Comments on: Smart Pointer System http://www.rakkar.org/blog/?p=278 Trials of a game developer Tue, 20 Jul 2010 22:23:25 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: David Sveningsson http://www.rakkar.org/blog/?p=278&cpage=1#comment-13803 David Sveningsson Thu, 13 Dec 2007 18:20:13 +0000 http://www.rakkar.org/blog/?p=278#comment-13803 I wouldn't really call this a smart pointer, what you have implemented is usually just called reference counting. Smart pointers is a wrapper around a regular c/c++ pointer. I wouldn’t really call this a smart pointer, what you have implemented is usually just called reference counting. Smart pointers is a wrapper around a regular c/c++ pointer.

]]> By: Rak'kar http://www.rakkar.org/blog/?p=278&cpage=1#comment-13766 Rak'kar Thu, 29 Nov 2007 15:35:37 +0000 http://www.rakkar.org/blog/?p=278#comment-13766 Thanks for the clarification! Not being able to do forward declarations of pointers is a major drawback. I read about their system and here's the summary as I understand it scoped_ptr is fast. If the pointer goes out of scope, the object is deleted. If the object is deleted, the pointer is set to null. Cannot point to arrays. scoped_array same as scoped_ptr, but for arrays shared_ptr is like what I have, but you can have incomplete types and don't have to have your class derive from anything. Cannot point to arrays. shared_array is like shared_ptr for arrays weak_ptr is a pointer that will be set to null if the object is deleted, but won't delete the object itself when the pointer goes out of scope. Also used if shared_ptr would result in cycles intrusive_ptr like what I have. Thanks for the clarification! Not being able to do forward declarations of pointers is a major drawback.

I read about their system and here’s the summary as I understand it

scoped_ptr is fast. If the pointer goes out of scope, the object is deleted. If the object is deleted, the pointer is set to null. Cannot point to arrays.

scoped_array same as scoped_ptr, but for arrays

shared_ptr is like what I have, but you can have incomplete types and don’t have to have your class derive from anything. Cannot point to arrays.

shared_array is like shared_ptr for arrays

weak_ptr is a pointer that will be set to null if the object is deleted, but won’t delete the object itself when the pointer goes out of scope. Also used if shared_ptr would result in cycles

intrusive_ptr like what I have.

]]> By: Aras PranckeviÄ?ius http://www.rakkar.org/blog/?p=278&cpage=1#comment-13765 Aras PranckeviÄ?ius Thu, 29 Nov 2007 10:52:38 +0000 http://www.rakkar.org/blog/?p=278#comment-13765 More problems arise because of exception safety. Also if you want to the reference counting to be thread-safe. More problems arise because of exception safety. Also if you want to the reference counting to be thread-safe.

]]> By: Zeux http://www.rakkar.org/blog/?p=278&cpage=1#comment-13764 Zeux Thu, 29 Nov 2007 07:48:54 +0000 http://www.rakkar.org/blog/?p=278#comment-13764 This code is so called intrusive ptr. It's the simplest one from refcounted smart pointers, if you look into boost library, you'll see that their class is just as simple as your is, though more extensible. However, this approach has some problems: 1. You have to derive from base class - in order to share something simple you'll have to wrap it into a class that's derived from base 2. You can't specify incomplete (forward-declared) class as a type, thus creating unnecessary include-dependencies 3. It is harder for people (in my experience) to use intrusive pointers than pointers with a separate refcount 4. You can't make a weak pointer easily with intrusive ptr The reasons why boost has smart pointer are: 1. No matter how simple the code is, it's better to have a (more or less) standard component than to write a new one 2. It's definitely not that simple if we're talking about separate reference counts (shared_ptr in Boost). It's not rocket science either, but again it's much better to have a tested version instead of needing to write one. As much as I don't like Boost, smart pointers library is one of the best things in it in terms of implementation. This code is so called intrusive ptr. It’s the simplest one from refcounted smart pointers, if you look into boost library, you’ll see that their class is just as simple as your is, though more extensible.

However, this approach has some problems:
1. You have to derive from base class – in order to share something simple you’ll have to wrap it into a class that’s derived from base
2. You can’t specify incomplete (forward-declared) class as a type, thus creating unnecessary include-dependencies
3. It is harder for people (in my experience) to use intrusive pointers than pointers with a separate refcount
4. You can’t make a weak pointer easily with intrusive ptr

The reasons why boost has smart pointer are:
1. No matter how simple the code is, it’s better to have a (more or less) standard component than to write a new one
2. It’s definitely not that simple if we’re talking about separate reference counts (shared_ptr in Boost). It’s not rocket science either, but again it’s much better to have a tested version instead of needing to write one.

As much as I don’t like Boost, smart pointers library is one of the best things in it in terms of implementation.

]]>