2
0
mirror of https://github.com/boostorg/python.git synced 2026-01-20 04:42:28 +00:00
Files
python/doc/cross_module.html
2001-04-17 19:49:42 +00:00

337 lines
12 KiB
HTML

<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.0//EN"
"http://www.w3.org/TR/REC-html40/strict.dtd">
<title>Cross-extension-module dependencies</title>
<div>
<img src="../../../c++boost.gif"
alt="c++boost.gif (8819 bytes)"
align="center"
width="277" height="86">
<hr>
<h1>Cross-extension-module dependencies</h1>
It is good programming practice to organize large projects as modules
that interact with each other via well defined interfaces. With
Boost.Python it is possible to reflect this organization at the C++
level at the Python level. This is, each logical C++ module can be
organized as a separate Python extension module.
<p>
At first sight this might seem natural and straightforward. However, it
is a fairly complex problem to establish cross-extension-module
dependencies while maintaining the same ease of use Boost.Python
provides for classes that are wrapped in the same extension module. To
a large extent this complexity can be hidden from the author of a
Boost.Python extension module, but not entirely.
<hr>
<h2>The recipe</h2>
Suppose there is an extension module that exposes certain instances of
the C++ <tt>std::vector</tt> template library such that it can be used
from Python in the following manner:
<pre>
import std_vector
v = std_vector.double([1, 2, 3, 4])
v.push_back(5)
v.size()
</pre>
Suppose the <tt>std_vector</tt> module is done well and reflects all
C++ functions that are useful at the Python level, for all C++ built-in
data types (<tt>std_vector.int</tt>, <tt>std_vector.long</tt>, etc.).
<p>
Suppose further that there is statistic module with a C++ class that
has constructors or member functions that use or return a
<tt>std::vector</tt>. For example:
<pre>
class xy {
public:
xy(const std::vector&lt;double&gt;&amp; x, const std::vector&lt;double&gt;&amp; y) : m_x(x), m_y(y) {}
const std::vector&lt;double&gt;&amp; x() const { return m_x; }
const std::vector&lt;double&gt;&amp; y() const { return m_y; }
double correlation();
private:
std::vector&lt;double&gt; m_x;
std::vector&lt;double&gt; m_y;
}
</pre>
What is more natural than reusing the <tt>std_vector</tt> extension
module to expose these constructors or functions to Python?
<p>
Unfortunately, what seems natural needs a little work in both the
<tt>std_vector</tt> and the <tt>statistics</tt> module.
<p>
In the <tt>std_vector</tt> extension module,
<tt>std::vector&lt;double&gt;</tt> is exposed to Python in the usual
way with the <tt>class_builder&lt;&gt;</tt> template. To also enable the
automatic conversion of <tt>std::vector&lt;double&gt;</tt> function
arguments or return values in other Boost.Python C++ modules, the
converters that convert a <tt>std::vector&lt;double&gt;</tt> C++ object
to a Python object and vice versa (i.e. the <tt>to_python()</tt> and
<tt>from_python()</tt> template functions) have to be exported. For
example:
<pre>
#include &lt;boost/python/cross_module.hpp&gt;
//...
class_builder&lt;std::vector&lt;double&gt; &gt; v_double(std_vector_module, &quot;double&quot;);
export_converters(v_double);
</pre>
In the extension module that wraps <tt>class xy</tt> we can now import
these converters with the <tt>import_converters&lt;&gt;</tt> template.
For example:
<pre>
#include &lt;boost/python/cross_module.hpp&gt;
//...
import_converters&lt;std::vector&lt;double&gt; &gt; v_double_converters(&quot;std_vector&quot;, &quot;double&quot;);
</pre>
That is all. All the attributes that are defined for
<tt>std_vector.double</tt> in the <tt>std_vector</tt> Boost.Python
module will be available for the returned objects of <tt>xy.x()</tt>
and <tt>xy.y()</tt>. Similarly, the constructor for <tt>xy</tt> will
accept objects that were created by the <tt>std_vector</tt>module.
<hr>
<h2>Placement of <tt>import_converters&lt;&gt;</tt> template instantiations</h2>
<tt>import_converts&lt;&gt;</tt> can be viewed as a drop-in replacement
for <tt>class_wrapper&lt;&gt;</tt>, and the recommendations for the
placement of <tt>class_wrapper&lt;&gt;</tt> template instantiations
also apply to to <tt>import_converts&lt;&gt;</tt>. In particular, it is
important that an instantiation of <tt>class_wrapper&lt;&gt;</tt> is
visible to any code which wraps a C++ function with a <tt>T</tt>,
<tt>T*</tt>, const <tt>T&amp;</tt>, etc. parameter or return value.
Therefore you may want to group all <tt>class_wrapper&lt;&gt;</tt> and
<tt>import_converts&lt;&gt;</tt> instantiations at the top of your
module's init function, then <tt>def()</tt> the member functions later
to avoid problems with inter-class dependencies.
<hr>
<h2>Non-copyable types</h2>
<tt>export_converters()</tt> instantiates C++ template functions that
invoke the copy constructor of the wrapped type. For a type that is
non-copyable this will result in compile-time error messages. In such a
case, <tt>export_converters_noncopyable()</tt> can be used to export
the converters that do not involve the copy constructor of the wrapped
type. For example:
<pre>
class_builder&lt;store&gt; py_store(your_module, &quot;store&quot;);
export_converters_noncopyable(py_store);
</pre>
The corresponding <tt>import_converters&lt;&gt;</tt> statement does not
need any special attention:
<pre>
import_converters&lt;store&gt; py_store(&quot;noncopyable_export&quot;, &quot;store&quot;);
</pre>
<hr>
<h2>Python module search path</h2>
The <tt>std_vector</tt> and <tt>statistics</tt> modules can now be used
in the following way:
<pre>
import std_vector
import statistics
x = std_vector.double([1, 2, 3, 4])
y = std_vector.double([2, 4, 6, 8])
xy = statistics.xy(x, y)
xy.correlation()
</pre>
In this example it is clear that Python has to be able to find both the
<tt>std_vector</tt> and the <tt>statistics</tt> extension module. In
other words, both extension modules need to be in the Python module
search path (<tt>sys.path</tt>).
<p>
The situation is not always this obvious. Suppose the
<tt>statistics</tt> module has a <tt>random()</tt> function that
returns a vector of random numbers with a given length:
<pre>
import statistics
x = statistics.random(5)
y = statistics.random(5)
xy = statistics.xy(x, y)
xy.correlation()
</pre>
A naive user will not easily anticipate that the <tt>std_vector</tt>
module is used to pass the <tt>x</tt> and <tt>y</tt> vectors around. If
the <tt>std_vector</tt> module is in the Python module search path,
this form of ignorance is of no harm. On the contrary, we are glad
that we do not have to bother the user with details like this.
<p>
If the <tt>std_vector</tt> module is not in the Python module search
path, a Python exception will be raised:
<pre>
Traceback (innermost last):
File &quot;foo.py&quot;, line 2, in ?
x = statistics.random(5)
ImportError: No module named std_vector
</pre>
As is the case with any system of a non-trivial complexity, it is
important that the setup is consistent and complete.
<hr>
<h2>Two-way module dependencies</h2>
Boost.Python supports two-way module dependencies. This is best
illustrated by a simple example.
<p>
Suppose there is a module <tt>ivect</tt> that implements vectors of
integers, and a similar module <tt>dvect</tt> that implements vectors
of doubles. We want to be able do convert an integer vector to a double
vector and vice versa. For example:
<pre>
import ivect
iv = ivect.ivect((1,2,3,4,5))
dv = iv.as_dvect()
</pre>
The last expression will implicitly import the <tt>dvect</tt> module in
order to enable the conversion of the C++ representation of
<tt>dvect</tt> to a Python object. The analogous is possible for a
<tt>dvect</tt>:
<pre>
import dvect
dv = dvect.dvect((1,2,3,4,5))
iv = dv.as_ivect()
</pre>
Now the <tt>ivect</tt> module is imported implicitly.
<p>
Note that the two-way dependencies are possible because the
dependencies are resolved only when needed. This is, the initialization
of the <tt>ivect</tt> module does not rely on the <tt>dvect</tt>
module, and vice versa. Only if <tt>as_dvect()</tt> or
<tt>as_ivect()</tt> is actually invoked will the corresponding module
be implicitly imported. This also means that, for example, the
<tt>dvect</tt> module does not have to be available at all if
<tt>as_dvect()</tt> is never used.
<hr>
<h2>Clarification of compile-time and link-time dependencies</h2>
Boost.Python's support for resolving cross-module dependencies at
runtime does not imply that compile-time dependencies are eliminated.
For example, the statistics extension module in the example above will
need to <tt>#include &lt;vector&gt;</tt>. This is immediately obvious
from the definition of <tt>class xy</tt>.
<p>
If a library is wrapped that consists of both header files and compiled
components (e.g. <tt>libdvect.a</tt>, <tt>dvect.lib</tt>, etc.), both
the Boost.Python extension module with the
<tt>export_converters()</tt> statement and the module with the
<tt>import_converters&lt;&gt;</tt> statement need to be linked against
the object library. Ideally one would build a shared library (e.g.
<tt>libdvect.so</tt>, <tt>dvect.dll</tt>, etc.). However, this
introduces the issue of having to configure the search path for the
dynamic loading correctly. For small libraries it is therefore often
more convenient to ignore the fact that the object files are loaded
into memory more than once.
<hr>
<h2>Summary of motivation for cross-module support</h2>
The main purpose of Boost.Python's cross-module support is to allow for
a modular system layout. With this support it is straightforward to
reflect C++ code organization at the Python level. Without the
cross-module support, a multi-purpose module like <tt>std_vector</tt>
would be impractical because the entire wrapper code would somehow have
to be duplicated in all extension modules that use it, making them
harder to maintain and harder to build.
<p>
Another motivation for the cross-module support is that two extension
modules that wrap the same class cannot both be imported into Python.
For example, if there are two modules <tt>A</tt> and <tt>B</tt> that
both wrap a given <tt>class X</tt>, this will work:
<pre>
import A
x = A.X()
</pre>
This will also work:
<pre>
import B
x = B.X()
</pre>
However, this will fail:
<pre>
import A
import B
python: /net/cci/rwgk/boost/boost/python/detail/extension_class.hpp:866:
static void boost::python::detail::class_registry&lt;X&gt;::register_class(boost::python::detail::extension_class_base *):
Assertion `static_class_object == 0' failed.
Abort
</pre>
A good solution is to wrap <tt>class X</tt> only once. Depending on the
situation, this could be done by module <tt>A</tt> or <tt>B</tt>, or an
additional small extension module that only wraps and exports
<tt>class X</tt>.
<p>
Finally, there can be important psychological or political reasons for
using the cross-module support. If a group of classes is lumped
together with many others in a huge module, the authors will have
difficulties in being identified with their work. The situation is
much more transparent if the work is represented by a module with a
recognizable name. This is not just a question of strong egos, but also
of getting credit and funding.
<hr>
<h2>Why not use <tt>export_converters()</tt> universally?</h2>
There is some overhead associated with the Boost.Python cross-module
support. Depending on the platform, the size of the code generated by
<tt>export_converters()</tt> is roughly 10%-20% of that generated
by <tt>class_builder&lt;&gt;</tt>. For a large extension module with
many wrapped classes, this could mean a significant difference.
Therefore the general recommendation is to use
<tt>export_converters()</tt> only for classes that are likely to
be used as function arguments or return values in other modules.
<hr>
&copy; Copyright Ralf W. Grosse-Kunstleve 2001. Permission to copy,
use, modify, sell and distribute this document is granted provided this
copyright notice appears in all copies. This document is provided "as
is" without express or implied warranty, and with no claim as to its
suitability for any purpose.
<p>
Updated: April 2001
</div>