overloading global dereference operator?

[TuxC0d3]

Hi!

I'm diving into the some more ++ specific aspects of c++ (and finally
accepting that c++ is more than "a plus added to c" , so that means
using namespaces, templates, std::strings, lists, vectors, operator
overloading and what not.. And i was wondering if there is a way to
override the global dereference operator, so to be able to check the
address that one tries to dereference. This gives the ability to throw
an exception when one would try to dereference NULL for instance.

I have tried several different ways to do it, but gcc doesn't like what
i'm asking of him...

This attempt is the most reasonable to me:
43 template<class T> T& operator * (T* ptr)
44 {
45 if (!ptr)
46 throw EAccessViolation();
47
48 return (*ptr);
49 }
(sorry about the line numbers.. Just the vim settings on my console)

Even though this seems right to me, g++ gives me the following
complaint when i compile that like thus:
[tim@server tests]$ g++ -Wall -o test operatornew.cpp > c.log 2>&1
[tim@server tests]$ cat c.log
operatornew.cpp:44: error: `T& operator*(T*)' must have an argument of
class or
enumerated type
[tim@server tests]$

Could it be that the compiler gets confused and/or thinks that i want
to overide the multiply operator? Or am i just beeing totaly clueless
here?

I have searched for this for what feels like all around the net, but i
can't seem to find anything like it. Everything that comes up about
dereference operator overloading is the operator->() for inside a
class.. I have also tried putting this global, which naturaly didn't
work becouse -> is obviously only used for members..

Can anyone give me a lead on this and/or provide some example code?
Tnx!

Tim.

 

[John Carson]

Hi!

I'm diving into the some more ++ specific aspects of c++ (and finally
accepting that c++ is more than "a plus added to c" , so that means
using namespaces, templates, std::strings, lists, vectors, operator
overloading and what not.. And i was wondering if there is a way to
override the global dereference operator, so to be able to check the
address that one tries to dereference. This gives the ability to throw
an exception when one would try to dereference NULL for instance.

I have tried several different ways to do it, but gcc doesn't like
what i'm asking of him...

This attempt is the most reasonable to me:
43 template<class T> T& operator * (T* ptr)
44 {
45 if (!ptr)
46 throw EAccessViolation();
47
48 return (*ptr);
49 }
(sorry about the line numbers.. Just the vim settings on my console)

Even though this seems right to me, g++ gives me the following
complaint when i compile that like thus:
[tim@server tests]$ g++ -Wall -o test operatornew.cpp > c.log 2>&1
[tim@server tests]$ cat c.log
operatornew.cpp:44: error: `T& operator*(T*)' must have an argument of
class or
enumerated type
[tim@server tests]$

Could it be that the compiler gets confused and/or thinks that i want
to overide the multiply operator? Or am i just beeing totaly clueless
here?

I have searched for this for what feels like all around the net, but i
can't seem to find anything like it. Everything that comes up about
dereference operator overloading is the operator->() for inside a
class.. I have also tried putting this global, which naturaly didn't
work becouse -> is obviously only used for members..

Can anyone give me a lead on this and/or provide some example code?
Tnx!

You can't overload operators that involve only built in types. At least one
operand must be of class/struct/enum type. Pointers are considered built in
types, even if they point to user-defined types. Thus you cannot overload
any operator that works exclusively on pointers. For example, the following
won't compile:

#include <iostream>

struct S
{};

void operator+(S*lptr, S*rptr)
{
std::cout << "New plus pointers operator\n";
}

 

[TuxC0d3]

Hmm, ok that sounds reasonable..

However, it would be nice if it could be made to work, for instance by
some kind of class-with-operator-overloading technique.. (Like wrapping
a pointer inside a class to make it a "smart" pointer, or such a
thing..)
That might make it work a decent bit slower, but when building in debug
mode, it could provide me with a lot of helpful information...

I'm thinking about something like
template(class T) class ptr
{
T* m_ptr;
public:
//... add default constructor, copy ctor and operator=, etc here
T& operator*()
{
if (!m_ptr)
throw EAccessViolation();
return *m_ptr;
}
};

Wouldn't something like that be able to solve it as well, or am i
probably thinking along the wrong lines? (for instance that this
wouldn't give me an object which acts like a pointer, but gives me a
ptr::* (or something like that))

Don't be afraid to speak tech by the way.. I've already got my pretty
little callback technique (based on a class with () operator
overloading, a template constructor and pointer-to-memberfunction)
working like a charm in only a couple of hours with almost no
experience in operator overloading, templates, etc etc. (Though i
do understand quite some things about working with pointers)

 

[John Carson]

Hmm, ok that sounds reasonable..

However, it would be nice if it could be made to work, for instance by
some kind of class-with-operator-overloading technique.. (Like
wrapping a pointer inside a class to make it a "smart" pointer, or
such a thing..)
That might make it work a decent bit slower, but when building in
debug mode, it could provide me with a lot of helpful information...

I'm thinking about something like
template(class T) class ptr
{
T* m_ptr;
public:
//... add default constructor, copy ctor and operator=, etc here
T& operator*()
{
if (!m_ptr)
throw EAccessViolation();
return *m_ptr;
}
};

Wouldn't something like that be able to solve it as well, or am i
probably thinking along the wrong lines? (for instance that this
wouldn't give me an object which acts like a pointer, but gives me a
ptr::* (or something like that))

Sure, wrapping a pointer in a class will give you something pretty close to
what you want. The following (quickly composed) class seems to work OK,
though it is still missing some needed operators (I haven't defined a copy
constructor or assignment operator since the compiler supplied ones seem
adequate):

#include <iostream>
using namespace std;

struct S
{
int x;
};


class EAccessViolation
{};

template<class T> class ptr
{
T* m_ptr;
public:
ptr() : m_ptr(0)
{}
ptr(T *rawptr) : m_ptr(rawptr)
{}
bool operator==(ptr &rhs)
{
return m_ptr == rhs.m_ptr;
}
T*operator->() const
{
if (!m_ptr)
throw EAccessViolation();
return m_ptr;
}
T& operator*() const
{
if (!m_ptr)
throw EAccessViolation();
return *m_ptr;
}
};


int main()
{
S s1,s2;
ptr<S> p1=&s1, p2, p3;

// p1 is initialised but p2 is not so should be able to
// dereference p1 OK but get an exception from p2

p3 = p1;
if(p1==p3)
cout << "p1 and p3 are equal\n";
else
cout << "p1 and p3 are unequal\n";


try
{
p1->x = 5;
cout << "Dereference of p1 successful\n";
}
catch(EAccessViolation&)
{
cout << "Access violation for p1\n";
}
try
{
p2->x = 5;
cout << "Dereference of p2 successful\n";
}
catch(EAccessViolation&)
{
cout << "Access violation for p2\n";
}
try
{
s2 = *p1;
cout << "Dereference successful for p1\n";
}
catch(EAccessViolation&)
{
cout << "Access violation for p1\n";
}
try
{
s2 = *p2;
cout << "Dereference successful for p2\n";
}
catch(EAccessViolation&)
{
cout << "Access violation for p2\n";
}
}

Note that a wrapped pointer is never quite the same as a raw pointer.
Consider:

struct Test
{
Test(int *pint)
{}
};

void foo(const Test &test)
{}

int main()
{
int *pint1;
foo(pint1); // will compile

ptr<int> pint2;
foo(pint2); // won't compile since two user defined
//conversions required
}

 

[Greg]

Sure, wrapping a pointer in a class will give you something pretty close to
what you want. The following (quickly composed) class seems to work OK,
though it is still missing some needed operators (I haven't defined a copy
constructor or assignment operator since the compiler supplied ones seem
adequate):

#include <iostream>
using namespace std;

struct S
{
int x;
};


class EAccessViolation
{};

template<class T> class ptr
{
T* m_ptr;
public:
ptr() : m_ptr(0)
{}
ptr(T *rawptr) : m_ptr(rawptr)
{}
bool operator==(ptr &rhs)
{
return m_ptr == rhs.m_ptr;
}
T*operator->() const
{
if (!m_ptr)
throw EAccessViolation();
return m_ptr;
}
T& operator*() const
{
if (!m_ptr)
throw EAccessViolation();
return *m_ptr;
}

Some implementations would elect to add these methods:

operator T*()
{
return m_ptr;
}

operator const T*() const
{
return m_ptr;
}

};


Note that a wrapped pointer is never quite the same as a raw pointer.
Consider:

struct Test
{
Test(int *pint)
{}
};

void foo(const Test &test)
{}

int main()
{
int *pint1;
foo(pint1); // will compile

ptr<int> pint2;
foo(pint2); // won't compile since two user defined
//conversions required

With the conversion methods added, there is one user-defined conversion
from ptr<int> to int* so the above function call will compile and work
correctly.

Although adding these conversion methods makes the smart pointer
virtually indistinguishable from a raw pointer, it can cause problems
if the smart pointer is maintaining a reference count. The implicit
conversion allows the smart pointer to be assigned to a raw pointer
without incrementing the reference count. For this (and other reasons)
an implicit conversion to the corresponding raw pointer is usually not
provided, or done so conditionally. And any client wanting to perform
such a conversion must do so explicitly.

Greg

 

[John Carson]

>
Some implementations would elect to add these methods:

operator T*()
{
return m_ptr;
}

operator const T*() const
{
return m_ptr;
}

With the conversion methods added, there is one user-defined
conversion from ptr<int> to int* so the above function call will
compile and work correctly.

Actually, no. In fact my comments implicitly assumed exactly that operator
(I should have made this explicit). foo does not take a int * argument; it
takes a reference to Test. Thus there are two conversions required:

1. int<ptr> to int *
2. int * to Test

Accordingly, the code will not compile.

Of course, one might not want the conversion from int<ptr> to Test, but that
is another issue. The issue at hand is the equivalence of int * and ptr<int>
and the foo function illustrates a non-equivalence.

 

[John Carson]

Thus there are two
conversions required:
1. int<ptr> to int *
2. int * to Test

That should be ptr<int>, not int<ptr>