Getting error while making delete operator private

[Premal]

Hi,

I tried to make delete operator private for my class. Strangely it is
giving me error if I compile that code in VC++.NET. But it compiles
successfully on VC++6.o. Can anybody give me inputs about it. I wanted
that on my class delete should not work. Object pointer should be
deleted using my function only which is taking care of reference count
for particular class.

Thanx in advance for your inputs.

Regards,
Premal Panchal

 

[Jim Langston]

Hi,

I tried to make delete operator private for my class. Strangely it is
giving me error if I compile that code in VC++.NET. But it compiles
successfully on VC++6.o. Can anybody give me inputs about it. I wanted
that on my class delete should not work. Object pointer should be
deleted using my function only which is taking care of reference count
for particular class.

Thanx in advance for your inputs.

Please show what you tried that didn't work.

 

[Premal]

Please show what you tried that didn't work.

Hi,
I tried following one:

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*);
public:
//methods
};
if i have above class implementation.Then in VC++6.0 it works. You can
allocate memory using new but you cannot delete that pointer in your
clilent code. You have to use some method provided by above class to
release the pointer.

Same thing doesnt work in VC++.NET. It clearly throws error that
making delete private cause memory leakage. May be VC++.NET compiler
become more stirct about this. .....

I hope you got my point. I hope you can give me some valuable input.

Regards,
Premal Panchal

 

[Ron Natalie]

Hi,
I tried following one:

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*);
public:
//methods
};
if i have above class implementation.Then in VC++6.0 it works. You can
allocate memory using new but you cannot delete that pointer in your
clilent code. You have to use some method provided by above class to
release the pointer.

Same thing doesnt work in VC++.NET. It clearly throws error that
making delete private cause memory leakage. May be VC++.NET compiler
become more stirct about this. .....

First off, that is not the delete operator (despite it's unfortunate
name) it is the memory deallocation function. Compilers are free to
issue whatever diagnostics they want over the required ones.

Why did you do this?

 

[James Kanze]

I tried following one:

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*);
public:
//methods};

if i have above class implementation.Then in VC++6.0 it works. You can
allocate memory using new but you cannot delete that pointer in your
clilent code. You have to use some method provided by above class to
release the pointer.

Same thing doesnt work in VC++.NET. It clearly throws error that
making delete private cause memory leakage. May be VC++.NET compiler
become more stirct about this. .....

The problem is that the new operator must call delete if the
constructor throws. So you can't use new outside of a member
function if operator delete is private. Since this behavior was
added during standardization, some older compilers (e.g. VC++
6.0) might not implement it.

The solution, of course, is to provide factory functions as
well: static member functions which do the new.

 

[James Kanze]

First off, that is not the delete operator (despite it's unfortunate
name) it is the memory deallocation function.

Yes, but access control for the delete operator is done on the
memory deallocation function which is associated with it.

In this case, of course, what's bothering him is that the access
control for the new operator also checks the memory deallocation
function (because the new operator must call the memory
deallocation function if the constructor exits via an
exception).

Compilers are free to
issue whatever diagnostics they want over the required ones.

In this case, the compiler is required to issue the diagnostic.
§5.3.4:

If the new-expression creates an object or an array of
objects of class type, access and ambiguity control are
done for the allocation function, the deallocation
function (12.5), and the constructor (12.1). If the new
expression creates an array of objects of class type,
access and ambiguity control are done for the destructor
(12.4).

Why did you do this?

As he said, he didn't want any one else to delete the class.
(Think about it for a moment. For most classes that you
allocate dynamically, anything but a delete this would be an
error.)

 

[Jim Langston]

Hi,
I tried following one:

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*);
public:
//methods
};
if i have above class implementation.Then in VC++6.0 it works. You can
allocate memory using new but you cannot delete that pointer in your
clilent code. You have to use some method provided by above class to
release the pointer.

Same thing doesnt work in VC++.NET. It clearly throws error that
making delete private cause memory leakage. May be VC++.NET compiler
become more stirct about this. .....

I hope you got my point. I hope you can give me some valuable input.

#include <iostream>

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*) { ::delete this };
public:
void Kill() { delete this; }
//methods};
};

int main()
{
RefCountImpl Foo;
RefCountImpl* Bar = new RefCountImpl;
Bar->Kill(); // Works
delete Bar; // Doesn't work

return 0;

}

Show how you are trying to use it. It does not fail for me unless I try to
delete the instance, are you trying to do this? Are you trying to make it
on the stack? How are you using it? Show the code that is actualy causing
the error when you compile.

 

[Premal]

#include <iostream>

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*) { ::delete this };
public:
void Kill() { delete this; }
//methods};

};

int main()
{
RefCountImpl Foo;
RefCountImpl* Bar = new RefCountImpl;
Bar->Kill(); // Works
delete Bar; // Doesn't work

return 0;

}

Show how you are trying to use it. It does not fail for me unless I try to
delete the instance, are you trying to do this? Are you trying to make it
on the stack? How are you using it? Show the code that is actualy causing
the error when you compile.- Hide quoted text -

- Show quoted text -

Hi Jim,

Sorry for replying late. I was out of the town. I may be not available
for two days.. But just let me know whether you have tried on VC++6.0
or VC++.NET. In case of VC++6.0 above one is working. But its not
working in VC++.NET. May be as James has mentioned that according new
standard it may be the problem.

Actually I have to work on both VC++6.0 and VC++.NET. Once I reach at
my place I will post you exact error. But it is giving error at
compile time itself in VC++.NET while making delete operator private.

Really thanx to both of you JIM and JAMES. Thanx for your input. If
you find something interesting about it then let me know. As I told
you I am implementing class which is extension to auto_ptr and it
works on reference count basis. If possible I will try to post you
whole design of that.

I hope you post soon something intereting about it.

Regards,
Premal

 

[Premal]

#include <iostream>

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*) { ::delete this };
public:
void Kill() { delete this; }
//methods};

};

int main()
{
RefCountImpl Foo;
RefCountImpl* Bar = new RefCountImpl;
Bar->Kill(); // Works
delete Bar; // Doesn't work

return 0;

}

Show how you are trying to use it. It does not fail for me unless I try to
delete the instance, are you trying to do this? Are you trying to make it
on the stack? How are you using it? Show the code that is actualy causing
the error when you compile.- Hide quoted text -

- Show quoted text -

Hi,

Jim and James can you give your email address so I can also email you
document about my development around Reference count object
implementation.

Regards,
Premal Panchal

 

[Ron Natalie]

In this case, the compiler is required to issue the diagnostic.
§5.3.4:

If the new-expression creates an object or an array of
objects of class type, access and ambiguity control are
done for the allocation function, the deallocation
function (12.5), and the constructor (12.1). If the new
expression creates an array of objects of class type,
access and ambiguity control are done for the destructor
(12.4).

Well, he didn't show any code that actually invoked new or the
allocation function, so it wasn't clear that such a diagnostic
was required.

 

[Premal]

Well, he didn't show any code that actually invoked new or the
allocation function, so it wasn't clear that such a diagnostic
was required.

Hi Ron,

Look at the example given by Jim. It was the compilation error in my
case. I invoke it in similar fashion, but error was that as delete
can't be private. So I wanted why it is so. It comes only in .NET not
in VC++6.0.

I hope you got my point.

Regards,
Premal Panchal

 

[James Kanze]

#include <iostream>

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*) { ::delete this };
public:
void Kill() { delete this; }
//methods};
};

int main()
{
RefCountImpl Foo;
RefCountImpl* Bar = new RefCountImpl;

The above line shouldn't work.

The original standard wasn't 100% clear about this; if the
constructor of RefCountImpl terminates with an exception, then
the code here must call delete, but in this case, the compiler
can easily determine that it cannot terminate with an exception.
Some compilers used this information to avoid requiring that
operator delete was accessible, and others didn't. (And some
older compilers, like I think VC++ 6.0, didn't even bother with
the delete -- a constructor terminates with an exception, and
you leak memory.) I believe in fact that there was a defect
report concerning this. At any rate, the current draft
explicitly says that delete must be accessible here, even if the
compiler can determine that the constructor can't possibly
throw. (It's a logical choice, because otherwise, how much
analysis should the compiler do to determine whether the
constructor can throw or not.)

 

[James Kanze]

[...]

Jim and James can you give your email address so I can also email you
document about my development around Reference count object
implementation.

Mine's in my .sig, but as soon as my current data base request
works, I'm going off line for about three weeks vacation. With
no computer anywhere in sight.

 

[Premal]

[...]

Jim and James can you give your email address so I can also email you
document about my development around Reference count object
implementation.

Mine's in my .sig, but as soon as my current data base request
works, I'm going off line for about three weeks vacation. With
no computer anywhere in sight.

--
James Kanze (GABI Software) email:[email protected]
Conseils en informatique orientée objet/
Beratung in objektorientierter Datenverarbeitung
9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34

Hi,

Sorry folks for troubling you. Finally I have reached at my place and
i aksed complete code from my junior. Mistake was he has just declared
the delete operator but didnt implement it.

Thats why we were getting error that delete is in private mode and
memory cannot be freed.

Sorry folks. Now I have changed the code and my design is working.
Thanks for your support.

Regards,
Premal Panchal

 

[Premal]

[...]

Jim and James can you give your email address so I can also email you
document about my development around Reference count object
implementation.

Mine's in my .sig, but as soon as my current data base request
works, I'm going off line for about three weeks vacation. With
no computer anywhere in sight.

--
James Kanze (GABI Software) email:[email protected]
Conseils en informatique orientée objet/
Beratung in objektorientierter Datenverarbeitung
9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34

Hi,

Sorry folks for troubling you. Finally I have reached at my place and
i aksed complete code from my junior. Mistake was he has just declared
the delete operator but didnt implement it.

Thats why we were getting error that delete is in private mode and
memory cannot be freed.

Sorry folks. Now I have changed the code and my design is working.
Thanks for your support.

Regards,
Premal Panchal- Hide quoted text -

- Show quoted text -

Hi,
Again sorry.
I did mistake. I ran that on VC++6.0... On VC++ .NET its still giving
me error. If I use "new" on RefCountImpl then it clearly says that "
operator::delete : cant access protected member". I dont know why this
strange things.

Sorry again, So pls start thinking about it and let me know if any
fruitful explaination is there or not.

If anybody has latest compiler other than .NET then also kindly try
the code given by JIM and see whether same kind of error there or
not.

Also Jim let me know which compiler you have used for your code
snippet.

Thanx and Regards,
Premal Panchal

 

[Jim Langston]

#include <iostream>

class RefCountImpl
{
private:
//data members
protected:
void operator delete(void*) { ::delete this };
public:
void Kill() { delete this; }
//methods};
};

int main()
{
RefCountImpl Foo;
RefCountImpl* Bar = new RefCountImpl;

The above line shouldn't work.

The original standard wasn't 100% clear about this; if the
constructor of RefCountImpl terminates with an exception, then
the code here must call delete, but in this case, the compiler
can easily determine that it cannot terminate with an exception.
Some compilers used this information to avoid requiring that
operator delete was accessible, and others didn't. (And some
older compilers, like I think VC++ 6.0, didn't even bother with
the delete -- a constructor terminates with an exception, and
you leak memory.) I believe in fact that there was a defect
report concerning this. At any rate, the current draft
explicitly says that delete must be accessible here, even if the
compiler can determine that the constructor can't possibly
throw. (It's a logical choice, because otherwise, how much
analysis should the compiler do to determine whether the
constructor can throw or not.)

=====

Well, it compiled in Microsoft VC++ .net 2003. Perhaps it won't compile in
2005.

 

[Premal]

The above line shouldn't work.

The original standard wasn't 100% clear about this; if the
constructor of RefCountImpl terminates with an exception, then
the code here must call delete, but in this case, the compiler
can easily determine that it cannot terminate with an exception.
Some compilers used this information to avoid requiring that
operator delete was accessible, and others didn't. (And some
older compilers, like I think VC++ 6.0, didn't even bother with
the delete -- a constructor terminates with an exception, and
you leak memory.) I believe in fact that there was a defect
report concerning this. At any rate, the current draft
explicitly says that delete must be accessible here, even if the
compiler can determine that the constructor can't possibly
throw. (It's a logical choice, because otherwise, how much
analysis should the compiler do to determine whether the
constructor can throw or not.)

=====

Well, it compiled in Microsoft VC++ .net 2003. Perhaps it won't compile in
2005.- Hide quoted text -

- Show quoted text -

Yes Jim,
May by you are right. It is not getting compiled. But interesting
thing is in VC++6.0 if you make delete operator protected and just
throw exception in constructor. And in your client code under try
catch block make the array of RefImplCount. In this case if you have
thrown exception after making two objects in array then it is calling
delete and hence the destructor of the class. So its very surprising
that why it is now restricted in VC++.NET 2005.

I hope you are getting my point.I will check version of .NET at my
place also.I am not able to find your email address. If you can
provide it then I will send my code to you so may be you can give
better suggestion.

I really appreciated response you have given. Thanx a lot for this.

Regards,
Premal Panchal