, merely enables one to be wrong with authority.' - The Doctor: The Wheel in Space

DaFrawg27-Oct-03 2:05

27-Oct-03 2:05

Thanks, it works really good.

David Crow3-Oct-03 10:07

David Crow

3-Oct-03 10:07

Try this:

class Bar
{
public:
    Bar();
    ~Bar();
    
    CString m_String;
};
 
class Foo
{
private:
    CString m_String;
 
public:
    Foo();
    Foo(CString);
    ~Foo();
    
    const Foo& operator=( const Bar& bar );
};
 
Foo::Foo (CString string)
{
    m_String = string;
}
 
const Foo& Foo::operator=(const Bar& BarInstance)
{
    m_String = BarInstance.m_String;
 
    return (*this);
}

Five birds are sitting on a fence.
Three of them decide to fly off.
How many are left?

DaFrawg4-Oct-03 2:23

4-Oct-03 2:23

It still doesn't work. I now have the real code I use:

<br />
class Foo<br />
{<br />
public:<br />
	Foo();<br />
	virtual ~Foo();<br />
private:<br />
	CString m_String;<br />
public:<br />
	friend const Foo& operator= (const class Bar& BarInst);    //2x C2801<br />
};<br />
<br />
class Bar<br />
{<br />
public:<br />
	Bar();<br />
	virtual ~Bar();<br />
private:<br />
	CString m_String;<br />
};<br />
<br />
const Foo& operator=(const class Bar& BarInst)   //C2801<br />
{<br />
	m_String = Bar.m_String;   //C2673, C2227 and C2248. Ignore C2248<br />
	return (this*);   //C2673 and C2059<br />
}<br />

The compiler is getting berzerk. Not only a C2801, but also two C2673 (Global functions do not have 'this' pointers), C2227 (left of m_String must point to class/struct/union), C2248 (but that's not a real problem) and C2059 (Syntax error: ';').

David Crow6-Oct-03 2:59

David Crow

6-Oct-03 2:59

Please refer to M. Dimmick's reply. Friend functions are not considered class members; they are normal external functions that are given special access privileges. Also, friends are not in the class’s scope, and they are not called using the member-selection operators (. and –>). The code snippet that I provided you is correct. If it is not the correct solution to your problem, however, then you must reconsider the design.

Five birds are sitting on a fence.
Three of them decide to fly off.
How many are left?

Mike Dimmick3-Oct-03 23:52

Mike Dimmick

3-Oct-03 23:52

It's not weird at all, it's standard: operator= must be implemented as a member function, according to the definition of the language.

The error message also tells you that any overloads of operator->, operator[] or operator() must also be implemented as a member function. I believe it also applies to compound assignments, such as +=, /=, etc.

operator= must be a function that modifies the object to which it's applied, i.e. the left-hand side of the assignment statement, which is pointed to by this inside the function body. It should return a reference to the object that was modified.

DaFrawg27-Oct-03 2:00