cyclic defination problem.

T

toton

Hi I have a class called Session, which stores a vector of Page, like
vector<Page> Also each Page need's to know the Session to which it is
attached.

Thus I have, (the classes has many other things, I copied only the
portion i needed)
in Session.hpp

#include "Page.hpp"
class Session {
private:
std::vector<Page> _pages;
};

while in Page.hpp
class Session;// forward decl
class Page{
private:
Session* _session;
public:
Page(Session* session) : _session(session) {}
};
Is it the best way to remove cyclic definition problem? Or any way I
can store a reference to Session rather than pointer to Session?

Thanks
abir
 
V

Victor Bazarov

toton said:
Hi I have a class called Session, which stores a vector of Page, like
vector<Page> Also each Page need's to know the Session to which it is
attached.

Thus I have, (the classes has many other things, I copied only the
portion i needed)
in Session.hpp

#include "Page.hpp"
class Session {
private:
std::vector<Page> _pages;
};

while in Page.hpp
class Session;// forward decl
class Page{
private:
Session* _session;
public:
Page(Session* session) : _session(session) {}
};
Is it the best way to remove cyclic definition problem? Or any way I
can store a reference to Session rather than pointer to Session?

Whether to store a reference to Session or a pointer to Session is up
to you, and either would require using some kind of forward declaration,
which you have done already. Keep in mind that if you store a reference
in 'Page' you will have to initialise it during construction of the Page
object and you will not be able to change it during the Page's lifetime.
If that fits your design, prefer the reference. Otherwise, go with the
pointer.

V
 
M

mlimber

toton said:
Hi I have a class called Session, which stores a vector of Page, like
vector<Page> Also each Page need's to know the Session to which it is
attached.

Thus I have, (the classes has many other things, I copied only the
portion i needed)
in Session.hpp

#include "Page.hpp"
class Session {
private:
std::vector<Page> _pages;
};

while in Page.hpp
class Session;// forward decl
class Page{
private:
Session* _session;
public:
Page(Session* session) : _session(session) {}
};
Is it the best way to remove cyclic definition problem? Or any way I
can store a reference to Session rather than pointer to Session?

You could have a reference (it's simple enough to test -- did you try
it?). See this article for more on forward declarations and breaking
dependencies:

http://www.gotw.ca/publications/mill04.htm

Cheers! --M
 
J

John Carson

mlimber said:
You could have a reference (it's simple enough to test -- did you try
it?).

Use of a reference is a little tricky.

Page is stored in a vector and objects stored in standard containers must be
assignable. The compiler won't generate an assignment operator for Page,
because a reference cannot be re-seated.

You could write an assignment operator for Page to get the code to compile,
but any user-defined assignment operator likewise will not re-seat a
reference, so there could be some unexpected behaviour.
 
M

mlimber

John said:
Use of a reference is a little tricky.

Page is stored in a vector and objects stored in standard containers must be
assignable. The compiler won't generate an assignment operator for Page,
because a reference cannot be re-seated.

You could write an assignment operator for Page to get the code to compile,
but any user-defined assignment operator likewise will not re-seat a
reference, so there could be some unexpected behaviour.

Right, and the same is true if one used a const pointer to a Session
object instead (const-correctness and all that). I took the OP's
question to be about whether a forward declaration would work with a
reference as well as a pointer, which, of course, it will.

Cheers! --M
 
T

toton

Victor said:
Whether to store a reference to Session or a pointer to Session is up
to you, and either would require using some kind of forward declaration,
which you have done already. Keep in mind that if you store a reference
in 'Page' you will have to initialise it during construction of the Page
object and you will not be able to change it during the Page's lifetime.
If that fits your design, prefer the reference. Otherwise, go with the
pointer.
I prefer to store reference. Infact I want to do it in ctor, as you
have mentioned. I prefer not to change it. I also prefer to have a
constant reference.
like
class Page{
private:
const Session& _session;
};
but, at the same time, I prefer to pass it in the constructor as
reference, rather than pointer.
like Page(const Session& session) : _session(session) {}
Here some cyclic definition creating problem. I am not sure whether
this can be done, i.e only forward defination will allow me to do so or
not, or in my case problem is coming from some other cyclic
definition.
Thanks...
 
V

Victor Bazarov

toton said:
[..]
I prefer to store reference. Infact I want to do it in ctor, as you
have mentioned. I prefer not to change it. I also prefer to have a
constant reference.
like
class Page{
private:
const Session& _session;
};
but, at the same time, I prefer to pass it in the constructor as
reference, rather than pointer.
like Page(const Session& session) : _session(session) {}
Here some cyclic definition creating problem. I am not sure whether
this can be done, i.e only forward defination will allow me to do so
or not, or in my case problem is coming from some other cyclic
definition.

Pull the _implementation_ of the constructor out of your 'Page' class
definition. Place it along with other Page's member functions in its
own implementation file (translation unit), and include both headers
in it (in any order) before defining all functions.

V
 
T

toton

Victor said:
toton said:
[..]
I prefer to store reference. Infact I want to do it in ctor, as you
have mentioned. I prefer not to change it. I also prefer to have a
constant reference.
like
class Page{
private:
const Session& _session;
};
but, at the same time, I prefer to pass it in the constructor as
reference, rather than pointer.
like Page(const Session& session) : _session(session) {}
Here some cyclic definition creating problem. I am not sure whether
this can be done, i.e only forward defination will allow me to do so
or not, or in my case problem is coming from some other cyclic
definition.

Pull the _implementation_ of the constructor out of your 'Page' class
definition. Place it along with other Page's member functions in its
own implementation file (translation unit), and include both headers
in it (in any order) before defining all functions.

Now cyclic definition problem is solved, but one more problem arises.
When I declare Session as pointer inside Page, and use a vector<Page>
inside session, it works fine.

However, when I declare Session as reference inside Page and use
vector<Page> in session, vector<Page> unable to find the assignment
operator for Page.
I hadn't defined the assignment operator for Page explicitly, ( I only
have a ctor, & virtual dtor, not defined copy ctor, assignment operator
or equality operator explicitly). It seems, that for the reference
case, Page class is not defining assignment operator automatically,
however it defines when I am passing pointer to Session in Page. and
thus vector<Page> is having problem inside Session class.
Do I need to define assignment operator explicitly in this case?
thanks
 
V

Victor Bazarov

toton said:
[..]
Now cyclic definition problem is solved, but one more problem arises.
When I declare Session as pointer inside Page, and use a vector<Page>
inside session, it works fine.

However, when I declare Session as reference inside Page and use
vector<Page> in session, vector<Page> unable to find the assignment
operator for Page.
I hadn't defined the assignment operator for Page explicitly, ( I only
have a ctor, & virtual dtor, not defined copy ctor, assignment
operator or equality operator explicitly). It seems, that for the
reference case, Page class is not defining assignment operator
automatically, however it defines when I am passing pointer to
Session in Page. and thus vector<Page> is having problem inside
Session class.
Do I need to define assignment operator explicitly in this case?
thanks

Yes, assignment operators are very tricky when your class has a member
that is a reference. References cannot be "reseated", i.e. made to
refer to another object than the one with which they were initialised.
So, imagine you have

class A {};
class B {
A & a;
public:
B(A& ra) : a(ra) {}
};

The compiler does not know how to generate the assignment operator
for class 'B', since the assignment semantics for references are to
assign the objects and since you're keeping the reference in 'B', you
clearly don't want to affect the objects you refer to. You can write
your own assignment operator and forgo doing anything special about
the reference member (like this:

B& operator =(const B& other_b) {
// do nothing
return *this;
}

) which means that when assigned new value, a B object will retain
the reference, IOW its 'a' member will still refer to what the object
was constructed to refer to. Is that what you want? I don't know.
If you want "reseat-ability", use pointers.

V
 
T

toton

Victor said:
toton said:
[..]
Now cyclic definition problem is solved, but one more problem arises.
When I declare Session as pointer inside Page, and use a vector<Page>
inside session, it works fine.

However, when I declare Session as reference inside Page and use
vector<Page> in session, vector<Page> unable to find the assignment
operator for Page.
I hadn't defined the assignment operator for Page explicitly, ( I only
have a ctor, & virtual dtor, not defined copy ctor, assignment
operator or equality operator explicitly). It seems, that for the
reference case, Page class is not defining assignment operator
automatically, however it defines when I am passing pointer to
Session in Page. and thus vector<Page> is having problem inside
Session class.
Do I need to define assignment operator explicitly in this case?
thanks

Yes, assignment operators are very tricky when your class has a member
that is a reference. References cannot be "reseated", i.e. made to
refer to another object than the one with which they were initialised.
So, imagine you have

class A {};
class B {
A & a;
public:
B(A& ra) : a(ra) {}
};

The compiler does not know how to generate the assignment operator
for class 'B', since the assignment semantics for references are to
assign the objects and since you're keeping the reference in 'B', you
clearly don't want to affect the objects you refer to. You can write
your own assignment operator and forgo doing anything special about
the reference member (like this:

B& operator =(const B& other_b) {
// do nothing
return *this;
}

) which means that when assigned new value, a B object will retain
the reference, IOW its 'a' member will still refer to what the object
was constructed to refer to. Is that what you want? I don't know.
If you want "reseat-ability", use pointers.

In my situation, the reference of Session (or pointer, still not
decided! ) to a Page, and Page can not change the Session, it can only
refer to the Session it belongs. Thus Page knows the Session it belongs
at construction (in ctor) , and do not have a method like setSession.
So, I don't think it is "reseat-able", a Page is associated with a
Session throughout it's lifetime, and Session can remove a Page from
its evctor<Page> if the processing is done. Similarly, a Char knows The
Page it belongs, and can refer it, but itself can't change the Page it
belongs. However a Page can remove the a Char from it's deque<Char>.
That is why I thught reference is a better option.
If I define the assignment operator as you have mentioned, does it mean
that for two page
Page p1,p2; if I write p1 = p2, then p1 will not get modified
irrespective of what p2 is or will it be same object as p2 ?
Also, will the copy constructor work in its usual way? As I have a
vector<Page> inside the session, copy contsructor for Page is needed
for Session to work properly.
thanks.
 
V

Victor Bazarov

toton said:
[..]
If I define the assignment operator as you have mentioned, does it
mean that for two page
Page p1,p2; if I write p1 = p2, then p1 will not get modified
irrespective of what p2 is or will it be same object as p2 ?

Modify anything you want, just don't touch the reference (in any way).
The assignment operator should perform all actions necessary to copy
data members. It's not possible to copy references (only copy objects
to which they refer), so usually you'd leave references alone. Thus,
the reference refers to the same object throughout its lifetime, no
matter whether the other members get changed along the way using the
assignment operator.
Also, will the copy constructor work in its usual way?

That should be fine.
As I have a
vector<Page> inside the session, copy contsructor for Page is needed
for Session to work properly.

Sure. Will not be a problem.

V
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

No members online now.

Forum statistics

Threads
473,769
Messages
2,569,580
Members
45,054
Latest member
TrimKetoBoost

Latest Threads

Top