Planet
navi homePPSaboutscreenshotsdownloaddevelopmentforum

source: downloads/boost_1_34_1/libs/python/doc/internals.html @ 29

Last change on this file since 29 was 29, checked in by landauf, 17 years ago

updated boost from 1_33_1 to 1_34_1
File size: 10.0 KB
Line 
1<?xml version="1.0" encoding="utf-8" ?>
2<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
3<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
4<head>
5<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
6<meta name="generator" content="Docutils 0.3.0: http://docutils.sourceforge.net/" />
7<title>Boost.Python Internals Boost</title>
8<link rel="stylesheet" href="../../../rst.css" type="text/css" />
9</head>
10<body>
11<div class="document" id="boost-python-internals-logo">
12<h1 class="title"><a class="reference" href="index.html">Boost.Python</a> Internals <a class="reference" href="../../../index.htm"><img alt="Boost" src="../../../boost.png" /></a></h1>
13<div class="section" id="a-conversation-between-brett-calcott-and-david-abrahams">
14<h1><a name="a-conversation-between-brett-calcott-and-david-abrahams">A conversation between Brett Calcott and David Abrahams</a></h1>
15<table class="field-list" frame="void" rules="none">
16<col class="field-name" />
17<col class="field-body" />
18<tbody valign="top">
19<tr class="field"><th class="field-name">copyright:</th><td class="field-body">Copyright David Abrahams and Brett Calcott 2003. See
20accompanying <a class="reference" href="../../../LICENSE_1_0.txt">license</a> for terms of use.</td>
21</tr>
22</tbody>
23</table>
24<p>In both of these cases, I'm quite capable of reading code - but the
25thing I don't get from scanning the source is a sense of the
26architecture, both structurally, and temporally (er, I mean in what
27order things go on).</p>
28<ol class="arabic">
29<li><p class="first">What happens when you do the following:</p>
30<pre class="literal-block">
31struct boring {};
32...etc...
33class_&lt;boring&gt;(&quot;boring&quot;)
34    ;
35</pre>
36</li>
37</ol>
38<p>There seems to be a fair bit going on.</p>
39<blockquote>
40<ul class="simple">
41<li>Python needs a new ClassType to be registered.</li>
42<li>We need to construct a new type that can hold our boring struct.</li>
43<li>Inward and outward converters need to be registered for the type.</li>
44</ul>
45</blockquote>
46<p>Can you gesture in the general direction where these things are done?</p>
47<blockquote>
48<p>I only have time for a &quot;off-the-top-of-my-head&quot; answer at the moment;
49I suggest you step through the code with a debugger after reading this
50to see how it works, fill in details, and make sure I didn't forget
51anything.</p>
52<blockquote>
53<p>A new (Python) subclass of Boost.Python.Instance (see
54libs/python/src/object/class.cpp) is created by invoking
55Boost.Python.class, the metatype:</p>
56<pre class="literal-block">
57&gt;&gt;&gt; boring = Boost.Python.class(
58...     'boring'
59...   , bases_tuple       # in this case, just ()
60...   , {
61...         '__module__' : module_name
62...       , '__doc__' : doc_string # optional
63...     }
64... )
65</pre>
66<p>A handle to this object is stuck in the m_class_object field
67of the registration associated with <tt class="literal"><span class="pre">typeid(boring)</span></tt>.  The
68registry will keep that object alive forever, even if you
69wipe out the 'boring' attribute of the extension module
70(probably not a good thing).</p>
71<p>Because you didn't specify <tt class="literal"><span class="pre">class&lt;boring,</span> <span class="pre">non_copyable,</span>
72<span class="pre">...&gt;</span></tt>, a to-python converter for boring is registered which
73copies its argument into a value_holder held by the the
74Python boring object.</p>
75<p>Because you didn't specify <tt class="literal"><span class="pre">class&lt;boring</span> <span class="pre">...&gt;(no_init)</span></tt>,
76an <tt class="literal"><span class="pre">__init__</span></tt> function object is added to the class
77dictionary which default-constructs a boring in a
78value_holder (because you didn't specify some smart pointer
79or derived wrapper class as a holder) held by the Python
80boring object.</p>
81<p><tt class="literal"><span class="pre">register_class_from_python</span></tt> is used to register a
82from-python converter for <tt class="literal"><span class="pre">shared_ptr&lt;boring&gt;</span></tt>.
83<tt class="literal"><span class="pre">boost::shared_ptr</span></tt>s are special among smart pointers
84because their Deleter argument can be made to manage the
85whole Python object, not just the C++ object it contains, no
86matter how the C++ object is held.</p>
87<p>If there were any <tt class="literal"><span class="pre">bases&lt;&gt;</span></tt>, we'd also be registering the
88relationship between these base classes and boring in the
89up/down cast graph (<tt class="literal"><span class="pre">inheritance.[hpp/cpp]</span></tt>).</p>
90<p>In earlier versions of the code, we'd be registering lvalue
91from-python converters for the class here, but now
92from-python conversion for wrapped classes is handled as a
93special case, before consulting the registry, if the source
94Python object's metaclass is the Boost.Python metaclass.</p>
95<p>Hmm, that from-python converter probably ought to be handled
96the way class converters are, with no explicit conversions
97registered.</p>
98</blockquote>
99</blockquote>
100<ol class="arabic" start="2">
101<li><p class="first">Can you give a brief overview of the data structures that are
102present in the registry</p>
103<blockquote>
104<p>The registry is simple: it's just a map from typeid -&gt;
105registration (see boost/python/converter/registrations.hpp).
106<tt class="literal"><span class="pre">lvalue_chain</span></tt> and <tt class="literal"><span class="pre">rvalue_chain</span></tt> are simple endogenous
107linked lists.</p>
108<p>If you want to know more, just ask.</p>
109<p>If you want to know about the cast graph, ask me something specific in
110a separate message.</p>
111</blockquote>
112<p>and an overview of the process that happens as a type makes its
113way from c++ to python and back again.</p>
114</li>
115</ol>
116<blockquote>
117<p>Big subject.  I suggest some background reading: look for relevant
118info in the LLNL progress reports and the messages they link to.
119Also,</p>
120<blockquote>
121<p><a class="reference" href="http://mail.python.org/pipermail/c++-sig/2002-May/001023.html">http://mail.python.org/pipermail/c++-sig/2002-May/001023.html</a></p>
122<p><a class="reference" href="http://mail.python.org/pipermail/c++-sig/2002-December/003115.html">http://mail.python.org/pipermail/c++-sig/2002-December/003115.html</a></p>
123<p><a class="reference" href="http://aspn.activestate.com/ASPN/Mail/Message/1280898">http://aspn.activestate.com/ASPN/Mail/Message/1280898</a></p>
124<p><a class="reference" href="http://mail.python.org/pipermail/c++-sig/2002-July/001755.html">http://mail.python.org/pipermail/c++-sig/2002-July/001755.html</a></p>
125</blockquote>
126<p>from c++ to python:</p>
127<blockquote>
128<p>It depends on the type and the call policies in use or, for
129<tt class="literal"><span class="pre">call&lt;&gt;(...)</span></tt>, <tt class="literal"><span class="pre">call_method&lt;&gt;(...)</span></tt>, or <tt class="literal"><span class="pre">object(...)</span></tt>, if
130<tt class="literal"><span class="pre">ref</span></tt> or <tt class="literal"><span class="pre">ptr</span></tt> is used.  There are also two basic
131categories to to-python conversion, &quot;return value&quot; conversion
132(for Python-&gt;C++ calls) and &quot;argument&quot; conversion (for
133C++-&gt;Python calls and explicit <tt class="literal"><span class="pre">object()</span></tt> conversions).  The
134behavior of these two categories differs subtly in various ways
135whose details I forget at the moment.  You can probably find
136the answers in the above references, and certainly in the code.</p>
137<p>The &quot;default&quot; case is by-value (copying) conversion, which uses
138to_python_value as a to-python converter.</p>
139<blockquote>
140<p>Since there can sensibly be only one way to convert any type
141to python (disregarding the idea of scoped registries for the
142moment), it makes sense that to-python conversions can be
143handled by specializing a template.  If the type is one of
144the types handled by a built-in conversion
145(builtin_converters.hpp), the corresponding template
146specialization of to_python_value gets used.</p>
147<p>Otherwise, to_python_value uses the <tt class="literal"><span class="pre">m_to_python</span></tt>
148function in the registration for the C++ type.</p>
149</blockquote>
150<p>Other conversions, like by-reference conversions, are only
151available for wrapped classes, and are requested explicitly by
152using <tt class="literal"><span class="pre">ref(...)</span></tt>, <tt class="literal"><span class="pre">ptr(...)</span></tt>, or by specifying different
153CallPolicies for a call, which can cause a different to-python
154converter to be used.  These conversions are never registered
155anywhere, though they do need to use the registration to find
156the Python class corresponding to the C++ type being referred
157to.  They just build a new Python instance and stick the
158appropriate Holder instance in it.</p>
159</blockquote>
160<p>from python to C++:</p>
161<blockquote>
162<p>Once again I think there is a distinction between &quot;return value&quot;
163and &quot;argument&quot; conversions, and I forget exactly what that is.</p>
164<p>What happens depends on whether an lvalue conversion is needed
165(see <a class="reference" href="http://mail.python.org/pipermail/c++-sig/2002-May/001023.html">http://mail.python.org/pipermail/c++-sig/2002-May/001023.html</a>)
166All lvalue conversions are also registered in a type's rvalue
167conversion chain, since when an rvalue will do, an lvalue is
168certainly good enough.</p>
169<p>An lvalue conversion can be done in one step (just get me the
170pointer to the object - it can be <tt class="literal"><span class="pre">NULL</span></tt> if no conversion is
171possible) while an rvalue conversion requires two steps to
172support wrapped function overloading and multiple converters for
173a given C++ target type: first tell me if a conversion is
174possible, then construct the converted object as a second step.</p>
175</blockquote>
176</blockquote>
177</div>
178</div>
179<hr class="footer"/>
180<div class="footer">
181<a class="reference" href="internals.rst">View document source</a>.
182Generated on: 2003-09-12 14:51 UTC.
183Generated by <a class="reference" href="http://docutils.sourceforge.net/">Docutils</a> from <a class="reference" href="http://docutils.sourceforge.net/rst.html">reStructuredText</a> source.
184</div>
185</body>
186</html>
Note: See TracBrowser for help on using the repository browser.