Subversion Repositories openrisc

<section id="std.util.memory.shared_ptr" xreflabel="shared_ptr">
<?dbhtml filename="shared_ptr.html"?>

<sectioninfo>
  <keywordset>
    <keyword>
      ISO C++
    </keyword>
    <keyword>
      shared_ptr
    </keyword>
  </keywordset>
</sectioninfo>

<title>shared_ptr</title>

<para>
The shared_ptr class template stores a pointer, usually obtained via new,
and implements shared ownership semantics.
</para>

<section id="shared_ptr.req">
<title>Requirements</title>

  <para>
  </para>

  <para>
    The standard deliberately doesn't require a reference-counted
    implementation, allowing other techniques such as a
    circular-linked-list.
  </para>

  <para>
    At the time of writing the C++0x working paper doesn't mention how
    threads affect shared_ptr, but it is likely to follow the existing
    practice set by <classname>boost::shared_ptr</classname>.  The
    shared_ptr in libstdc++ is derived from Boost's, so the same rules
    apply.
  </para>

  <para>
  </para>
</section>

<section id="shared_ptr.design_issues">
<title>Design Issues</title>


  <para>
The <classname>shared_ptr</classname> code is kindly donated to GCC by the Boost
project and the original authors of the code. The basic design and
algorithms are from Boost, the notes below describe details specific to
the GCC implementation. Names have been uglified in this implementation,
but the design should be recognisable to anyone familiar with the Boost
1.32 shared_ptr.
  </para>

  <para>
The basic design is an abstract base class, <code>_Sp_counted_base</code> that
does the reference-counting and calls virtual functions when the count
drops to zero.
Derived classes override those functions to destroy resources in a context
where the correct dynamic type is known. This is an application of the
technique known as type erasure.
  </para>

</section>

<section id="shared_ptr.impl">
<title>Implementation</title>

  <section>
    <title>Class Hierarchy</title>

    <para>
A <classname>shared_ptr&lt;T&gt;</classname> contains a pointer of
type <type>T*</type> and an object of type
<classname>__shared_count</classname>. The shared_count contains a
pointer of type <type>_Sp_counted_base*</type> which points to the
object that maintains the reference-counts and destroys the managed
resource.
    </para>

<variablelist>

<varlistentry>
  <term><classname>_Sp_counted_base&lt;Lp&gt;</classname></term>
  <listitem>
    <para>
The base of the hierarchy is parameterized on the lock policy alone.
_Sp_counted_base doesn't depend on the type of pointer being managed,
it only maintains the reference counts and calls virtual functions when
the counts drop to zero. The managed object is destroyed when the last
strong reference is dropped, but the _Sp_counted_base itself must exist
until the last weak reference is dropped.
    </para>
  </listitem>
</varlistentry>

<varlistentry>
  <term><classname>_Sp_counted_base_impl&lt;Ptr, Deleter, Lp&gt;</classname></term>
  <listitem>
    <para>
Inherits from _Sp_counted_base and stores a pointer of type <type>Ptr</type>
and a deleter of type <code>Deleter</code>.  <code>_Sp_deleter</code> is
used when the user doesn't supply a custom deleter. Unlike Boost's, this
default deleter is not "checked" because GCC already issues a warning if
<function>delete</function> is used with an incomplete type.
This is the only derived type used by <classname>shared_ptr&lt;Ptr&gt;</classname>
and it is never used by <classname>shared_ptr</classname>, which uses one of
the following types, depending on how the shared_ptr is constructed.
    </para>
  </listitem>
</varlistentry>

<varlistentry>
  <term><classname>_Sp_counted_ptr&lt;Ptr, Lp&gt;</classname></term>
  <listitem>
    <para>
Inherits from _Sp_counted_base and stores a pointer of type <type>Ptr</type>,
which is passed to <function>delete</function> when the last reference is dropped.
This is the simplest form and is used when there is no custom deleter or
allocator.
    </para>
  </listitem>
</varlistentry>

<varlistentry>
  <term><classname>_Sp_counted_deleter&lt;Ptr, Deleter, Alloc&gt;</classname></term>
  <listitem>
    <para>
Inherits from _Sp_counted_ptr and adds support for custom deleter and
allocator. Empty Base Optimization is used for the allocator. This class
is used even when the user only provides a custom deleter, in which case
<classname>allocator</classname> is used as the allocator.
    </para>
  </listitem>
</varlistentry>

<varlistentry>
  <term><classname>_Sp_counted_ptr_inplace&lt;Tp, Alloc, Lp&gt;</classname></term>
  <listitem>
    <para>
Used by <code>allocate_shared</code> and <code>make_shared</code>.
Contains aligned storage to hold an object of type <type>Tp</type>,
which is constructed in-place with placement <function>new</function>.
Has a variadic template constructor allowing any number of arguments to
be forwarded to <type>Tp</type>'s constructor.
Unlike the other <classname>_Sp_counted_*</classname> classes, this one is parameterized on the
type of object, not the type of pointer; this is purely a convenience
that simplifies the implementation slightly.
    </para>
  </listitem>
</varlistentry>

</variablelist>

  </section>

  <section>
    <title>Thread Safety</title>

    <para>
The interface of <classname>tr1::shared_ptr</classname> was extended for C++0x
with support for rvalue-references and the other features from
N2351. As with other libstdc++ headers shared by TR1 and C++0x,
boost_shared_ptr.h uses conditional compilation, based on the macros
<constant>_GLIBCXX_INCLUDE_AS_CXX0X</constant> and
<constant>_GLIBCXX_INCLUDE_AS_TR1</constant>, to enable and disable
features.
    </para>

    <para>
C++0x-only features are: rvalue-ref/move support, allocator support,
aliasing constructor, make_shared &amp; allocate_shared. Additionally,
the constructors taking <classname>auto_ptr</classname> parameters are
deprecated in C++0x mode.
    </para>

<para>
The
<ulink url="http://boost.org/libs/smart_ptr/shared_ptr.htm#ThreadSafety">Thread
Safety</ulink> section of the Boost shared_ptr documentation says "shared_ptr
objects offer the same level of thread safety as built-in types."
The implementation must ensure that concurrent updates to separate shared_ptr
instances are correct even when those instances share a reference count e.g.
</para>

<programlisting>
shared_ptr&lt;A&gt; a(new A);
shared_ptr&lt;A&gt; b(a);

// Thread 1     // Thread 2
   a.reset();      b.reset();
</programlisting>

<para>
The dynamically-allocated object must be destroyed by exactly one of the
threads. Weak references make things even more interesting.
The shared state used to implement shared_ptr must be transparent to the
user and invariants must be preserved at all times.
The key pieces of shared state are the strong and weak reference counts.
Updates to these need to be atomic and visible to all threads to ensure
correct cleanup of the managed resource (which is, after all, shared_ptr's
job!)
On multi-processor systems memory synchronisation may be needed so that
reference-count updates and the destruction of the managed resource are
race-free.
</para>

<para>
The function <function>_Sp_counted_base::_M_add_ref_lock()</function>, called when
obtaining a shared_ptr from a weak_ptr, has to test if the managed
resource still exists and either increment the reference count or throw
<classname>bad_weak_ptr</classname>.
In a multi-threaded program there is a potential race condition if the last
reference is dropped (and the managed resource destroyed) between testing
the reference count and incrementing it, which could result in a shared_ptr
pointing to invalid memory.
</para>
<para>
The Boost shared_ptr (as used in GCC) features a clever lock-free
algorithm to avoid the race condition, but this relies on the
processor supporting an atomic <emphasis>Compare-And-Swap</emphasis>
instruction. For other platforms there are fall-backs using mutex
locks.  Boost (as of version 1.35) includes several different
implementations and the preprocessor selects one based on the
compiler, standard library, platform etc. For the version of
shared_ptr in libstdc++ the compiler and library are fixed, which
makes things much simpler: we have an atomic CAS or we don't, see Lock
Policy below for details.
</para>

  </section>

  <section>
    <title>Selecting Lock Policy</title>

    <para>
    </para>

    <para>
There is a single <classname>_Sp_counted_base</classname> class,
which is a template parameterized on the enum
<type>__gnu_cxx::_Lock_policy</type>.  The entire family of classes is
parameterized on the lock policy, right up to
<classname>__shared_ptr</classname>, <classname>__weak_ptr</classname> and
<classname>__enable_shared_from_this</classname>. The actual
<classname>std::shared_ptr</classname> class inherits from
<classname>__shared_ptr</classname> with the lock policy parameter
selected automatically based on the thread model and platform that
libstdc++ is configured for, so that the best available template
specialization will be used. This design is necessary because it would
not be conforming for <classname>shared_ptr</classname> to have an
extra template parameter, even if it had a default value.  The
available policies are:
    </para>

   <orderedlist>
     <listitem>
       <para>
       <type>_S_Atomic</type>
       </para>
       <para>
Selected when GCC supports a builtin atomic compare-and-swap operation
on the target processor (see <ulink url="http://gcc.gnu.org/onlinedocs/gcc/Atomic-Builtins.html">Atomic
Builtins</ulink>.)  The reference counts are maintained using a lock-free
algorithm and GCC's atomic builtins, which provide the required memory
synchronisation.
       </para>
     </listitem>

     <listitem>
       <para>
       <type>_S_Mutex</type>
       </para>
       <para>
The _Sp_counted_base specialization for this policy contains a mutex,
which is locked in add_ref_lock(). This policy is used when GCC's atomic
builtins aren't available so explicit memory barriers are needed in places.
       </para>
     </listitem>

     <listitem>
       <para>
       <type>_S_Single</type>
       </para>
       <para>
This policy uses a non-reentrant add_ref_lock() with no locking. It is
used when libstdc++ is built without <literal>--enable-threads</literal>.
       </para>
     </listitem>

   </orderedlist>
     <para>
       For all three policies, reference count increments and
       decrements are done via the functions in
       <filename>ext/atomicity.h</filename>, which detect if the program
       is multi-threaded.  If only one thread of execution exists in
       the program then less expensive non-atomic operations are used.
     </para>
  </section>

  <section>
    <title>Dual C++0x and TR1 Implementation</title>

<para>
The classes derived from <classname>_Sp_counted_base</classname> (see Class Hierarchy
below) and <classname>__shared_count</classname> are implemented separately for C++0x
and TR1, in <filename>bits/boost_sp_shared_count.h</filename> and
<filename>tr1/boost_sp_shared_count.h</filename> respectively.  All other classes
including <classname>_Sp_counted_base</classname> are shared by both implementations.
</para>

<para>
The TR1 implementation is considered relatively stable, so is unlikely to
change unless bug fixes require it.  If the code that is common to both
C++0x and TR1 modes needs to diverge further then it might be necessary to
duplicate additional classes and only make changes to the C++0x versions.
</para>
</section>

<section>
<title>Related functions and classes</title>

<variablelist>

<varlistentry>
  <term><code>dynamic_pointer_cast</code>, <code>static_pointer_cast</code>,
<code>const_pointer_cast</code></term>
  <listitem>
    <para>
As noted in N2351, these functions can be implemented non-intrusively using
the alias constructor.  However the aliasing constructor is only available
in C++0x mode, so in TR1 mode these casts rely on three non-standard
constructors in shared_ptr and __shared_ptr.
In C++0x mode these constructors and the related tag types are not needed.
    </para>
  </listitem>
</varlistentry>

<varlistentry>
  <term><code>enable_shared_from_this</code></term>
  <listitem>
    <para>
The clever overload to detect a base class of type
<code>enable_shared_from_this</code> comes straight from Boost.
There is an extra overload for <code>__enable_shared_from_this</code> to
work smoothly with <code>__shared_ptr&lt;Tp, Lp&gt;</code> using any lock
policy.
    </para>
  </listitem>
</varlistentry>

<varlistentry>
  <term><code>make_shared</code>, <code>allocate_shared</code></term>
  <listitem>
    <para>
<code>make_shared</code> simply forwards to <code>allocate_shared</code>
with <code>std::allocator</code> as the allocator.
Although these functions can be implemented non-intrusively using the
alias constructor, if they have access to the implementation then it is
possible to save storage and reduce the number of heap allocations. The
newly constructed object and the _Sp_counted_* can be allocated in a single
block and the standard says implementations are "encouraged, but not required,"
to do so. This implementation provides additional non-standard constructors
(selected with the type <code>_Sp_make_shared_tag</code>) which create an
object of type <code>_Sp_counted_ptr_inplace</code> to hold the new object.
The returned <code>shared_ptr&lt;A&gt;</code> needs to know the address of the
new <code>A</code> object embedded in the <code>_Sp_counted_ptr_inplace</code>,
but it has no way to access it.
This implementation uses a "covert channel" to return the address of the
embedded object when <code>get_deleter&lt;_Sp_make_shared_tag&gt;()</code>
is called.  Users should not try to use this.
As well as the extra constructors, this implementation also needs some
members of _Sp_counted_deleter to be protected where they could otherwise
be private.
    </para>
  </listitem>
</varlistentry>

</variablelist>

</section>

</section>

<!--- XXX
     <listitem>
       <type>_Sp_counted_base&lt;Lp&gt;</type>
       <para>
The base of the hierarchy is parameterized on the lock policy alone.
_Sp_counted_base doesn't depend on the type of pointer being managed,
it only maintains the reference counts and calls virtual functions when
the counts drop to zero. The managed object is destroyed when the last
strong reference is dropped, but the _Sp_counted_base itself must exist
until the last weak reference is dropped.
       </para>
     </listitem>

     <listitem>
       <type>_Sp_counted_base_impl&lt;Ptr, Deleter, Lp&gt;</type>
       <para>
Inherits from _Sp_counted_base and stores a pointer of type <code>Ptr</code>
and a deleter of type <code>Deleter</code>.  <code>_Sp_deleter</code> is
used when the user doesn't supply a custom deleter. Unlike Boost's, this
default deleter is not "checked" because GCC already issues a warning if
<code>delete</code> is used with an incomplete type.
This is the only derived type used by <code>tr1::shared_ptr&lt;Ptr&gt;</code>
and it is never used by <code>std::shared_ptr</code>, which uses one of
the following types, depending on how the shared_ptr is constructed.
       </para>
     </listitem>
-->

<section id="shared_ptr.using">
<title>Use</title>

  <section>
    <title>Examples</title>
    <para>
      Examples of use can be found in the testsuite, under
      <filename class="directory">testsuite/tr1/2_general_utilities/shared_ptr</filename>.
    </para>
  </section>

  <section>
    <title>Unresolved Issues</title>
    <para>
      The resolution to C++ Standard Library issue <ulink url="http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-active.html#674">674</ulink>,
      "shared_ptr interface changes for consistency with N1856" will
      need to be implemented after it is accepted into the working
      paper. Issue <ulink url="http://www.open-std.org/jtc1/sc22/wg21/docs/lwg-active.html#743">743</ulink>
      might also require changes.
    </para>

    <para>
      The <type>_S_single</type> policy uses atomics when used in MT
      code, because it uses the same dispatcher functions that check
      <function>__gthread_active_p()</function>. This could be
      addressed by providing template specialisations for some members
      of <classname>_Sp_counted_base&lt;_S_single&gt;</classname>.
    </para>

    <para>
      Unlike Boost, this implementation does not use separate classes
      for the pointer+deleter and pointer+deleter+allocator cases in
      C++0x mode, combining both into _Sp_counted_deleter and using
      <classname>allocator</classname> when the user doesn't specify
      an allocator.  If it was found to be beneficial an additional
      class could easily be added.  With the current implementation,
      the _Sp_counted_deleter and __shared_count constructors taking a
      custom deleter but no allocator are technically redundant and
      could be removed, changing callers to always specify an
      allocator. If a separate pointer+deleter class was added the
      __shared_count constructor would be needed, so it has been kept
      for now.
    </para>

    <para>
      The hack used to get the address of the managed object from
      <function>_Sp_counted_ptr_inplace::_M_get_deleter()</function>
      is accessible to users. This could be prevented if
      <function>get_deleter&lt;_Sp_make_shared_tag&gt;()</function>
      always returned NULL, since the hack only needs to work at a
      lower level, not in the public API. This wouldn't be difficult,
      but hasn't been done since there is no danger of accidental
      misuse: users already know they are relying on unsupported
      features if they refer to implementation details such as
      _Sp_make_shared_tag.
    </para>

    <para>
      tr1::_Sp_deleter could be a private member of tr1::__shared_count but it
      would alter the ABI.
    </para>

    <para>
      Exposing the alias constructor in TR1 mode could simplify the
      *_pointer_cast functions.  Constructor could be private in TR1
      mode, with the cast functions as friends.
    </para>
  </section>

</section>

<section id="shared_ptr.ack">
<title>Acknowledgments</title>

  <para>
    The original authors of the Boost shared_ptr, which is really nice
    code to work with, Peter Dimov in particular for his help and
    invaluable advice on thread safety.  Phillip Jordan and Paolo
    Carlini for the lock policy implementation.
  </para>

</section>

<bibliography id="shared_ptr.biblio">
<title>Bibliography</title>

  <biblioentry>
    <biblioid class="uri">
      <ulink url="http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2351.htm">
        <citetitle>
          Improving shared_ptr for C++0x, Revision 2
        </citetitle>
      </ulink>
    </biblioid>
    <subtitle>
      N2351
    </subtitle>
  </biblioentry>

  <biblioentry>
    <biblioid class="uri">
      <ulink url="http://open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2456.html">
        <citetitle>
          C++ Standard Library Active Issues List
        </citetitle>
      </ulink>
    </biblioid>
    <subtitle>
      N2456
    </subtitle>
  </biblioentry>

  <biblioentry>
    <biblioid class="uri">
      <ulink url="http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2461.pdf">
        <citetitle>
          Working Draft, Standard for Programming Language C++
        </citetitle>
      </ulink>
    </biblioid>
    <subtitle>
      N2461
    </subtitle>
  </biblioentry>

  <biblioentry>
    <biblioid class="uri">
      <ulink url="http://boost.org/libs/smart_ptr/shared_ptr.htm">shared_ptr
        <citetitle>
          Boost C++ Libraries documentation, shared_ptr
        </citetitle>
      </ulink>
    </biblioid>
    <subtitle>
      N2461
    </subtitle>
  </biblioentry>

</bibliography>

</section>