[ceph.git] / ceph / src / boost / libs / smart_ptr / sp_techniques.html

<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
    <head>
        <title>Smart Pointer Programming Techniques</title>
        <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
    </head>
    <body text="#000000" bgcolor="#ffffff" link="#0000ff" vlink="#0000ff">
        <h1><img height="86" alt="boost.png (6897 bytes)" src="../../boost.png" 
            width="277" align="middle" border="0">Smart Pointer Programming Techniques</h1>
        <p><A href="#incomplete">Using incomplete classes for implementation hiding</A><br>
            <A href="#pimpl">The "Pimpl" idiom</A><br>
            <A href="#abstract">Using abstract classes for implementation hiding</A><br>
            <A href="#preventing_delete">Preventing <code>delete px.get()</code></A><br>
            <A href="#array">Using a <code>shared_ptr</code> to hold a pointer to an array</A><br>
            <A href="#encapsulation">Encapsulating allocation details, wrapping factory 
                functions</A><br>
            <A href="#static">Using a <code>shared_ptr</code> to hold a pointer to a statically 
                allocated object</A><br>
            <A href="#com">Using a <code>shared_ptr</code> to hold a pointer to a COM object</A><br>
            <A href="#intrusive">Using a <code>shared_ptr</code> to hold a pointer to an object 
                with an embedded reference count</A><br>
            <A href="#another_sp">Using a <code>shared_ptr</code> to hold another shared 
                ownership smart pointer</A><br>
            <A href="#from_raw">Obtaining a <code>shared_ptr</code> from a raw pointer</A><br>
            <A href="#in_constructor">Obtaining a <code>shared_ptr</code> (<code>weak_ptr</code>) 
                to <code>this</code> in a constructor</A><br>
            <A href="#from_this">Obtaining a <code>shared_ptr</code> to <code>this</code></A><br>
            <A href="#handle">Using <code>shared_ptr</code> as a smart counted handle</A><br>
            <A href="#on_block_exit">Using <code>shared_ptr</code> to execute code on block 
                exit</A><br>
            <A href="#pvoid">Using <code>shared_ptr&lt;void&gt;</code> to hold an arbitrary 
                object</A><br>
            <A href="#extra_data">Associating arbitrary data with heterogeneous <code>shared_ptr</code>
                instances</A><br>
            <A href="#as_lock">Using <code>shared_ptr</code> as a CopyConstructible mutex lock</A><br>
            <A href="#wrapper">Using <code>shared_ptr</code> to wrap member function calls</A><br>
            <A href="#delayed">Delayed deallocation</A><br>
            <A href="#weak_without_shared">Weak pointers to objects not managed by a <code>shared_ptr</code></A><br>
        </p>
        <h2><A name="incomplete">Using incomplete classes for implementation hiding</A></h2>
        <p>A proven technique (that works in C, too) for separating interface from 
            implementation is to use a pointer to an incomplete class as an opaque handle:</p>
        <pre>class FILE;

FILE * fopen(char const * name, char const * mode);
void fread(FILE * f, void * data, size_t size);
void fclose(FILE * f);
</pre>
        <p>It is possible to express the above interface using <code>shared_ptr</code>, 
            eliminating the need to manually call <code>fclose</code>:</p>
        <pre>class FILE;

shared_ptr&lt;FILE&gt; fopen(char const * name, char const * mode);
void fread(shared_ptr&lt;FILE&gt; f, void * data, size_t size);
</pre>
        <p>This technique relies on <code>shared_ptr</code>'s ability to execute a custom 
            deleter, eliminating the explicit call to <code>fclose</code>, and on the fact 
            that <code>shared_ptr&lt;X&gt;</code> can be copied and destroyed when <code>X</code>
            is incomplete.</p>
        <h2><A name="pimpl">The "Pimpl" idiom</A></h2>
        <p>A C++ specific variation of the incomplete class pattern is the "Pimpl" idiom. 
            The incomplete class is not exposed to the user; it is hidden behind a 
            forwarding facade. <code>shared_ptr</code> can be used to implement a "Pimpl":</p>
        <pre>// file.hpp:

class file
{
private:

    class impl;
    shared_ptr&lt;impl&gt; pimpl_;

public:

    file(char const * name, char const * mode);

    // compiler generated members are fine and useful

    void read(void * data, size_t size);
};
</pre>
        <pre>// file.cpp:

#include "file.hpp"

class file::impl
{
private:

    impl(impl const &amp;);
    impl &amp; operator=(impl const &amp;);

    // private data

public:

    impl(char const * name, char const * mode) { ... }
    ~impl() { ... }
    void read(void * data, size_t size) { ... }
};

file::file(char const * name, char const * mode): pimpl_(new impl(name, mode))
{
}

void file::read(void * data, size_t size)
{
    pimpl_-&gt;read(data, size);
}
</pre>
        <p>The key thing to note here is that the compiler-generated copy constructor, 
            assignment operator, and destructor all have a sensible meaning. As a result, <code>
                file</code> is <code>CopyConstructible</code> and <code>Assignable</code>, 
            allowing its use in standard containers.</p>
        <h2><A name="abstract">Using abstract classes for implementation hiding</A></h2>
        <p>Another widely used C++ idiom for separating inteface and implementation is to 
            use abstract base classes and factory functions. The abstract classes are 
            sometimes called "interfaces" and the pattern is known as "interface-based 
            programming". Again, <code>shared_ptr</code> can be used as the return type of 
            the factory functions:</p>
        <pre>// X.hpp:

class X
{
public:

    virtual void f() = 0;
    virtual void g() = 0;

protected:

    ~X() {}
};

shared_ptr&lt;X&gt; createX();
</pre>
        <pre>-- X.cpp:

class X_impl: public X
{
private:

    X_impl(X_impl const &amp;);
    X_impl &amp; operator=(X_impl const &amp;);

public:

    virtual void f()
    {
      // ...
    }

    virtual void g()
    {
      // ...
    }
};

shared_ptr&lt;X&gt; createX()
{
    shared_ptr&lt;X&gt; px(new X_impl);
    return px;
}
</pre>
        <p>A key property of shared_ptr is that the allocation, construction, deallocation, 
            and destruction details are captured at the point of construction, inside the 
            factory function. Note the protected and nonvirtual destructor in the example 
            above. The client code cannot, and does not need to, delete a pointer to <code>X</code>; 
            the <code>shared_ptr&lt;X&gt;</code> instance returned from <code>createX</code>
            will correctly call <code>~X_impl</code>.</p>
        <h2><A name="preventing_delete">Preventing <code>delete px.get()</code></A></h2>
        <p>It is often desirable to prevent client code from deleting a pointer that is 
            being managed by <code>shared_ptr</code>. The previous technique showed one 
            possible approach, using a protected destructor. Another alternative is to use 
            a private deleter:</p>
        <pre>class X
{
private:

    ~X();

    class deleter;
    friend class deleter;

    class deleter
    {
    public:

        void operator()(X * p) { delete p; }
    };

public:

    static shared_ptr&lt;X&gt; create()
    {
        shared_ptr&lt;X&gt; px(new X, X::deleter());
        return px;
    }
};
</pre>
        <h2><A name="array">Using a <code>shared_ptr</code> to hold a pointer to an array</A></h2>
        <p>A <code>shared_ptr</code> can be used to hold a pointer to an array allocated 
            with <code>new[]</code>:</p>
        <pre>shared_ptr&lt;X&gt; px(new X[1], <A href="../utility/checked_delete.html" >checked_array_deleter</A>&lt;X&gt;());
</pre>
        <p>Note, however, that <code><A href="shared_array.htm">shared_array</A></code> is 
            often preferable, if this is an option. It has an array-specific interface, 
            without <code>operator*</code> and <code>operator-&gt;</code>, and does not 
            allow pointer conversions.</p>
        <h2><A name="encapsulation">Encapsulating allocation details, wrapping factory 
                functions</A></h2>
        <p><code>shared_ptr</code> can be used in creating C++ wrappers over existing C 
            style library interfaces that return raw pointers from their factory functions 
            to encapsulate allocation details. As an example, consider this interface, 
            where <code>CreateX</code> might allocate <code>X</code> from its own private 
            heap, <code>~X</code> may be inaccessible, or <code>X</code> may be incomplete:</p>
        <pre>X * CreateX();
void DestroyX(X *);
</pre>
        <p>The only way to reliably destroy a pointer returned by <code>CreateX</code> is 
            to call <code>DestroyX</code>.</p>
        <P>Here is how a <code>shared_ptr</code>-based wrapper may look like:</P>
        <pre>shared_ptr&lt;X&gt; createX()
{
    shared_ptr&lt;X&gt; px(CreateX(), DestroyX);
    return px;
}
</pre>
        <p>Client code that calls <code>createX</code> still does not need to know how the 
            object has been allocated, but now the destruction is automatic.</p>
        <h2><A name="static">Using a <code>shared_ptr</code> to hold a pointer to a statically 
                allocated object</A></h2>
        <p>Sometimes it is desirable to create a <code>shared_ptr</code> to an already 
            existing object, so that the <code>shared_ptr</code> does not attempt to 
            destroy the object when there are no more references left. As an example, the 
            factory function:</p>
        <pre>shared_ptr&lt;X&gt; createX();
</pre>
        <p>in certain situations may need to return a pointer to a statically allocated <code>X</code>
            instance.</p>
        <P>The solution is to use a custom deleter that does nothing:</P>
        <pre>struct null_deleter
{
    void operator()(void const *) const
    {
    }
};

static X x;

shared_ptr&lt;X&gt; createX()
{
    shared_ptr&lt;X&gt; px(&amp;x, null_deleter());
    return px;
}
</pre>
        <p>The same technique works for any object known to outlive the pointer.</p>
        <h2><A name="com">Using a <code>shared_ptr</code> to hold a pointer to a COM Object</A></h2>
        <p>Background: COM objects have an embedded reference count and two member 
            functions that manipulate it. <code>AddRef()</code> increments the count. <code>Release()</code>
            decrements the count and destroys itself when the count drops to zero.</p>
        <P>It is possible to hold a pointer to a COM object in a <code>shared_ptr</code>:</P>
        <pre>shared_ptr&lt;IWhatever&gt; make_shared_from_COM(IWhatever * p)
{
    p-&gt;AddRef();
    shared_ptr&lt;IWhatever&gt; pw(p, <A href="../bind/mem_fn.html" >mem_fn</A>(&amp;IWhatever::Release));
    return pw;
}
</pre>
        <p>Note, however, that <code>shared_ptr</code> copies created from <code>pw</code> will 
            not "register" in the embedded count of the COM object; they will share the 
            single reference created in <code>make_shared_from_COM</code>. Weak pointers 
            created from <code>pw</code> will be invalidated when the last <code>shared_ptr</code>
            is destroyed, regardless of whether the COM object itself is still alive.</p>
        <P>As <A href="../bind/mem_fn.html#Q3">explained</A> in the <code>mem_fn</code> documentation, 
            you need to <A href="../bind/mem_fn.html#stdcall">#define 
                BOOST_MEM_FN_ENABLE_STDCALL</A> first.</P>
        <h2><A name="intrusive">Using a <code>shared_ptr</code> to hold a pointer to an object 
                with an embedded reference count</A></h2>
        <p>This is a generalization of the above technique. The example assumes that the 
            object implements the two functions required by <code><A href="intrusive_ptr.html">intrusive_ptr</A></code>,
            <code>intrusive_ptr_add_ref</code> and <code>intrusive_ptr_release</code>:</p>
        <pre>template&lt;class T&gt; struct intrusive_deleter
{
    void operator()(T * p)
    {
        if(p) intrusive_ptr_release(p);
    }
};

shared_ptr&lt;X&gt; make_shared_from_intrusive(X * p)
{
    if(p) intrusive_ptr_add_ref(p);
    shared_ptr&lt;X&gt; px(p, intrusive_deleter&lt;X&gt;());
    return px;
}
</pre>
        <h2><A name="another_sp">Using a <code>shared_ptr</code> to hold another shared 
                ownership smart pointer</A></h2>
        <p>One of the design goals of <code>shared_ptr</code> is to be used in library 
            interfaces. It is possible to encounter a situation where a library takes a <code>shared_ptr</code>
            argument, but the object at hand is being managed by a different reference 
            counted or linked smart pointer.</p>
        <P>It is possible to exploit <code>shared_ptr</code>'s custom deleter feature to 
            wrap this existing smart pointer behind a <code>shared_ptr</code> facade:</P>
        <pre>template&lt;class P&gt; struct smart_pointer_deleter
{
private:

    P p_;

public:

    smart_pointer_deleter(P const &amp; p): p_(p)
    {
    }

    void operator()(void const *)
    {
        p_.reset();
    }
    
    P const &amp; get() const
    {
        return p_;
    }
};

shared_ptr&lt;X&gt; make_shared_from_another(another_ptr&lt;X&gt; qx)
{
    shared_ptr&lt;X&gt; px(qx.get(), smart_pointer_deleter&lt; another_ptr&lt;X&gt; &gt;(qx));
    return px;
}
</pre>
        <p>One subtle point is that deleters are not allowed to throw exceptions, and the 
            above example as written assumes that <code>p_.reset()</code> doesn't throw. If 
            this is not the case, <code>p_.reset()</code> should be wrapped in a <code>try {} 
                catch(...) {}</code> block that ignores exceptions. In the (usually 
            unlikely) event when an exception is thrown and ignored, <code>p_</code> will 
            be released when the lifetime of the deleter ends. This happens when all 
            references, including weak pointers, are destroyed or reset.</p>
        <P>Another twist is that it is possible, given the above <code>shared_ptr</code> instance, 
            to recover the original smart pointer, using <code><A href="shared_ptr.htm#get_deleter">
                    get_deleter</A></code>:</P>
        <pre>void extract_another_from_shared(shared_ptr&lt;X&gt; px)
{
    typedef smart_pointer_deleter&lt; another_ptr&lt;X&gt; &gt; deleter;

    if(deleter const * pd = get_deleter&lt;deleter&gt;(px))
    {
        another_ptr&lt;X&gt; qx = pd-&gt;get();
    }
    else
    {
        // not one of ours
    }
}
</pre>
        <h2><A name="from_raw">Obtaining a <code>shared_ptr</code> from a raw pointer</A></h2>
        <p>Sometimes it is necessary to obtain a <code>shared_ptr</code> given a raw 
            pointer to an object that is already managed by another <code>shared_ptr</code> 
            instance. Example:</p>
        <pre>void f(X * p)
{
    shared_ptr&lt;X&gt; px(<i>???</i>);
}
</pre>
        <p>Inside <code>f</code>, we'd like to create a <code>shared_ptr</code> to <code>*p</code>.</p>
        <P>In the general case, this problem has no solution. One approach is to modify <code>f</code>
            to take a <code>shared_ptr</code>, if possible:</P>
        <pre>void f(shared_ptr&lt;X&gt; px);
</pre>
        <p>The same transformation can be used for nonvirtual member functions, to convert 
            the implicit <code>this</code>:</p>
        <pre>void X::f(int m);
</pre>
        <p>would become a free function with a <code>shared_ptr</code> first argument:</p>
        <pre>void f(shared_ptr&lt;X&gt; this_, int m);
</pre>
        <p>If <code>f</code> cannot be changed, but <code>X</code> uses intrusive counting, 
            use <code><A href="#intrusive">make_shared_from_intrusive</A></code> described 
            above. Or, if it's known that the <code>shared_ptr</code> created in <code>f</code>
            will never outlive the object, use <A href="#static">a null deleter</A>.</p>
        <h2><A name="in_constructor">Obtaining a <code>shared_ptr</code> (<code>weak_ptr</code>) 
                to <code>this</code> in a constructor</A></h2>
        <p>Some designs require objects to register themselves on construction with a 
            central authority. When the registration routines take a shared_ptr, this leads 
            to the question how could a constructor obtain a shared_ptr to this:</p>
        <pre>class X
{
public:

    X()
    {
        shared_ptr&lt;X&gt; this_(<i>???</i>);
    }
};
</pre>
        <p>In the general case, the problem cannot be solved. The <code>X</code> instance 
            being constructed can be an automatic variable or a static variable; it can be 
            created on the heap:</p>
        <pre>shared_ptr&lt;X&gt; px(new X);</pre>
        <P>but at construction time, <code>px</code> does not exist yet, and it is 
            impossible to create another <code>shared_ptr</code> instance that shares 
            ownership with it.</P>
        <P>Depending on context, if the inner <code>shared_ptr</code> <code>this_</code> doesn't 
            need to keep the object alive, use a <code>null_deleter</code> as explained <A href="#static">
                here</A> and <A href="#weak_without_shared">here</A>. If <code>X</code> is 
            supposed to always live on the heap, and be managed by a <code>shared_ptr</code>, 
            use a static factory function:</P>
        <pre>class X
{
private:

    X() { ... }

public:

    static shared_ptr&lt;X&gt; create()
    {
        shared_ptr&lt;X&gt; px(new X);
        // use px as 'this_'
        return px;
    }
};
</pre>
        <h2><A name="from_this">Obtaining a <code>shared_ptr</code> to <code>this</code></A></h2>
        <p>Sometimes it is needed to obtain a <code>shared_ptr</code> from <code>this</code>
            in a virtual member function under the assumption that <code>this</code> is 
            already managed by a <code>shared_ptr</code>. The transformations <A href="#from_raw">
                described in the previous technique</A> cannot be applied.</p>
        <P>A typical example:</P>
        <pre>class X
{
public:

    virtual void f() = 0;

protected:

    ~X() {}
};

class Y
{
public:

    virtual shared_ptr&lt;X&gt; getX() = 0;

protected:

    ~Y() {}
};

// --

class impl: public X, public Y
{
public:

    impl() { ... }

    virtual void f() { ... }

    virtual shared_ptr&lt;X&gt; getX()
    {
        shared_ptr&lt;X&gt; px(<i>???</i>);
        return px;
    }
};
</pre>
        <p>The solution is to keep a weak pointer to <code>this</code> as a member in <code>impl</code>:</p>
        <pre>class impl: public X, public Y
{
private:

    weak_ptr&lt;impl&gt; weak_this;

    impl(impl const &amp;);
    impl &amp; operator=(impl const &amp;);

    impl() { ... }

public:

    static shared_ptr&lt;impl&gt; create()
    {
        shared_ptr&lt;impl&gt; pi(new impl);
        pi-&gt;weak_this = pi;
        return pi;
    }

    virtual void f() { ... }

    virtual shared_ptr&lt;X&gt; getX()
    {
        shared_ptr&lt;X&gt; px(weak_this);
        return px;
    }
};
</pre>
        <p>The library now includes a helper class template <code><A href="enable_shared_from_this.html">
                    enable_shared_from_this</A></code> that can be used to encapsulate the 
            solution:</p>
        <pre>class impl: public X, public Y, public enable_shared_from_this&lt;impl&gt;
{
public:

    impl(impl const &amp;);
    impl &amp; operator=(impl const &amp;);

public:

    virtual void f() { ... }

    virtual shared_ptr&lt;X&gt; getX()
    {
        return shared_from_this();
    }
}
</pre>
        <p>Note that you no longer need to manually initialize the <code>weak_ptr</code> member 
            in <code><A href="enable_shared_from_this.html">enable_shared_from_this</A></code>. 
            Constructing a <code>shared_ptr</code> to <code>impl</code> takes care of that.</p>
        <h2><A name="handle">Using <code>shared_ptr</code> as a smart counted handle</A></h2>
        <p>Some library interfaces use opaque handles, a variation of the <A href="#incomplete">
                incomplete class technique</A> described above. An example:</p>
        <pre>typedef void * HANDLE;
HANDLE CreateProcess();
void CloseHandle(HANDLE);
</pre>
        <p>Instead of a raw pointer, it is possible to use <code>shared_ptr</code> as the 
            handle and get reference counting and automatic resource management for free:</p>
        <pre>typedef shared_ptr&lt;void&gt; handle;

handle createProcess()
{
    shared_ptr&lt;void&gt; pv(CreateProcess(), CloseHandle);
    return pv;
}
</pre>
        <h2><A name="on_block_exit">Using <code>shared_ptr</code> to execute code on block exit</A></h2>
        <p><code>shared_ptr&lt;void&gt;</code> can automatically execute cleanup code when 
            control leaves a scope.</p>
        <UL>
            <LI>
                Executing <code>f(p)</code>, where <code>p</code> is a pointer:</LI></UL>
        <pre>    shared_ptr&lt;void&gt; guard(p, f);
</pre>
        <UL>
            <LI>
                Executing arbitrary code: <code>f(x, y)</code>:</LI></UL>
        <pre>    shared_ptr&lt;void&gt; guard(static_cast&lt;void*&gt;(0), <A href="../bind/bind.html" >bind</A>(f, x, y));
</pre>
        <P>For a more thorough treatment, see the article "Simplify Your Exception-Safe 
            Code" by Andrei Alexandrescu and Petru Marginean, available online at <A href="http://www.cuj.com/experts/1812/alexandr.htm?topic=experts">
                http://www.cuj.com/experts/1812/alexandr.htm?topic=experts</A>.</P>
        <h2><A name="pvoid">Using <code>shared_ptr&lt;void&gt;</code> to hold an arbitrary 
                object</A></h2>
        <p><code>shared_ptr&lt;void&gt;</code> can act as a generic object pointer similar 
            to <code>void*</code>. When a <code>shared_ptr&lt;void&gt;</code> instance 
            constructed as:</p>
        <pre>    shared_ptr&lt;void&gt; pv(new X);
</pre>
        <p>is destroyed, it will correctly dispose of the <code>X</code> object by 
            executing <code>~X</code>.</p>
        <p>This propery can be used in much the same manner as a raw <code>void*</code> is 
            used to temporarily strip type information from an object pointer. A <code>shared_ptr&lt;void&gt;</code>
            can later be cast back to the correct type by using <code><A href="shared_ptr.htm#static_pointer_cast">
                    static_pointer_cast</A></code>.</p>
        <h2><A name="extra_data">Associating arbitrary data with heterogeneous <code>shared_ptr</code>
                instances</A></h2>
        <p><code>shared_ptr</code> and <code>weak_ptr</code> support <code>operator&lt;</code>
            comparisons required by standard associative containers such as <code>std::map</code>. 
            This can be used to non-intrusively associate arbitrary data with objects 
            managed by <code>shared_ptr</code>:</p>
        <pre>typedef int Data;

std::map&lt; shared_ptr&lt;void&gt;, Data &gt; userData;
// or std::map&lt; weak_ptr&lt;void&gt;, Data &gt; userData; to not affect the lifetime

shared_ptr&lt;X&gt; px(new X);
shared_ptr&lt;int&gt; pi(new int(3));

userData[px] = 42;
userData[pi] = 91;
</pre>
        <h2><A name="as_lock">Using <code>shared_ptr</code> as a CopyConstructible mutex lock</A></h2>
        <p>Sometimes it's necessary to return a mutex lock from a function, and a 
            noncopyable lock cannot be returned by value. It is possible to use <code>shared_ptr</code>
            as a mutex lock:</p>
        <pre>class mutex
{
public:

    void lock();
    void unlock();
};

shared_ptr&lt;mutex&gt; lock(mutex &amp; m)
{
    m.lock();
    return shared_ptr&lt;mutex&gt;(&amp;m, mem_fn(&amp;mutex::unlock));
}
</pre>
        <p>Better yet, the <code>shared_ptr</code> instance acting as a lock can be 
            encapsulated in a dedicated <code>shared_lock</code> class:</p>
        <pre>class shared_lock
{
private:

    shared_ptr&lt;void&gt; pv;

public:

    template&lt;class Mutex&gt; explicit shared_lock(Mutex &amp; m): pv((m.lock(), &amp;m), mem_fn(&amp;Mutex::unlock)) {}
};
</pre>
        <p><code>shared_lock</code> can now be used as:</p>
        <pre>    shared_lock lock(m);
</pre>
        <p>Note that <code>shared_lock</code> is not templated on the mutex type, thanks to <code>
                shared_ptr&lt;void&gt;</code>'s ability to hide type information.</p>
        <h2><A name="wrapper">Using <code>shared_ptr</code> to wrap member function calls</A></h2>
        <p><code>shared_ptr</code> implements the ownership semantics required from the <code>Wrap</code>/<code>CallProxy</code>
            scheme described in Bjarne Stroustrup's article "Wrapping C++ Member Function 
            Calls" (available online at <A href="http://www.stroustrup.com/wrapper.pdf">http://www.stroustrup.com/wrapper.pdf</A>). 
            An implementation is given below:</p>
        <pre>template&lt;class T&gt; class pointer
{
private:

    T * p_;

public:

    explicit pointer(T * p): p_(p)
    {
    }

    shared_ptr&lt;T&gt; operator-&gt;() const
    {
        p_-&gt;prefix();
        return shared_ptr&lt;T&gt;(p_, <A href="../bind/mem_fn.html" >mem_fn</A>(&amp;T::suffix));
    }
};

class X
{
private:

    void prefix();
    void suffix();
    friend class pointer&lt;X&gt;;
    
public:

    void f();
    void g();
};

int main()
{
    X x;

    pointer&lt;X&gt; px(&amp;x);

    px-&gt;f();
    px-&gt;g();
}
</pre>
        <h2><A name="delayed">Delayed deallocation</A></h2>
        <p>In some situations, a single <code>px.reset()</code> can trigger an expensive 
            deallocation in a performance-critical region:</p>
        <pre>class X; // ~X is expensive

class Y
{
    shared_ptr&lt;X&gt; px;

public:

    void f()
    {
        px.reset();
    }
};
</pre>
        <p>The solution is to postpone the potential deallocation by moving <code>px</code> 
            to a dedicated free list that can be periodically emptied when performance and 
            response times are not an issue:</p>
        <pre>vector&lt; shared_ptr&lt;void&gt; &gt; free_list;

class Y
{
    shared_ptr&lt;X&gt; px;

public:

    void f()
    {
        free_list.push_back(px);
        px.reset();
    }
};

// periodically invoke free_list.clear() when convenient
</pre>
        <p>Another variation is to move the free list logic to the construction point by 
            using a delayed deleter:</p>
        <pre>struct delayed_deleter
{
    template&lt;class T&gt; void operator()(T * p)
    {
        try
        {
            shared_ptr&lt;void&gt; pv(p);
            free_list.push_back(pv);
        }
        catch(...)
        {
        }
    }
};
</pre>
        <h2><A name="weak_without_shared">Weak pointers to objects not managed by a <code>shared_ptr</code></A></h2>
        <p>Make the object hold a <code>shared_ptr</code> to itself, using a <code>null_deleter</code>:</p>
        <pre>class X
{
private:

    shared_ptr&lt;X&gt; this_;
    int i_;

public:

    explicit X(int i): this_(this, null_deleter()), i_(i)
    {
    }

    // repeat in all constructors (including the copy constructor!)

    X(X const &amp; rhs): this_(this, null_deleter()), i_(rhs.i_)
    {
    }

    // do not forget to not assign this_ in the copy assignment

    X &amp; operator=(X const &amp; rhs)
    {
        i_ = rhs.i_;
    }

    weak_ptr&lt;X&gt; get_weak_ptr() const { return this_; }
};
</pre>
        <p>When the object's lifetime ends, <code>X::this_</code> will be destroyed, and 
            all weak pointers will automatically expire.</p>
        <hr>
        <p>$Date$</p>
        <p><small>Copyright &copy; 2003 Peter Dimov. Distributed under the Boost Software License, Version 
            1.0. See accompanying file <A href="../../LICENSE_1_0.txt">LICENSE_1_0.txt</A> or 
            copy at <A href="http://www.boost.org/LICENSE_1_0.txt">http://www.boost.org/LICENSE_1_0.txt</A>.</small></p>
    </body>
</html>