vector with deep copy

[Gunnar G]

Is there anything like vector in STL, that performes deep copy of the
elements it contains?

I hope this will appear in future releases of STL :)

[Jeff Schwab]

Re: vector with deep copy

Gunnar G wrote:[color=blue]
> Is there anything like vector in STL, that performes deep copy of the
> elements it contains?
>
> I hope this will appear in future releases of STL :)[/color]

That depends how deep you want the copy to be... If a vector contains
pointers, do you want the new copy to contain new pointers to copies of
the originally referenced elements? Can you guarantee that none of the
original pointers was null, and that all of them pointed to valid
objects? If the original pointers were of type pointer-to-A, but some
of the objects actually were of some subclass of A, how do you avoid
slicing?

[Gunnar G]

Re: vector with deep copy

> That depends how deep you want the copy to be... If a vector contains[color=blue]
> pointers, do you want the new copy to contain new pointers to copies of
> the originally referenced elements?[/color]
That's whats meant by deep copy, isn't it?
Anyway, that is what I want.
[color=blue]
> Can you guarantee that none of the original pointers was null,
> and that all of them pointed to valid objects?[/color]
Null pointers should never be stored in the container, so yes, that is
guaranteed.
[color=blue]
> If the original pointers were of type pointer-to-A, but some
> of the objects actually were of some subclass of A, how do you avoid
> slicing?[/color]
Well, I have to read more about slicing, but I would certainly want to have
an object of the subclass in the copy if that is what was in the original.
Can't I assume that the user is using virtual functions in their classes,
and then everything is perfectly happy?

I've read something about a virtual constructors in Bjarne's book, guess
I'll have to read more.

[tom_usenet]

Re: vector with deep copy

On Mon, 10 May 2004 11:05:29 GMT, Gunnar G <debian@comhem. se> wrote:
[color=blue]
>Is there anything like vector in STL, that performes deep copy of the
>elements it contains?[/color]

std::vector does perform deep copies - it copies contained elements
using their copy constructors.

Tom
--
C++ FAQ: http://www.parashift.com/c++-faq-lite/
C FAQ: http://www.eskimo.com/~scs/C-faq/top.html

[Jeff Schwab]

Re: vector with deep copy

Gunnar G wrote:[color=blue][color=green]
>>That depends how deep you want the copy to be... If a vector contains
>>pointers, do you want the new copy to contain new pointers to copies of
>>the originally referenced elements?[/color]
>
> That's whats meant by deep copy, isn't it?[/color]

Not necessarily. The elements "contained" by a std::vector aren't
actually members of the container class, they're held in memory pointed
to by a member variable. Since the referenced memory is copied, a
vector already is doing a deep copy.
[color=blue]
> Anyway, that is what I want.
>
>[color=green]
>>Can you guarantee that none of the original pointers was null,
>>and that all of them pointed to valid objects?[/color]
>
> Null pointers should never be stored in the container, so yes, that is
> guaranteed.[/color]

You mean under your particular circumstances, or are you asserting that
null pointers should never be stored in std::vector?
[color=blue][color=green]
>>If the original pointers were of type pointer-to-A, but some
>>of the objects actually were of some subclass of A, how do you avoid
>>slicing?[/color]
>
> Well, I have to read more about slicing, but I would certainly want to have
> an object of the subclass in the copy if that is what was in the original.
> Can't I assume that the user is using virtual functions in their classes,
> and then everything is perfectly happy?[/color]

No, everything is not then perfectly happy.
[color=blue]
> I've read something about a virtual constructors in Bjarne's book, guess
> I'll have to read more.[/color]

At this point, you're already imposing some pretty tough rules on the
types of element that can be stored in the vector. It sounds like you
need an "intrusive" container, very different from std::vector. Why
don't you implement your own, possibly using std::vector as part of the
implementation, and post your code here for feedback?

[Gunnar G]

Re: vector with deep copy

> std::vector does perform deep copies - it copies contained elements[color=blue]
> using their copy constructors.[/color]
and if I copy a

vector<Foo*> x ?

Will it not only copy the addresses of the Foo elements? That is AFAIK a
shallow copy.

[Daniel T.]

Re: vector with deep copy

In article <ZlJnc.58692$mU 6.237676@newsb. telia.net>,
Gunnar G <debian@comhem. se> wrote:
[color=blue]
>Is there anything like vector in STL, that performes deep copy of the
>elements it contains?
>
>I hope this will appear in future releases of STL :)[/color]

vector already does this. Simply hold a vector of objects that perform
deep copy when copied and vector will perform a deep copy of each object.

[Marcin Kalicinski]

Re: vector with deep copy


Uzytkownik "Gunnar G" <debian@comhem. se> napisal w wiadomosci
news:oXJnc.9213 9$dP1.288032@ne wsc.telia.net.. .[color=blue][color=green]
> > std::vector does perform deep copies - it copies contained elements
> > using their copy constructors.[/color]
> and if I copy a
>
> vector<Foo*> x ?
>
> Will it not only copy the addresses of the Foo elements? That is AFAIK a
> shallow copy.[/color]

Yes it will copy only addresses. But that is what it contains - so this is
actually a 'deep copy'. std::vector does not dereference the stored pointers
in any way and treats them exactly as if they were, say, ints.

If you want std::vector to copy objects, create a vector with objects, not
with pointers to them:

std::vector<Foo > x;

regards,
Marcin

[Gunnar G]

Re: vector with deep copy

> At this point, you're already imposing some pretty tough rules on the[color=blue]
> types of element that can be stored in the vector. It sounds like you
> need an "intrusive" container, very different from std::vector. Why
> don't you implement your own, possibly using std::vector as part of the
> implementation, and post your code here for feedback?[/color]

I guess we have been talking about different things, Yes I'm gonna write my
own container, I just asked if someone had done this before me so I didn't
have to reinvent the wheel :-)

Anyway, std::vector does not make a deep copy (as I define deep copy), at
least not with gcc-3.0. If it did a deep copy, it would give different
adresses below.

#include <iostream>
#include <vector>

using namespace std;

class Foo{
public:
int x;
Foo(){x=1;}
Foo(int q):x(q){}
};

int main(){
vector<Foo*> a,b;
Foo* t=new Foo(3);
a.push_back(t);
cout<<"Adress "<<a[0]<<endl;
b=a;
cout<<"Adress "<<b[0]<<endl;
}

[Howard Hinnant]

Re: vector with deep copy

In article <DsKnc.92143$dP 1.287973@newsc. telia.net>,
Gunnar G <debian@comhem. se> wrote:
[color=blue][color=green]
> > At this point, you're already imposing some pretty tough rules on the
> > types of element that can be stored in the vector. It sounds like you
> > need an "intrusive" container, very different from std::vector. Why
> > don't you implement your own, possibly using std::vector as part of the
> > implementation, and post your code here for feedback?[/color]
>
> I guess we have been talking about different things, Yes I'm gonna write my
> own container, I just asked if someone had done this before me so I didn't
> have to reinvent the wheel :-)
>
> Anyway, std::vector does not make a deep copy (as I define deep copy), at
> least not with gcc-3.0. If it did a deep copy, it would give different
> adresses below.
>
> #include <iostream>
> #include <vector>
>
> using namespace std;
>
> class Foo{
> public:
> int x;
> Foo(){x=1;}
> Foo(int q):x(q){}
> };
>
> int main(){
> vector<Foo*> a,b;
> Foo* t=new Foo(3);
> a.push_back(t);
> cout<<"Adress "<<a[0]<<endl;
> b=a;
> cout<<"Adress "<<b[0]<<endl;
> }[/color]

Another approach you might consider besides implementing a container is
implementing a smart pointer: copy_ptr<T>:

copy_ptr<Foo> a(new Foo(3));
copy_ptr<Foo> b(a); // makes copy with new Foo(const Foo&)
assert(a != b);

Then you can just:

vector<copy_ptr <Foo> > va;
va.push_back(co py_ptr<Foo>(new Foo));

An advantage to this approach is that if you decide you need different
semantics on copy, it is easier to change to a new smart pointer rather
than rewrite your container. For example if Foo is a base class with a
virtual clone function (typical virtual copy ctor) you could create a
clone_ptr:

clone_ptr<Foo> a(new DerivedFromFoo( 3));
clone_ptr<Foo> b(a); // makes copy with a->clone()
assert(a != b);
....
vector<clone_pt r<Foo> > vec_of_base_poi nters;

A high quality refcounted pointer is available at boost (www.boost.org).
The same design is also in the first library technical report and thus
may be available with your C++ compiler under namespace std::tr1 (e.g.
Metrowerks Pro 9). So you could:

vector<std::tr1 ::shared_ptr<Fo o> > shared_vec;

-Howard