automatically disassociating pointers

**underthesun** · 07-15-2009

I was wondering, what is the official/more likely name of this? The first I thought it'd be called is "automatically dereferencing smart poitners" but I guess that's wrong.

This is what it is : a smart pointer template class, where if it's pointing to B, and B is destroyed (and then the destructor called etc), the pointer class would automatically set its contents to NULL (instead of &B).

Does this exist, say, in boost library or anything like that? Otherwise, I'm thinking of my own implementation, but yeah, would be nice to have a good idea of how it's supposed to be done.

**laserlight** · 07-15-2009

Perhaps you are thinking of something like std::tr1::weak_ptr?

**underthesun** · 07-15-2009

is it standard c++? The first reference I got on the net was on msdn.. and it seems that you'll need a shared pointer to work with that?

**m37h0d** · 07-15-2009

this may be relevant to your interests

Reference counting - Wikipedia, the free encyclopedia

**laserlight** · 07-15-2009

Originally Posted by underthesun

is it standard c++?

Not yet, which is why it is currently in the std::tr1 namespace rather than the std namespace.

Originally Posted by underthesun

and it seems that you'll need a shared pointer to work with that?

Yes. The shared_ptr(s) have ownership, the weak_ptr(s) do not, but are notified when there are no owners left.

**underthesun** · 07-15-2009

Thanks..

what I've done is made a special smart container class using a template, where the stuff contained has to be a subclass of this base class, which will call the container to dissociate when it gets destructed. I think this should be ok, it's just strange for this not covered by stl. I guess I might have been imagining things..

**Elysia** · 07-15-2009

Don't be surprised. It's been 10 years or so since the last C++ standard came out and a lot of things change in 10 years. Plus there's only so much time they have to make new additions to the library.
Hopefully, this will pick up pace as the committee is exploring making Technical Reports with new additions to the standard library which compiler vendors can implement on a shorter period of time. They are then integrated into the main standard once its next revision comes around.

**underthesun** · 07-15-2009

Hmm, you are right. And then it'll take time for g++ to support all these new stuff, and even more time for it to be fully robust in supporting them I guess..

But while making that special pointer class I just couldn't help but think how nicely all the c++ stuff, those templates and operator overloading fit together.. Bjarne Stroustrup (sic?) is truly a genius.

**Elysia** · 07-15-2009

Ah, don't lay all the praise on him. He is just the founder of the language.
You should thank the C++ committee and all of the people who help out by suggesting changes.
And C++ still has its problems. Templates are very tricky stuff and there are still some things missing in that area...

**underthesun** · 07-15-2009

Ah that's true, fair enough. But without him, c++ wouldn't even be here.. how much did it end up changing from his initial vision anyways?

btw is there a way to have a "label" as a template argument? Would rather use them than #define macros..

**Elysia** · 07-15-2009

Originally Posted by underthesun

Ah that's true, fair enough. But without him, c++ wouldn't even be here.. how much did it end up changing from his initial vision anyways?

I think he's pretty happy with how it has evolved. There may have been a few things he'd rather see, but the committee lacks the funding to do.
So yes, praise be to him, but not all of it

btw is there a way to have a "label" as a template argument? Would rather use them than #define macros..

I'm not sure what you mean.
This is possible:

Code:

const int x = 10;
template<int N> void myfunc()
{
    std::cout << N << std::endl;
}

If not this, then I don't know what you mean...

**laserlight** · 07-15-2009

Originally Posted by underthesun

btw is there a way to have a "label" as a template argument? Would rather use them than #define macros..

What do you have in mind? You might want to give an example of the macro that you intend to do away with.

**CornedBee** · 07-16-2009

Originally Posted by underthesun

what I've done is made a special smart container class using a template, where the stuff contained has to be a subclass of this base class, which will call the container to dissociate when it gets destructed. I think this should be ok, it's just strange for this not covered by stl.

I distrust code that requires such shenanigans. Who holds this pointer, and why wouldn't he know that the pointee is gone through some other means? In a perfect ownership model, it should be technically impossible for an object to be destroyed while an observer holds a reference to it.

Also, given that your pointer is an observing pointer and it's usually quite possible that there's multiple observers for a single object, does that mean that the base class must hold a list of every observer?

**underthesun** · 07-16-2009

Yeah, it does hold a list of observers, using a list<Holder*>. The template class has a holder inside it. i.e

Code:

class Pointee{
   list<PointerHolder*> holders;
   // plus destructor methods that dissociates this
};

class PointerHolder{
   Pointee * thepointed;
   // plus destructor methods that dissociates with the pointed pointees
};

template <class SubclassOfPointee>
class Holder{
   PointerHolder holder;
   // plus a bunch of convenient operators such as:
   Holder<SubclassOfPointee>& operator=(SubclassOfPointee * thepointee){
      holder.assign(thepointee);
      return *this;
   }
};

That list of pointers might seem overkill, but I was going to implement it that way for what I needed for keeping track of a bunch of enemies in a game, where they could die any second, and these observers do not own the pointer in any way.

**underthesun** · 07-16-2009

What do you have in mind? You might want to give an example of the macro that you intend to do away with.

Was thinking of allowing

Code:

template <class M, label moo>
void do100times(M& m){
   m.moo();
}

Then again you can probably do most of that with #define macros, except they seem a bit messy.. sorry was going off topic with the thread.

Thread: automatically disassociating pointers

Thread Tools

Search Thread

Display

automatically disassociating pointers

Similar Threads

pointers

Pointers to function(function pointers)

API "Clean Up" Functions & delete Pointers :: Winsock

Pointers pointers pointers...