File: tutorial.html

package info (click to toggle)
boost 1.33.1-10
links: PTS
area: main
in suites: etch, etch-m68k
size: 100,948 kB
ctags: 145,103
sloc: cpp: 573,492; xml: 49,055; python: 15,626; ansic: 13,588; sh: 2,099; yacc: 858; makefile: 660; perl: 427; lex: 111; csh: 6
file content (373 lines) | stat: -rw-r--r-- 15,040 bytes
<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<meta name="generator" content="Docutils 0.3.9: http://docutils.sourceforge.net/" />
<title>Boost Pointer Container Library</title>
<link rel="stylesheet" href="default.css" type="text/css" />
</head>
<body>
<div class="document" id="boost-pointer-container-library">
<h1 class="title"><img alt="Boost" src="boost.png" /> Pointer Container Library</h1>
<h2 class="subtitle" id="tutorial">Tutorial</h2>
<p>The tutorial shows you the most simple usage of the
library. It is assumed that the reader is familiar
with the use of standard containers. Although
the tutorial is devided into sections, it is recommended
that you read it all from top to bottom.</p>
<ul class="simple">
<li><a class="reference" href="#basic-usage">Basic usage</a></li>
<li><a class="reference" href="#indirected-interface">Indirected interface</a></li>
<li><a class="reference" href="#sequence-containers">Sequence containers</a></li>
<li><a class="reference" href="#associative-containers">Associative containers</a></li>
<li><a class="reference" href="#null-values">Null values</a></li>
<li><a class="reference" href="#clonability">Clonability</a></li>
<li><a class="reference" href="#new-functions">New functions</a></li>
<li><a class="reference" href="#algorithms">Algorithms</a></li>
</ul>
<div class="section" id="basic-usage">
<h1><a name="basic-usage">Basic usage</a></h1>
<p>The most important aspect of a pointer container is that it manages
memory for you. This means that you in most cases do not need to worry
about deleting memory.</p>
<p>Let us assume that we have an OO-hierarchy of animals</p>
<pre class="literal-block">
class animal : <a class="reference" href="http://www.boost.org/libs/utility/utility.htm#Class_noncopyable">boost::noncopyable</a>
{
public:
    virtual      ~animal() {}
    virtual void eat() = 0;
    // ...
};

class mammal : public animal
{
    // ...
};

class bird : public animal
{
    // ...
};
</pre>
<p>Then the managing of the animals is straight-forward. Imagine a 
Zoo:</p>
<pre class="literal-block">
class zoo
{
    boost::ptr_vector&lt;animal&gt; the_animals;
public:

    void add_animal( animal* a )
    {
        the_animals.push_back( a );
    }
};
</pre>
<p>Notice how just pass the class name to the container; there
is no <tt class="docutils literal"><span class="pre">*</span></tt> to indicate it is a pointer.
With this declaration we can now say:</p>
<pre class="literal-block">
zoo the_zoo;
the_zoo.add_animal( new mammal(&quot;joe&quot;) );
the_zoo.add_animal( new bird(&quot;dodo&quot;) );
</pre>
<p>Thus we heap-allocate all elements of the container
and never rely on copy-semantics.</p>
</div>
<div class="section" id="indirected-interface">
<h1><a name="indirected-interface">Indirected interface</a></h1>
<p>As particular feature of the pointer containers is that
the query interface is indirected. For example,</p>
<pre class="literal-block">
boost::ptr_vector&lt;animal&gt; vec;
vec.push_back( new animal ); // you add it as pointer ...
vec[0].eat();                // but get a reference back
</pre>
<p>This indirection also happens to iterators, so</p>
<pre class="literal-block">
typedef std::vector&lt;animal*&gt; std_vec;
std_vec vec;
...
std_vec::iterator i = vec.begin();
(*i)-&gt;eat(); // '*' needed
</pre>
<p>now becomes</p>
<pre class="literal-block">
typedef boost::ptr_vector&lt;animal&gt;  ptr_vec;
ptr_vec vec;
ptr_vec::iterator i = vec.begin();
i-&gt;eat(); // no indirection needed
</pre>
</div>
<div class="section" id="sequence-containers">
<h1><a name="sequence-containers">Sequence containers</a></h1>
<p>The sequence containers used when you do not need to
keep an ordering on your elements. You can basically
expect all operations of the normal standard containers
to be available. So, for example, with a  <tt class="docutils literal"><span class="pre">ptr_deque</span></tt>
and <tt class="docutils literal"><span class="pre">ptr_list</span></tt> object you can say:</p>
<pre class="literal-block">
boost::ptr_deque&lt;animal&gt; deq;
deq.push_front( new animal );    
deq.pop_front();
</pre>
<p>because <tt class="docutils literal"><span class="pre">std::deque</span></tt> and <tt class="docutils literal"><span class="pre">std::list</span></tt> has <tt class="docutils literal"><span class="pre">push_front()</span></tt>
and <tt class="docutils literal"><span class="pre">pop_front</span></tt> members.</p>
<p>If the standard sequence support
random access, so does the pointer container; for example:</p>
<pre class="literal-block">
for( boost::ptr_deque&lt;animal&gt;::size_type i = 0u;
     i != deq.size(); ++i )
     deq[i].eat();
</pre>
<p>The <tt class="docutils literal"><span class="pre">ptr_vector</span></tt> also allows you to specify the size of
the buffer to allocate; for example</p>
<pre class="literal-block">
boost::ptr_vector&lt;animal&gt; animals( 10u );
</pre>
<p>will reserve room for 10 animals.</p>
</div>
<div class="section" id="associative-containers">
<h1><a name="associative-containers">Associative containers</a></h1>
<p>To keep an ordering on our animals, we could use a <tt class="docutils literal"><span class="pre">ptr_set</span></tt>:</p>
<pre class="literal-block">
boost::ptr_set&lt;animal&gt; set;
set.insert( new monkey(&quot;bobo&quot;) );
set.insert( new whale(&quot;anna&quot;) );
...
</pre>
<p>This requires that <tt class="docutils literal"><span class="pre">operator&lt;()</span></tt> is defined for animals. One
way to do this could be</p>
<pre class="literal-block">
inline bool operator&lt;( const animal&amp; l, const animal&amp; r )
{
    return l.name() &lt; r.name();
}
</pre>
<p>if we wanted to keep the animals sorted by name.</p>
<p>Maybe you want to keep all the animals in zoo ordered wrt.
their name, but it so happens that many animals have the
same name. We can then use a <tt class="docutils literal"><span class="pre">ptr_multimap</span></tt>:</p>
<pre class="literal-block">
typedef boost::ptr_multimap&lt;std::string,animal&gt; zoo_type;
zoo_type zoo;
std::string bobo = &quot;bobo&quot;,
            anna = &quot;anna&quot;;
zoo.insert( bobo, new monkey(bobo) );
zoo.insert( bobo, new elephant(bobo) );
zoo.insert( anna, new whale(anna) );
zoo.insert( anna, new emu(anna) );
</pre>
<p>Note that must create the key as an lvalue 
(due to exception-safety issues); the following would not 
have compiled</p>
<pre class="literal-block">
zoo.insert( &quot;bobo&quot;, // this is bad, but you get compile error
            new monkey(&quot;bobo&quot;) );
</pre>
<p>If a multimap is not needed, we can use <tt class="docutils literal"><span class="pre">operator[]()</span></tt>
to avoid the clumsiness:</p>
<pre class="literal-block">
boost::ptr_map&lt;std::string,animal&gt; animals;
animals[&quot;bobo&quot;].set_name(&quot;bobo&quot;);
</pre>
<p>This requires a default constructor for animals and
a function to do the initialization, in this case <tt class="docutils literal"><span class="pre">set_name()</span></tt>;</p>
</div>
<div class="section" id="null-values">
<h1><a name="null-values">Null values</a></h1>
<p>By default, if you try to insert null into a container, an exception
is thrown. If you want to allow nulls, then you must
say so explicitly when declaring the container variable</p>
<pre class="literal-block">
boost::ptr_vector&lt; boost::nullable&lt;animal&gt; &gt; animals_type;
animals_type animals;
...
animals.insert( animals.end(), new dodo(&quot;fido&quot;) );
animals.insert( animals.begin(), 0 ) // ok
</pre>
<p>Once you have inserted a null into the container, you must
always check if the value is null before accessing the object</p>
<pre class="literal-block">
for( animals_type::iterator i = animals.begin();
     i != animals.end(); ++i )
{
    if( !boost::is_null(i) ) // always check for validity
        i-&gt;eat();
}
</pre>
<p>If the container support random access, you may also check this as</p>
<pre class="literal-block">
for( animals_type::size_type i = 0u; 
     i != animals.size(); ++i )
{
    if( !animals.is_null(i) )
         animals[i].eat();
}
</pre>
<p>Note that it is meaningless to insert
null into <tt class="docutils literal"><span class="pre">ptr_set</span></tt> and <tt class="docutils literal"><span class="pre">ptr_multiset</span></tt>.</p>
</div>
<div class="section" id="clonability">
<h1><a name="clonability">Clonability</a></h1>
<p>In OO programming it is typical to prohibit copying of objects; the 
objects may sometimes be allowed to be clonable; for example,:</p>
<pre class="literal-block">
animal* animal::clone() const
{
    return do_clone(); // implemented by private virtual function
}
</pre>
<p>If the OO hierarchy thus allows cloning, we need to tell the 
pointer containers how cloning is to be done. This is simply
done by defining a free-standing function, <tt class="docutils literal"><span class="pre">new_clone()</span></tt>, 
in the same namespace as
the object hierarchy:</p>
<pre class="literal-block">
inline animal* new_clone( const animal&amp; a )
{
    return a.clone();
}
</pre>
<p>That is all, now a lot of functions in a pointer container
can exploit the clonability of the animal objects. For example</p>
<pre class="literal-block">
typedef boost::ptr_list&lt;animal&gt; zoo_type;
zoo_type zoo, another_zoo;
...
another_zoo.assign( zoo.begin(), zoo.end() );
</pre>
<p>will fill another zoo with clones of the first zoo. Similarly,
insert() can now insert clones into your pointer container</p>
<pre class="literal-block">
another_zoo.insert( another_zoo.begin(), zoo.begin(), zoo.end() );
</pre>
<p>The whole container can now also be cloned</p>
<pre class="literal-block">
zoo_type yet_another_zoo = zoo.clone();
</pre>
</div>
<div class="section" id="new-functions">
<h1><a name="new-functions">New functions</a></h1>
<p>Given that we know we are working with pointers, a few new functions
make sense. For example, say you want to remove an
animal from the zoo</p>
<pre class="literal-block">
zoo_type::auto_type the_animal = zoo.release( zoo.begin() );
the_animal-&gt;eat();
animal* the_animal_ptr = the_animal.release(); // now this is not deleted
zoo.release(2); // for random access containers
</pre>
<p>You can think of <tt class="docutils literal"><span class="pre">auto_type</span></tt> as a non-copyable form of 
<tt class="docutils literal"><span class="pre">std::auto_ptr</span></tt>. Notice that when you release an object, the
pointer is removed from the container and the containers size
shrinks. You can also release the entire container if you
want to return it from a function</p>
<pre class="literal-block">
std::auto_ptr&lt; boost::ptr_deque&lt;animal&gt; &gt; get_zoo()
{
    boost::ptr_deque&lt;animal&gt;  result;
    ...
    return result.release(); // give up ownership
}
...
boost::ptr_deque&lt;animal&gt; animals = get_zoo();    
</pre>
<p>Let us assume we want to move an animal object from
one zoo to another. In other words, we want to move the 
animal and the responsibility of it to another zoo</p>
<pre class="literal-block">
another_zoo.transfer( another_zoo.end(), // insert before end 
                      zoo.begin(),       // insert this animal ...
                      zoo );             // from this container
</pre>
<p>This kind of &quot;move-semantics&quot; is different from
normal value-based containers. You can think of <tt class="docutils literal"><span class="pre">transfer()</span></tt>
as the same as <tt class="docutils literal"><span class="pre">splice()</span></tt> on <tt class="docutils literal"><span class="pre">std::list</span></tt>.</p>
<p>If you want to replace an element, you can easily do so</p>
<pre class="literal-block">
zoo_type::auto_type old_animal = zoo.replace( zoo.begin(), new monkey(&quot;bibi&quot;) ); 
zoo.replace( 2, old_animal.release() ); // for random access containers
</pre>
<p>A map is a little different to iterator over than standard maps.
Now we say</p>
<pre class="literal-block">
typedef boost::ptr_map&lt;std::string, boost::nullable&lt;animal&gt; &gt; animal_map;
animal_map map;
...
for( animal_map::iterator i = map.begin();
     i != map.end(); ++i )
{
    std::cout &lt;&lt; &quot;\n key: &quot; &lt;&lt; i.key();
    std::cout &lt;&lt; &quot;\n age: &quot;;
    
    if( boost::is_null(i) )
        std::cout &lt;&lt; &quot;unknown&quot;;
    else
        std::cout &lt;&lt; i-&gt;age(); 
 }
</pre>
<p>Maps can also be indexed with bounds-checking</p>
<pre class="literal-block">
try
{
    animal&amp; bobo = map.at(&quot;bobo&quot;);
}
catch( boost::bad_ptr_container_operation&amp; e )
{
    // &quot;bobo&quot; not found
}        
</pre>
</div>
<div class="section" id="algorithms">
<h1><a name="algorithms">Algorithms</a></h1>
<p>Unfortunately it is not possible to use pointer containers with
mutating algorithms from the standard library. However,
the most useful ones
are instead provided as member functions:</p>
<pre class="literal-block">
boost::ptr_vector&lt;animal&gt; zoo;
...
zoo.sort();                               // assume 'bool operator&lt;( const animal&amp;, const animal&amp; )'
zoo.sort( std::less&lt;animal&gt;() );          // the same, notice no '*' is present
zoo.sort( zoo.begin(), zoo.begin() + 5 ); // sort selected range
</pre>
<p>Notice that predicates are automatically wrapped in an <a class="reference" href="indirect_fun.html">indirect_fun</a> object.</p>
<p>You can remove equal and adjacent elements using <tt class="docutils literal"><span class="pre">unique()</span></tt>:</p>
<pre class="literal-block">
zoo.unique();                             // assume 'bool operator==( const animal&amp;, const animal&amp; )'
zoo.unique( zoo.begin(), zoo.begin() + 5, my_comparison_predicate() ); 
</pre>
<p>If you just want to remove certain elements, use <tt class="docutils literal"><span class="pre">erase_if</span></tt>:</p>
<pre class="literal-block">
zoo.erase_if( my_predicate() );
</pre>
<p>Finally you may want to merge together two sorted containers:</p>
<pre class="literal-block">
boost::ptr_vector&lt;animal&gt; another_zoo = ...;
another_zoo.sort();                      // sorted wrt. to same order as 'zoo'
zoo.merge( another_zoo );
BOOST_ASSERT( another_zoo.empty() );    
</pre>
<p>That is all; now you have learned all the basics!</p>
<p><strong>Navigate</strong></p>
<blockquote>
<ul class="simple">
<li><a class="reference" href="ptr_container.html">home</a></li>
<li><a class="reference" href="examples.html">examples</a></li>
</ul>
</blockquote>
<table class="docutils field-list" frame="void" rules="none">
<col class="field-name" />
<col class="field-body" />
<tbody valign="top">
<tr class="field"><th class="field-name">copyright:</th><td class="field-body">Thorsten Ottosen 2004-2005.</td>
</tr>
</tbody>
</table>
</div>
</div>
</body>
</html>