polymorphism and protected help

[xxx]

I'm having a little trouble understanding why a derivative class cannot
access a protected member of the base class in the following code:


#include <stdio.h>

class CBase
{
protected:
int x;
public:
CBase () {x = 123;}
public:
operator int () {return x;}
};

class CDerived : public CBase
{
public:
CDerived ()
{
CBase* p = this;
p->x = 456; // error: cannot access protected member
}
};

void main ()
{
printf ("CBase=%d\nCDerived=%d\n", CBase (), CDerived ());
}


I don't see why CDerived shouldn't be able to access the integer "x" in
CBase without having CBase to declare CDerived as a friend. Can someone
explain to me why this is? Thank you.

 

[Dietmar Kuehl]

class CBase
{
protected:
int x;
};

class CDerived : public CBase
{
public:
CDerived ()
{
CBase* p = this;
p->x = 456; // error: cannot access protected member

In this context the compiler does not know that 'p' is actually
of type 'CDerived'. Thus, it assumes that you are possibly
playing with the private parts of a sibling. I doubt that this
would be appreciated in real life and is consequently also not
allowed in C++. If I remember correctly, authors of other
languages have a different attitude towards this...

BTW:

void main ()

The above declaration is illegal according to the C standard:
'main()' has to return an 'int'.

{
printf ("CBase=%d\nCDerived=%d\n", CBase (), CDerived ());
}

Implicit conversions are not applied when passing arguments to
a variable parameter list.

I don't see why CDerived shouldn't be able to access the integer "x" in
CBase without having CBase to declare CDerived as a friend. Can someone
explain to me why this is? Thank you.

Actually, this question is also answered in the FAQ.

 

[Rolf Magnus]

I'm having a little trouble understanding why a derivative class cannot
access a protected member of the base class in the following code:


#include <stdio.h>

class CBase
{
protected:
int x;
public:
CBase () {x = 123;}
public:
operator int () {return x;}
};

class CDerived : public CBase
{
public:
CDerived ()
{
CBase* p = this;
p->x = 456; // error: cannot access protected member

Just write:
x = 456;

}
};

void main ()

main() must return int.

{
printf ("CBase=%d\nCDerived=%d\n", CBase (), CDerived ());

You cannot pass objects through variable argument list, and the compiler
doesn't know that it has to use your operator int(), so you have to cast
your objects to int before giving them to printf:

printf ("CBase=%d\nCDerived=%d\n", static_cast<int>(CBase ()),
static_cast<int>(CDerived ()));

Or just use cout (#include <iostream> at the top):

std::cout << "CBase=" << CBase() << "\nCDerived=" << CDerived() << "\n";

 

[Matthias =?ISO-8859-1?Q?K=E4ppler?=]

#include <stdio.h>

I know that's legal, but shouldn't this rather be #include <cstdio> ?

Just curious.

Cheers,
Matthias

 

[Dietmar Kuehl]

I know that's legal, but shouldn't this rather be #include <cstdio> ?

Just curious.

Probably, <stdio.h> is the better alternative: the reality is that
effectively no C++ standard library declares the C functions in
namespace 'std' and then imports them in the ".h" header via using
directives. Typically, it is done the other way around, i.e. the
names are defined in the global namespace and then made available
in namespace 'std'. Unfortunately, this causes some potential errors
to go undetected, e.g.:

#include <cstdio>
int main() { printf("hello, world\n"); }

The above program will compile with several different C++ library
implementations - but not with a standard conforming one. As a
consequence, it is a safer approach to use <stdio.h> in the first
place because a program compiling with this header on either a
broken or conforming library implementation will also compile on
the other one (where I make, of course, certain assumption about
the broken implementations...).

 

[Ricky Corsi]

xxx wrote:




Just write:
x = 456;




main() must return int.




You cannot pass objects through variable argument list, and the compiler
doesn't know that it has to use your operator int(), so you have to cast
your objects to int before giving them to printf:

printf ("CBase=%d\nCDerived=%d\n", static_cast<int>(CBase ()),
static_cast<int>(CDerived ()));

Sorry, why do you say that the compiler is not able to implicitly
convert a classed passed as a parameter of a function??

I wrote this simple code and both compiles and works. Can you explain me
what you mean? thanks_ricky

#include <iostream>

using namespace std;

class Converse
{
public:
Converse(int init = 0) : x(init) {};
operator int () {return x;};

private:
int x;
};

int summa(int x, int y)
{
return x+y;
}


int main()
{
Converse c1(7);
int val = 3;
int result = summa(val, c1);

cout << result << endl; // this prints exactly 10

return 0;
}

 

[Rolf Magnus]

Sorry, why do you say that the compiler is not able to implicitly
convert a classed passed as a parameter of a function??

I didn't say that. I said that it doesn't know which type a function expects
as part of a _variable_ argument list and therefore doesn't know it has to
convert the argument in the above code to int.

I wrote this simple code and both compiles and works.

That code doesn't use a variable argument list.

 

[Matthias =?ISO-8859-1?Q?K=E4ppler?=]

The above program will compile with several different C++ library
implementations - but not with a standard conforming one.

That doesn't really make sense.
From what I know, the <cxyz> headers are the ISO-C++ counterparts to the
classic xyz.h headers and were introduced in the standardization process to
make clear which are C headers and which are not. It's actually highly
encouraged to use them whenever you want to include C headers.

This is directly copied from <cstdio>:

//
// ISO C++ 14882: 27.8.2 C Library files
//

/** @file cstdio
* This is a Standard C++ Library file. You should @c #include this file
* in your programs, rather than any of the "*.h" implementation files.
*
* This is the C++ version of the Standard C Library header @c stdio.h,
* and its contents are (mostly) the same as that header, but are all
* contained in the namespace @c std.
*/

Regards,
Matthias

 

[xxx]

The intention was to have access to CBase's protected member from CDerived
without having to write a bunch of public accessors in CBase--it would
defeat the purpose to be protected.

Based on the assumption that the compiler preserves the location of CBase in
CDerived:


#include <stdio.h>

class CBase
{
protected:
int x;
public:
CBase () {x = 123;}
public:
operator int () {return x;}
};

class CDerived : public CBase
{
public:
CDerived () {x = 456;}
public:
void mymethod (CBase* p_base)
{
// p->x = x * 666; // error: cannot access protected member

CDerived* p_semi_derived = static_cast<CDerived*>(p_base); //
blasphemy!
p_semi_derived->x = 777;
}
};

void main ()
{
CBase obj_A;
CDerived obj_B;

printf ("CBase=%d\nCDerived=%d\n", static_cast<int>(obj_A),
static_cast<int>(obj_B));
obj_B.mymethod (&obj_A);
printf ("CBase=%d\nCDerived=%d\n", static_cast<int>(obj_A),
static_cast<int>(obj_B));
}


/*
output:

CBase=123
CDerived=456
CBase=777
CDerived=456
*/


But this will break between different implementations of compilers and
perhaps classes with different orderings of base classes--a.k.a. a hack job.

My intention was to have a base class with numerous protected data members
that only derived classes can access and modify. Some nodes of an AST parse
tree may need to modify other node attributes based on some type. I thought
it would be clear to make data members protected because I have multiple
parsers in the same project. If we take a look at "mymethod(...)" from the
above code, then it would represent some derivation of AST Node trying to
modify another derivation of AST Node. Using friends is quite painful the
same way as it is not safe to give away private access when not absolutely
necessary.

Has anyone any other ideas? (Please link me to a URL if there's something I
should read.) Much appreciated!

 

[Ricky Corsi]

Ricky Corsi wrote:




I didn't say that. I said that it doesn't know which type a function expects
as part of a _variable_ argument list and therefore doesn't know it has to
convert the argument in the above code to int.

Oh right, I see now what you mean. I didn't pay attention to the
function (printf) you were referring to...
thanks_ricky

 

[Dietmar Kuehl]

That doesn't really make sense.

It doesn't? Well, I think it really does: the replacement headers were
a nice idea which did not work out in practise. The reality is that
the C++ library implementer has no control over the ".h" versions but
is required to make sure that certain definitions from these headers
are made in namespace 'std'. The only available approach short of
keeping both versions in sync (which is not viable at all) is
something like this:

| // <cstdio>
| namespace std {
| #include "/usr/include/stdio.h"
| }

| // <stdio.h>
| #include <cstdio>
| using std:rintf;
| // ...

(assuming the C version can be included from the file
/usr/include/stdio.h). However, this does not work because there are
loads of things defined in the standard C headers which are unknown
to either the C or the C++ standard and which other headers, e.g.
<unistd.h> rely upon being available in the global namespace.

The intention of putting away the declarations of the C library into
namespace 'std' was all noble - but it simply is not workable under
realistic conditions. Effectively, most implementations actually use
an approach like this:

| // <cfile>
| #include <stdio.h>
| namespace std {
| using :rintf;
| // ...
| }

.... and leave <stdio.h> unchanged (well, not really: it is still
necessary to add a few overloads to some of the headers e.g. for
const correctness but these declarations can easily be bolted on).

From what I know, the <cxyz> headers are the ISO-C++ counterparts to the
classic xyz.h headers and were introduced in the standardization process to
make clear which are C headers and which are not. It's actually highly
encouraged to use them whenever you want to include C headers.

Yup, I know. Actually, I have spread this bad recommendation myself
in the past. At this time I was convinced that I can implement the C++
headers in a conforming way - and I can, as long as you don't want to
include any non-standard headers like <unistd.h>, too (well, actually,
<ctime> really gave me a headache when trying to encapsulate glibc's
<time.h> for some reason). Since the non-standard headers are
generally necessary in real live in some translation units, you can
only implement conforming C++ <c...> headers if you also have
control over the ".h" headers or at least their contents. I'm not
aware of any C++ library vendor who really has.

This is directly copied from <cstdio>:

//
// ISO C++ 14882: 27.8.2 C Library files
//

/** @file cstdio
* This is a Standard C++ Library file. You should @c #include this file
* in your programs, rather than any of the "*.h" implementation files.
*
* This is the C++ version of the Standard C Library header @c stdio.h,
* and its contents are (mostly) the same as that header, but are all
* contained in the namespace @c std.
*/

I know that not all C++ library implementations are able to tag
all C definitions into namespace 'std': that's an easy one for
someone who has implemented most of the standard C++ library and
whose implementation does not do it for practical reasons. I'm
sure that my implementation is not the only one. Actually, there
is an open issue concerning this exact stuff (see
<http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-active.html#456>).

Although I recommended differently in the past, I recommend that
you use the ".h" version for the reasons given before.

 

[Matthias =?ISO-8859-1?Q?K=E4ppler?=]

I wasn't aware of this problem. Thanks.

It doesn't? Well, I think it really does: the replacement headers were
a nice idea which did not work out in practise. The reality is that
the C++ library implementer has no control over the ".h" versions but
is required to make sure that certain definitions from these headers
are made in namespace 'std'. The only available approach short of
keeping both versions in sync (which is not viable at all) is
something like this:

| // <cstdio>
| namespace std {
| #include "/usr/include/stdio.h"
| }

| // <stdio.h>
| #include <cstdio>
| using std:rintf;
| // ...

(assuming the C version can be included from the file
/usr/include/stdio.h). However, this does not work because there are
loads of things defined in the standard C headers which are unknown
to either the C or the C++ standard and which other headers, e.g.
<unistd.h> rely upon being available in the global namespace.

The intention of putting away the declarations of the C library into
namespace 'std' was all noble - but it simply is not workable under
realistic conditions. Effectively, most implementations actually use
an approach like this:

| // <cfile>
| #include <stdio.h>
| namespace std {
| using :rintf;
| // ...
| }

... and leave <stdio.h> unchanged (well, not really: it is still
necessary to add a few overloads to some of the headers e.g. for
const correctness but these declarations can easily be bolted on).


Yup, I know. Actually, I have spread this bad recommendation myself
in the past. At this time I was convinced that I can implement the C++
headers in a conforming way - and I can, as long as you don't want to
include any non-standard headers like <unistd.h>, too (well, actually,
<ctime> really gave me a headache when trying to encapsulate glibc's
<time.h> for some reason). Since the non-standard headers are
generally necessary in real live in some translation units, you can
only implement conforming C++ <c...> headers if you also have
control over the ".h" headers or at least their contents. I'm not
aware of any C++ library vendor who really has.


I know that not all C++ library implementations are able to tag
all C definitions into namespace 'std': that's an easy one for
someone who has implemented most of the standard C++ library and
whose implementation does not do it for practical reasons. I'm
sure that my implementation is not the only one. Actually, there
is an open issue concerning this exact stuff (see
<http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-active.html#456>).

Although I recommended differently in the past, I recommend that
you use the ".h" version for the reasons given before.

 

[Dietmar Kuehl]

The intention was to have access to CBase's protected member from CDerived
without having to write a bunch of public accessors in CBase--it would
defeat the purpose to be protected.

A member function of a derived class has access to the protected members
of objects it knows to be derived. It does not have access to protected
members of sibling classes. I think this is a reasonable approach.
Otherwise, the protection would be rather thin: a could easily derive
from base and access protected members from static functions of his
derived class without ever even creating an object! This would be
practically useless. If you need to access base members of sibling
classes, you effectively need public access.