Files
serialization/doc/special.html
Robert Ramey db18e22b35 reconcilation with release 1.33.1
[SVN r31982]
2005-12-11 06:26:04 +00:00

395 lines
18 KiB
HTML

<!doctype HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN" "http://www.w3.org/TR/html4/loose.dtd">
<html>
<!--
(C) Copyright 2002-4 Robert Ramey - http://www.rrsd.com .
Use, modification and distribution is subject to the Boost Software
License, Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
http://www.boost.org/LICENSE_1_0.txt)
-->
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<link rel="stylesheet" type="text/css" href="../../../boost.css">
<link rel="stylesheet" type="text/css" href="style.css">
<title>Serialization - Special Considerations</title>
</head>
<body link="#0000ff" vlink="#800080">
<table border="0" cellpadding="7" cellspacing="0" width="100%" summary="header">
<tr>
<td valign="top" width="300">
<h3><a href="../../../index.htm"><img height="86" width="277" alt="C++ Boost" src="../../../boost.png" border="0"></a></h3>
</td>
<td valign="top">
<h1 align="center">Serialization</h1>
<h2 align="center">Special Considerations</h2>
</td>
</tr>
</table>
<hr>
<dl class="page-index">
<dt><a href="#objecttracking">Object Tracking</a>
<dt><a href="#export">Exporting Class Serialization</a>
<dt><a href="#classinfo">Class Information</a>
<dt><a href="#portability">Archive Portability</a>
<dl class="page-index">
<dt><a href="#numerics">Numerics</a>
<dt><a href="#traits">Traits</a>
</dl>
<dt><a href="#binary_archives">Binary Archives</a>
<dt><a href="#xml_archives">XML Archives</a>
<dt><a href="exceptions.html">Archive Exceptions</a>
<dt><a href="exception_safety.html">Exception Safety</a>
</dl>
<h3><a name="objecttracking">Object Tracking</a></h3>
Depending on how the class is used and other factors, serialized objects
may be tracked by memory address. This prevents the same object from being
written to or read from an archive multiple times. These stored addresses
can also be used to delete objects created during a loading process
that has been interrupted by throwing of an exception.
<p>
This could cause problems in
progams where the copies of different objects are saved from the same address.
<pre><code>
template&lt;class Archive&gt;
void save(boost::basic_oarchive &amp; ar, const unsigned int version) const
{
for(int i = 0; i &lt; 10; ++i){
A x = a[i];
ar &lt;&lt; x;
}
}
</code></pre>
In this case, the data to be saved exists on the stack. Each iteration
of the loop updates the value on the stack. So although the data changes
each iteration, the address of the data doesn't. If a[i] is an array of
objects being tracked by memory address, the library will skip storing
objects after the first as it will be assumed that objects at the same address
are really the same object.
<p>
To help detect such cases, output archive operators expect to be passed
<code style="white-space: normal">const</code> reference arguments.
<p>
Given this, the above code will invoke a compile time assertion.
The obvious fix in this example is to use
<pre><code>
template&lt;class Archive&gt;
void save(boost::basic_oarchive &amp; ar, const unsigned int version) const
{
for(int i = 0; i &lt; 10; ++i){
ar &lt;&lt; a[i];
}
}
</code></pre>
which will compile and run without problem.
The usage of <code style="white-space: normal">const</code> by the output archive operators
will ensure that the process of serialization doesn't
change the state of the objects being serialized. An attempt to do this
would constitute augmentation of the concept of saving of state with
some sort of non-obvious side effect. This would almost surely be a mistake
and a likely source of very subtle bugs.
<p>
Unfortunately, implementation issues currently prevent the detection of this kind of
error when the data item is wrapped as a name-value pair.
<p>
A similar problem can occur when different objects are loaded to and address
which is different from the final location:
<pre><code>
template&lt;class Archive&gt;
void load(boost::basic_oarchive &amp; ar, const unsigned int version) const
{
for(int i = 0; i &lt; 10; ++i){
A x;
ar &gt;&gt; x;
std::m_set.insert(x);
}
}
</code></pre>
In this case, the address of <code>x</code> is the one that is tracked rather than
the address of the new item added to the set. Left unaddressed
this will break the features that depend on tracking such as loading object through a pointer.
Subtle bugs will be introduced into the program. This can be
addressed by altering the above code thusly:
<pre><code>
template&lt;class Archive&gt;
void load(boost::basic_iarchive &amp; ar, const unsigned int version) const
{
for(int i = 0; i &lt; 10; ++i){
A x;
ar &gt;&gt; x;
std::pair&lt;std::set::const_iterator, bool&gt; result;
result = std::m_set.insert(x);
ar.reset_object_address(& (*result.first), &x);
}
}
</code></pre>
This will adjust the tracking information to reflect the final resting place of
the moved variable and thereby rectify the above problem.
<p>
If it is known a priori that no pointer
values are duplicated, overhead associated with object tracking can
be eliminated by setting the object tracking class serialization trait
appropriately.
<p>
By default, data types designated primitive by
<a target="detail" href="traits.html#level">Implementation Level</a>
class serialization trait are never tracked. If it is desired to
track a shared primitive object through a pointer (e.g. a
<code style="white-space: normal">long</code> used as a reference count), It should be wrapped
in a class/struct so that it is an identifiable type.
The alternative of changing the implementation level of a <code style="white-space: normal">long</code>
would affect all <code style="white-space: normal">long</code>s serialized in the whole
program - probably not what one would intend.
<p>
It is possible that we may want to track addresses even though
the object is never serialized through a pointer. For example,
a virtual base class need be saved/loaded only once. By setting
this serialization trait to <code style="white-space: normal">track_always</code>, we can suppress
redundant save/load operations.
<pre><code>
BOOST_CLASS_TRACKING(my_virtual_base_class, boost::serialization::track_always)
</code></pre>
<h3><a name="export">Exporting Class Serialization</a></h3>
<a target="detail" href="traits.html#export">Elsewhere</a> in this manual, we have described
<code style="white-space: normal">BOOST_CLASS_EXPORT</code>. This is used to make the serialization library aware
that code should be instantiated for serialization of a given class even though the
class hasn't been otherwise referred to by the program.
<p>
There are several ways <code style="white-space: normal">BOOST_CLASS_EXPORT</code> could have been
implemented.
<p>
One approach would be to instantiate serialization code for all archive classes included in the library.
This would add to each executable a large amount of code that is most likely never called.
Also it would needlessly slow down compilations of any program that uses the library. Finally,
the list of archives would be "built-in" to the library which would compilicate the addition of
new or custom archive classes.
<p>
Another approach would be for the library user to somehow explicitly instantiate which archive classes
code should be instantiated for each class to be serialized. Users would have to include
header files corresponding the archive classes to be instantiated.
The list of instantiated archive classes would have to be manually kept in sync with the
archive class headers actually included. This was considered burdensome and error prone.
<p>
This implementation of <code style="white-space: normal">BOOST_CLASS_EXPORT</code> works in the
following way:
<ul>
<li>All header modules of the form &lt;boost/archive/*archive.hpp&gt; are required to precede
the header module <a href="../../../boost/serialization/export.hpp" target="export_hpp">export.hpp</a>.
<li>The header <a href="../../../boost/serialization/export.hpp" target="export_hpp">export.hpp</a>
builds a list of archive classes whose header modules have been previously included.
It does this by checking to see which inclusion guard constants have been defined.
<li><code style="white-space: normal">BOOST_CLASS_EXPORT(my_class)</code> explicitly instantiates
serialization code for <code style="white-space: normal">my_class</code> for each archive in the list.
</ul>
Serialization code will be instantiated for a given archive class
if and only if the module that defines that archive class has been included in the program.
Given this, our program will contain all necessary code instantiations and no other.
<p>
For many styles of code organization this header sequencing requirement presents little problem.
Serialization code organized by class headers that are designed to be independent of archive
implementations will look something like the following:
<code><pre>
// A.hpp
// Note:to preserve independence from any particular archive implementation,
// no headers from &lt;boost/archive/...&gt; are included.
// Headers can be included in any order.
#include &lt;boost/serialization/...&gt;
#include &lt;boost/serialization/export.hpp&gt;
... // include other headers that A depends upon
class A {
...
};
BOOST_CLASS_EXPORT(A) // note: the export name of this class
</pre></code>
This style:
<ul>
<li>permits the header to include all aspects of the serialization implementation.
<li>permits the header to be included anywhere else as part of some other class declaration.
<li>reflects the concept of headers as a "library of types" which
can be used independently in other programs or other parts of the same program.
<li>reflects a fundamental principle of the serialization library design in that the
specification of serialization of any class is independent of any archive implementation.
</ul>
However, it might not always be possible or convenient to conform to the above style. Something
like the following might be required or preferred:
<code><pre>
// A.hpp
// headers can be included in any order
#include &lt;boost/archive/text_oarchive.hpp&gt;
#include &lt;boost/archive/text_iarchive.hpp&gt;
...
#include &lt;boost/serialization/...&gt;
...
// can't do the following because then A.hpp couldn't be included somewhere else
// #include &lt;boost/serialization/export.hpp&gt;
class A {
...
};
// can't do the following because export.hpp is not included !!
//BOOST_CLASS_EXPORT(A) // note: the export name of this class
</pre></code>
As noted in the comments, this would work. But
<code style="white-space: normal">#include &lt;.../export.hpp&gt;</code> can't be used
without conflicting with other modules which use
<code style="white-space: normal">#include &lt;.../*archive.hpp&gt;</code>. In this
case we can move the export to an implementation file:
<code><pre>
// A.cpp
#include "A.hpp"
...
// export.hpp header should be last;
#include &lt;boost/serialization/export.hpp&gt;
...
BOOST_CLASS_EXPORT(A)
...
</pre></code>
<h3><a name="classinfo">Class Information</a></h3>
By default, for each class serialized, class information is written to the archive.
This information includes version number, implementation level and tracking
behavior. This is necessary so that the archive can be correctly
deserialized even if a subsequent version of the program changes
some of the current trait values for a class. The space overhead for
this data is minimal. There is a little bit of runtime overhead
since each class has to be checked to see if it has already had its
class information included in the archive. In some cases, even this
might be considered too much. This extra overhead can be eliminated
by setting the
<a target="detail" href="traits.html#level">implementation level</a>
class trait to: <code style="white-space: normal">boost::serialization::object_serializable</code>.
<p>
<i>Turning off tracking and class information serialization will result
in pure template inline code that in principle could be optimised down
to a simple stream write/read.</i> Elimination of all serialization overhead
in this manner comes at a cost. Once archives are released to users, the
class serialization traits cannot be changed without invalidating the old
archives. Including the class information in the archive assures us
that they will be readable in the future even if the class definition
is revised. A light weight structure such as display pixel might be
declared in a header like this:
<pre><code>
#include &lt;boost/serialization/serialization.hpp&gt;
#include &lt;boost/serialization/level.hpp&gt;
#include &lt;boost/serialization/tracking.hpp&gt;
// a pixel is a light weight struct which is used in great numbers.
struct pixel
{
unsigned char red, green, blue;
template&lt;class Archive&gt;
void serialize(Archive &amp; ar, const unsigned int /* version */){
ar &lt;&lt; red &lt;&lt; green &lt;&lt; blue;
}
};
// elminate serialization overhead at the cost of
// never being able to increase the version.
BOOST_CLASS_IMPLEMENTATION(pixel, boost::serialization::object_serializable);
// eliminate object tracking (even if serialized through a pointer)
// at the risk of a programming error creating duplicate objects.
BOOST_CLASS_TRACKING(pixel, boost::serialization::track_never)
</code></pre>
<h3><a name="portability">Archive Portability</a></h3>
Several archive classes create their data in the form of text or portable a binary format.
It should be possible to save such an of such a class on one platform and load it on another.
This is subject to a couple of conditions.
<h4><a name="numerics">Numerics</a></h4>
The architecture of the machine reading the archive must be able hold the data
saved. For example, the gcc compiler reserves 4 bytes to store a variable of type
<code style="white-space: normal">wchar_t</code> while other compilers reserve only 2 bytes.
So its possible that a value could be written that couldn't be represented by the loading program. This is a
fairly obvious situation and easily handled by using the numeric types in
<a target="cstding" href="../../../boost/cstdint.hpp">&lt;boost/cstdint.hpp&gt;</a>
<h4><a name="traits">Traits</a></h4>
Another potential problem is illustrated by the following example:
<pre><code>
template&lt;class T&gt;
struct my_wrapper {
template&lt;class Archive&gt;
Archive & serialize ...
};
...
class my_class {
wchar_t a;
short unsigned b;
template<&lt;class Archive&gt;
Archive & serialize(Archive & ar, unsigned int version){
ar & my_wrapper(a);
ar & my_wrapper(b);
}
};
</code></pre>
If <code style="white-space: normal">my_wrapper</code> uses default serialization
traits there could be a problem. With the default traits, each time a new type is
added to the archive, bookkeeping information is added. So in this example, the
archive would include such bookkeeping information for
<code style="white-space: normal">my_wrapper&lt;wchar_t&gt;</code> and for
<code style="white-space: normal">my_wrapper&lt;short_unsigned&gt;</code>.
Or would it? What about compilers that treat
<code style="white-space: normal">wchar_t</code> as a
synonym for <code style="white-space: normal">unsigned short</code>?
In this case there is only one distinct type - not two. If archives are passed between
programs with compilers that differ in their treatment
of <code style="white-space: normal">wchar_t</code> the load operation will fail
in a catastrophic way.
<p>
One remedy for this is to assign serialization traits to the template
<code style="white-space: normal">my_template</code> such that class
information for instantiations of this template is never serialized. This
process is described <a target="detail" href="traits.html#templates">above</a> and
has been used for <a target="detail" href="wrappers.html#nvp"><strong>Name-Value Pairs</strong></a>.
Wrappers would typically be assigned such traits.
<p>
Another way to avoid this problem is to assign serialization traits
to all specializations of the template <code style="white-space: normal">my_wrapper</code>
for all primitive types so that class information is never saved. This is what has
been done for our implementation of serializations for STL collections.
<h3><a name="binary_archives">Binary Archives</a></h3>
Standard stream i/o on some systems will expand linefeed characters to carriage-return/linefeed
on output. This creates a problem for binary archives. The easiest way to handle this is to
open streams for binary archives in "binary mode" by using the flag
<code style="white-space: normal">ios::binary</code>. If this is not done, the archive generated
will be unreadable.
<p>
Unfortunately, no way has been found to detect this error before loading the archive. Debug builds
will assert when this is detected so that may be helpful in catching this error.
<h3><a name="xml_archives">XML Archives</a></h3>
XML archives present a somewhat special case.
XML format has a nested structure that maps well to the "recursive class member visitor" pattern
used by the serialization system. However, XML differs from other formats in that it
requires a name for each data member. Our goal is to add this information to the
class serialization specification while still permiting the the serialization code to be
used with any archive. This is achived by requiring that all data serialized to an XML archive
be serialized as a <a target="detail" href="wrappers.html#nvp">name-value pair</a>.
The first member is the name to be used as the XML tag for the
data item while the second is a reference to the data item itself. Any attempt to serialize data
not wrapped in a in a <a target="detail" href="wrappers.html#nvp">name-value pair</a> will
be trapped at compile time. The system is implemented in such a way that for other archive classes,
just the value portion of the data is serialized. The name portion is discarded during compilation.
So by always using <a target="detail" href="wrappers.html#nvp">name-value pairs</a>, it will
be guarenteed that all data can be serialized to all archive classes with maximum efficiency.
<h3><a href="exceptions.html">Archive Exceptions</a></h3>
<h3><a href="exception_safety.html">Exception Safety</a></h3>
<hr>
<p><i>&copy; Copyright <a href="http://www.rrsd.com">Robert Ramey</a> 2002-2004.
Distributed under the Boost Software License, Version 1.0. (See
accompanying file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
</i></p>
</body>
</html>