Basic problem with Inheritance

[Matthias Pfeifer]

Hi there,

i am implementing a Point and Vector class and found that they have some
things in common. Therefore i introduced a common parent class called
ndimobj, that hosts an array of n values, offers some constructors and
the following operator

template<size_t n, typename T>
class ndimobj
{
<SNIP>
public:
ndimobj<n, T>& operator= (const T s);
};

ndimobj is templated by the type of the elements that live in a given "n
dimensional space" (doubles, polynomials, etc...) and the dimension n.

Point is now derived from ndimobj like this

template<size_t n=3, typename T=float>
class Point : public ndimobj<n, T>
{
public:
double distance(Point< n,Tp)
{ <snip}
};

If i do (in a test) program

Point <4, doubleq;
q=3.0;

my compiler (g++) complains:

test.cc: 49: error no match for 'operator=' in 'q=3.0e+0'
Point.hpp: 22: note: candidates are: Point<4u, double>& Point<4u,
double>::operat or=(const Point<4u, double>&)

Why is operator= not known. and why does g++ find this unsuitable
candidate?!
Please let me know if didn't give sufficiently information and i'll post
the complete interface of ndimobj and Point if it helps.

Thanks in advance

matthias

[=?ISO-8859-1?Q?Erik_Wikstr=F6m?=]

On 2007-08-16 11:55, Matthias Pfeifer wrote:

Hi there,

i am implementing a Point and Vector class and found that they have some
things in common. Therefore i introduced a common parent class called
ndimobj, that hosts an array of n values, offers some constructors and
the following operator

template<size_t n, typename T>
class ndimobj
{
<SNIP>
public:
ndimobj<n, T>& operator= (const T s);
};

ndimobj is templated by the type of the elements that live in a given "n
dimensional space" (doubles, polynomials, etc...) and the dimension n.

Point is now derived from ndimobj like this

template<size_t n=3, typename T=float>
class Point : public ndimobj<n, T>
{
public:
double distance(Point< n,Tp)
{ <snip}
};

If i do (in a test) program

Point <4, doubleq;
q=3.0;

my compiler (g++) complains:

test.cc: 49: error no match for 'operator=' in 'q=3.0e+0'
Point.hpp: 22: note: candidates are: Point<4u, double>& Point<4u,
double>::operat or=(const Point<4u, double>&)

Why is operator= not known. and why does g++ find this unsuitable
candidate?!
Please let me know if didn't give sufficiently information and i'll post
the complete interface of ndimobj and Point if it helps.

Because you have not defined the = operator, what exactly do you expect
q=3.0; to do? If this was a point in the mathematical sense what would
that statement mean? Nothing as far as I know, you can't assign a scalar
to a vector (an a point is a vector as far as math is concerned).

The candidate found by gcc is used like this:

Point <4, doublep;
Point <4, doubleq;
p = q;

this = operator is automatically generated by the compiler for your
convenience.

--
Erik Wikström

[Matthias Pfeifer]

Because you have not defined the = operator,

my ndimobj class has "operator= (const T s)" shouldn't class Point have
that inhereted?

template<size_t n, typename T>
class ndimobj
{
<SNIP>
public:
ndimobj<n, T>& operator= (const T s);
};

matt

[LR]

Matthias Pfeifer wrote:

>Because you have not defined the = operator,

my ndimobj class has "operator= (const T s)" shouldn't class Point have
that inhereted?

template<size_t n, typename T>
class ndimobj
{
<SNIP>
public:
ndimobj<n, T>& operator= (const T s);
};

The compiler will generate a Point &operator=(cons t Point &).
Consider this code:

class A {
public:
A &x(const int x) {
return *this;
}
};

class B : public A {
public:
B &x(const B &x) {
return *this;
}
};

int main() {
B q;
B p;
q.x(1); // error
q.x(p);
}

LR

[=?ISO-8859-1?Q?Erik_Wikstr=F6m?=]

On 2007-08-16 13:31, Matthias Pfeifer wrote:

>Because you have not defined the = operator,

my ndimobj class has "operator= (const T s)" shouldn't class Point have
that inhereted?

template<size_t n, typename T>
class ndimobj
{
<SNIP>
public:
ndimobj<n, T>& operator= (const T s);
};

Sorry, missed that. No it will not inherit that, you should get the same
error when trying to compile this:

struct A {
A& operator=(int i_);
};

struct B : public A {
};

int main() {
B b;
b = 1;
}

--
Erik Wikström

[Frank Birbacher]

Hi!

Matthias Pfeifer schrieb:

Why is operator= not known.

There is something called "hiding" base class members. This is what is
happening here. Because the Point class has the automatically generated
"operator = (Point const&)" for self assignment, the base class
"operator =" is hidden. You can make it visible again by a "using"
declaration:

//in class Point
using ndimobj<n, T>::operator =;

HTH,
Frank

[Matthias]

Frank Birbacher schrieb:

Hi!

Matthias Pfeifer schrieb:

>Why is operator= not known.

There is something called "hiding" base class members. This is what is
happening here. Because the Point class has the automatically generated
"operator = (Point const&)" for self assignment, the base class
"operator =" is hidden. You can make it visible again by a "using"
declaration:

//in class Point
using ndimobj<n, T>::operator =;

HTH,
Frank

thank you all for your answers. Using

using ndimobj<n, T>::operator =;

my compiler is satisfied - i am however not. I know about copy
constructors, which should be the automagically generated "Point<n,
T>::operator= (Point const&)". I am confused, because "my" operator is
"ndimobj<n, T>::operator= (const T)". Where i like to point that he has
a completely different argument (not const Point&, but const T). Please.
Why is the operator hidden?

sincerely
Matthias

[BobR]

LR <lr***@superlin k.netwrote in message...

>
class A {
public:
A &x(const int x) {
return *this;
}
};

class B : public A {
public:
B &x(const B &x) {
return *this;
}
};

int main() {
B q;
B p;
q.x(1); // error
q.x(p);
}

LR

Thanks for the spaces vs. tabs. Looks like it should now (indented). <G>

--
Bob R
POVrookie

[Frank Birbacher]

Hi!

Matthias schrieb:

my compiler is satisfied - i am however not. I know about copy
constructors, which should be the automagically generated "Point<n,
T>::operator= (Point const&)".

In fact, this is not a "constructo r". It is the (automatically
generated" "assignment operator". The copy constructor is
"Point<n,T>::Po int(Point const&)".

I am confused, because "my" operator is
"ndimobj<n, T>::operator= (const T)". Where i like to point that he has
a completely different argument (not const Point&, but const T). Please.
Why is the operator hidden?

Hiding is not about argument types. This is not like overloading. Hiding
is solely done by function NAME. In your case the function name is
"operator =". And "Point" has its own "operator =" (automatically
generated) which hides the "ndimobj::opera tor ="s (both, yours and the
automatic one).

Example:

struct Base
{
void foo();
};

struct Dev : Base
{
void foo(int);
};

int main()
{
Dev d;
d.foo(); //error
}

"void Base::foo()" is hidden by "void Dev::foo(int)" although they have
different arguments and may otherwise be overloads.
struct Base2
{
void foo();
};

struct Dev2 : Base2
{
using Base2::foo;
void foo(int);
};

int main()
{
Dev d;
d.foo(); //works
}

Here the "using Base2::foo" makes the base class functions visible
again. Now "foo" is an overloaded function.

Your case (simplified):

struct ndimobj
{
//ndimobj& operator = (ndimobj const&); //automatic
ndimobj& operator = (int); //your operator =
};

struct Point : ndimobj
{
//this "operator =" hides "operator =" of ndimobj:
//Point& operator = (Point const&); //automatic
};

The automatic operator = in Point hides the base class operator =.

HTH,
Frank