2
0
mirror of https://github.com/boostorg/python.git synced 2026-01-19 16:32:16 +00:00
Files
python/doc/special.html
Dave Abrahams 2e5d8dbff7 This commit was generated by cvs2svn to compensate for changes in r711,
which included commits to RCS files with non-trunk default branches.


[SVN r8328]
2000-11-26 15:49:26 +00:00

889 lines
32 KiB
HTML
Raw Blame History

<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.0//EN"
"http://www.w3.org/TR/REC-html40/strict.dtd">
<title>
Special Method and Operator Support
</title>
<div>
<h1>
<img width="277" height="86" id="_x0000_i1025" align="center" src=
"../../../c++boost.gif" alt="c++boost.gif (8819 bytes)">Special Method and
Operator Support
</h1>
<h2>
Overview
</h2>
<p>
Py_cpp supports all of the standard <a href=
"http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/specialnames.html">
special method names</a> supported by real Python class instances <em>
except</em> <code>__complex__</code> (more on the reasons <a href=
"#reasons">below</a>). In addition, it can quickly and easily expose
suitable C++ functions and operators as Python operators. The following
categories of special method names are supported:
<ul>
<li><a href="#general">Basic Customization</a>
<li><a href="#numeric">Numeric Operators</a>
<li><a href="#sequence_and_mapping">Sequence and Mapping protocols</a>
<li><a href="#getter_setter">Attribute Getters and Setters</a>
</ul>
<h2><a name="general">Basic Customization</a></h2>
<p>
Python provides a number of special operators for basic customization of a
class. Only a brief description is provided below; more complete
documentation can be found <a
href="http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/customization.html">here</a>.
<dl>
<dt>
<b><tt class='method'>__init__</tt></b>(<i>self</i>)
<dd>
Initialize the class instance. For extension classes not subclassed in
Python, this is provided by the
<code>boost::python::constructor<...>()</code> construct and should <i>not</i> be explicitly <code>def</code>ed.
<dt>
<b><tt class='method'>__del__</tt></b>(<i>self</i>)
<dd>
Called when the extension instance is about to be destroyed.
<dt>
<b><tt class='method'>__repr__</tt></b>(<i>self</i>)
<dd>
Create a string representation from which the object can be
reconstructed.
<dt>
<b><tt class='method'>__str__</tt></b>(<i>self</i>)
<dd>
Create a string representation which is suitable for printing.
<dt>
<b><tt class='method'>__cmp__</tt></b>(<i>self, other</i>)
<dd>
Three-way compare function, used to implement comparison operators
(&lt; etc.) Should return a negative integer if <code> self < other
</code> , zero if <code> self == other </code> , a positive integer if
<code> self > other </code>.
<dt>
<b><tt class='method'>__hash__</tt></b>(<i>self</i>)
<dd>
Called for the key object for dictionary operations, and by the
built-in function hash(). Should return a 32-bit integer usable as a
hash value for dictionary operations (only allowed if __cmp__ is also
defined)
<dt>
<b><tt class='method'>__nonzero__</tt></b>(<i>self</i>)
<dd>
called if the object is used as a truth value (e.g. in an if
statement)
<dt>
<b><tt class='method'>__call__</tt></b> (<var>self</var><big>[</big><var>, args...</var><big>]</big>)
<dd>
Called when the instance is &ldquo;called&rdquo; as a function; if this method
is defined, <code><var>x</var>(arg1, arg2, ...)</code> is a shorthand for
<code><var>x</var>.__call__(arg1, arg2, ...)</code>.
</dl>
If we have a suitable C++ function that supports any of these features,
we can export it like any other function, using its Python special name.
For example, suppose that class <code>Foo</code> provides a string
conversion function:
<blockquote><pre>
std::string to_string(Foo const&amp; f)
{
std::ostringstream s;
s &lt;&lt; f;
return s.str();
}
</pre></blockquote>
This function would be wrapped like this:
<blockquote><pre>
boost::python::class_builder&lt;Foo&gt; foo_class(my_module, "Foo");
foo_class.def(&amp;to_string, "__str__");
</pre></blockquote>
Note that py_cpp also supports <em>automatic wrapping</em> of
<code>__str__</code> and <code>__cmp__</code>. This is explained in the <a
href="#numeric">next section</a> and the <a href="#numeric_table">Table of
Automatically Wrapped Methods</a>.
<h2><a name="numeric">Numeric Operators</a></h2>
<p>
Numeric operators can be exposed manually, by <code>def</code>ing C++
[member] functions that support the standard Python <a
href="http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/numeric-types.html">numeric
protocols</a>. This is the basic same technique used to expose
<code>to_string()</code> as <code>__str__()</code> above, and is <a
href="#numeric_manual">covered in detail below</a>. Py_cpp also supports
<i>automatic wrapping</i> of numeric operators whenever they have already
been defined in C++.
<h3><a name="numeric_auto">Exposing C++ Operators Automatically</a></h3>
<p>
Supose we wanted to expose a C++ class
<code>BigNum</code> which supports addition, so that we can write (in C++):
<blockquote><pre>
BigNum a, b, c;
...
c = a + b;
</pre></blockquote>
<p>
To enable the same functionality in Python, we first wrap the <code>
BigNum</code> class as usual:
<blockquote><pre>
boost::python::class_builder&lt;BigNum&gt; bignum_class(my_module, "BigNum");
bignum_class.def(boost::python::constructor&lt;&gt;());
...
</pre></blockquote>
Then we export the addition operator like this:
<blockquote><pre>
bignum_class.def(boost::python::operators&lt;boost::python::op_add&gt;());
</pre></blockquote>
Since BigNum also supports subtraction, multiplication, and division, we
want to export those also. This can be done in a single command by
&ldquo;or&rdquo;ing the operator identifiers together (a complete list of these
identifiers and the corresponding operators can be found in the <a href=
"#numeric_table">Table of Automatically Wrapped Methods</a>):
<blockquote><pre>
bignum_class.def(boost::python::operators&lt;(boost::python::op_sub | boost::python::op_mul | boost::python::op_div)&gt;());
</pre></blockquote>
[Note that the or-expression must be enclosed in parentheses.]
<p>This form of operator definition can be used to wrap unary and
homogeneous binary operators (a <i>homogeneous</i> operator has left and
right operands of the same type). Now suppose that our C++ library also
supports addition of BigNums and plain integers:
<blockquote><pre>
BigNum a, b;
int i;
...
a = b + i;
a = i + b;
</pre></blockquote>
To wrap these heterogeneous operators, we need to specify a different type for
one of the operands. This is done using the <code>right_operand</code>
and <code>left_operand</code> templates:
<blockquote><pre>
bignum_class.def(boost::python::operators&lt;boost::python::op_add&gt;(), boost::python::right_operand&lt;int&gt;());
bignum_class.def(boost::python::operators&lt;boost::python::op_add&gt;(), boost::python::left_operand&lt;int&gt;());
</pre></blockquote>
Py_cpp uses overloading to register several variants of the same
operation (more on this in the context of <a href="#coercion">
coercion</a>). Again, several operators can be exported at once:
<blockquote><pre>
bignum_class.def(boost::python::operators&lt;(boost::python::op_sub | boost::python::op_mul | boost::python::op_div)&gt;(),
boost::python::right_operand&lt;int&gt;());
bignum_class.def(boost::python::operators&lt;(boost::python::op_sub | boost::python::op_mul | boost::python::op_div)&gt;(),
boost::python::left_operand&lt;int&gt;());
</pre></blockquote>
The type of the operand not mentioned is taken from the class being wrapped. In
our example, the class object is <code>bignum_class</code>, and thus the
other operand's type is &ldquo;<code>BigNum const&amp;</code>&rdquo;. You can override
this default by explicitly specifying a type in the <code>
operators</code> template:
<blockquote><pre>
bignum_class.def(boost::python::operators&lt;boost::python::op_add, BigNum&gt;(), boost::python::right_operand&lt;int&gt;());
</pre></blockquote>
<p>
Note that automatic wrapping uses the <em>expression</em>
&ldquo;<code>left + right</code>&rdquo; and can be used uniformly
regardless of whether the C++ operators are supplied as free functions
<blockquote><pre>
BigNum operator+(BigNum, BigNum)
</pre></blockquote>
or as member
functions <blockquote><pre>
BigNum::operator+(BigNum).
</blockquote></pre>
<p>
For the Python built-in functions <code>pow()</code> and
<code>abs()</code>, there is no corresponding C++ operator. Instead,
automatic wrapping attempts to wrap C++ functions of the same name. This
only works if those functions are known in namespace
<code>python</code>. On some compilers (e.g. MSVC) it might be
necessary to add a using declaration prior to wrapping:
<blockquote><pre>
namespace boost { namespace python {
using my_namespace::pow;
using my_namespace::abs;
}
</pre></blockquote>
<h3><a name="numeric_manual">Wrapping Numeric Operators Manually</a></h3>
<p>
In some cases, automatic wrapping of operators may be impossible or
undesirable. Suppose, for example, that the modulo operation for BigNums
is defined by a set of functions <code>mod()</code> (for automatic
wrapping, we would need <code>operator%()</code>):
<blockquote><pre>
BigNum mod(BigNum const&amp; left, BigNum const&amp; right);
BigNum mod(BigNum const&amp; left, int right);
BigNum mod(int left, BigNum const&amp; right);
</pre></blockquote>
<p>
In order to create the Python operator "__mod__" from these functions, we
have to wrap them manually:
<blockquote><pre>
bignum_class.def((BigNum (*)(BigNum const&amp;, BigNum const&amp;))&amp;mod, "__mod__");
bignum_class.def((BigNum (*)(BigNum const&amp;, int))&amp;mod, "__mod__");
</pre></blockquote>
<p>
The third form (with <code>int</code> as left operand) cannot be wrapped
this way. We must first create a function <code>rmod()</code> with the
operands reversed:
<blockquote><pre>
BigNum rmod(BigNum const&amp; right, int left)
{
return mod(left, right);
}
</pre></blockquote>
This function must be wrapped under the name "__rmod__":
<blockquote><pre>
bignum_class.def(&amp;rmod, "__rmod__");
</pre></blockquote>
Many of the possible operator names can be found in the <a href=
"#numeric_table">Table of Automatically Wrapped Methods</a>. Special treatment is
necessary to export the <a href="#ternary_pow">ternary pow</a> operator.
<p>
Automatic and manual wrapping can be mixed arbitrarily. Note that you
cannot overload the same operator for a given extension class on both
&ldquo;<code>int</code>&rdquo; and &ldquo;<code>float</code>&rdquo;, because Python implicitly
converts these types into each other. Thus, the overloaded variant
found first (be it &ldquo;<code>int</code>&ldquo; or &ldquo;<code>float</code>&rdquo;) will be
used for either of the two types.
<h3><a name="coercion">Coercion</a></h3>
Plain Python can only execute operators with identical types on the left
and right hand side. If it encounters an expression where the types of
the left and right operand differ, it tries to coerce these type to a
common type before invoking the actual operator. Implementing good
coercion functions can be difficult if many type combinations must be
supported.
<p>
Py_cpp solves this problem the same way that C++ does: with <em><a
href="overloading.html">overloading</a></em>. This technique drastically
simplifies the code neccessary to support operators: you just register
operators for all desired type combinations, and py_cpp automatically
ensures that the correct function is called in each case; there is no
need for user-defined coercion functions. To enable operator
overloading, py_cpp provides a standard coercion which is <em>implicitly
registered</em> whenever automatic operator wrapping is used.
<p>
If you wrap all operator functions manually, but still want to use
operator overloading, you have to register the standard coercion
function explicitly:
<blockquote><pre>
// this is not necessary if automatic operator wrapping is used
bignum_class.def_standard_coerce();
</pre></blockquote>
If you encounter a situation where you absolutely need a customized
coercion, you can overload the "__coerce__" operator itself. The signature
of a coercion function should look like one of the following (the first is
the safest):
<blockquote><pre>
boost::python::tuple custom_coerce(boost::python::reference left, boost::python::reference right);
boost::python::tuple custom_coerce(PyObject* left, PyObject* right);
PyObject* custom_coerce(PyObject* left, PyObject* right);
</pre></blockquote>
The resulting <code>tuple</code> must contain two elements which
represent the values of <code>left</code> and <code>right</code>
converted to the same type. Such a function is wrapped as usual:
<blockquote><pre>
some_class.def(&amp;custom_coerce, "__coerce__");
</pre></blockquote>
Note that the later use of automatic operator wrapping on a
<code>class_builder</code> or a call to
&ldquo;<code>some_class.def_standard_coerce()</code>&rdquo; will cause any
custom coercion function to be replaced by the standard one.
<h3><a name="ternary_pow">The Ternary <code>pow()</code> Operator</a></h3>
<p>
In addition to the usual binary <code>pow(x, y)</code> operator (meaning
<i>x<sup>y</sup></i>), Python also provides a ternary variant that implements
<i>x<sup>y</sup> <b>mod</b> z</i>, presumably using a more efficient algorithm than
concatenation of power and modulo operators. Automatic operator wrapping
can only be used with the binary variant. Ternary <code>pow()</code> must
always be wrapped manually. For a homgeneous ternary <code>pow()</code>,
this is done as usual:
<blockquote><pre>
BigNum power(BigNum const&amp; first, BigNum const&amp; second, BigNum const&amp; module);
typedef BigNum (ternary_function1)(const BigNum&amp;, const BigNum&amp;, const BigNum&amp;);
...
bignum_class.def((ternary_function1)&amp;power, "__pow__");
</pre></blockquote>
If you want to support this function with non-uniform argument
types, wrapping is a little more involved. Suppose you have to wrap:
<blockquote><pre>
BigNum power(BigNum const&amp; first, int second, int modulus);
BigNum power(int first, BigNum const&amp; second, int modulus);
BigNum power(int first, int second, BigNum const&amp; modulus);
</pre></blockquote>
The first variant can be wrapped as usual:
<blockquote><pre>
typedef BigNum (ternary_function2)(const BigNum&amp;, int, int);
bignum_class.def((ternary_function2)&amp;power, "__pow__");
</pre></blockquote>
In the second variant, however, <code>BigNum</code> appears only as second
argument, and in the last one it is the third argument. These functions
must be presented to py_cpp such that that the <code>BigNum</code>
argument appears in first position:
<blockquote><pre>
BigNum rpower(BigNum const&amp; second, int first, int modulus)
{
return power(first, second, third);
}
BigNum rrpower(BigNum const&amp; third, int first, int second)
{
return power(first, second, third);
}
</pre></blockquote>
<p>These functions must be wrapped under the names "__rpow__" and "__rrpow__"
respectively:
<blockquote><pre>
bignum_class.def((ternary_function2)&amp;rpower, "__rpow__");
bignum_class.def((ternary_function2)&amp;rrpower, "__rrpow__");
</pre></blockquote>
Note that "__rrpow__" is an extension not present in plain Python.
<h2><a name="numeric_table">Table of Automatically Wrapped Methods</a></h2>
<p>
Py_cpp can automatically wrap the following <a href=
"http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/specialnames.html">
special methods</a>:
<p>
<table summary="special numeric methods" cellpadding="5" border="1"
width="100%">
<tr>
<td align="center">
<b>Python Operator Name</b>
<td align="center">
<b>Python Expression</b>
<td align="center">
<b>C++ Operator Id</b>
<td align="center">
<b>C++ Expression Used For Automatic Wrapping</b><br>
with <code>cpp_left = from_python(left,
type&lt;Left&gt;())</code>,<br>
<code>cpp_right = from_python(right,
type&lt;Right&gt;())</code>,<br>
and <code>cpp_oper = from_python(oper, type&lt;Oper&gt;())</code>
<tr>
<td>
<code>__add__, __radd__</code>
<td>
<code>left + right</code>
<td>
<code>op_add</code>
<td>
<code>cpp_left + cpp_right</code>
<tr>
<td>
<code>__sub__, __rsub__</code>
<td>
<code>left - right</code>
<td>
<code>op_sub</code>
<td>
<code>cpp_left - cpp_right</code>
<tr>
<td>
<code>__mul__, __rmul__</code>
<td>
<code>left * right</code>
<td>
<code>op_mul</code>
<td>
<code>cpp_left * cpp_right</code>
<tr>
<td>
<code>__div__, __rdiv__</code>
<td>
<code>left / right</code>
<td>
<code>op_div</code>
<td>
<code>cpp_left / cpp_right</code>
<tr>
<td>
<code>__mod__, __rmod__</code>
<td>
<code>left % right</code>
<td>
<code>op_mod</code>
<td>
<code>cpp_left % cpp_right</code>
<tr>
<td>
<code>__divmod__, __rdivmod__</code>
<td>
<code>(quotient, remainder)<br>
= divmod(left, right)</code>
<td>
<code>op_divmod</code>
<td>
<code>cpp_left / cpp_right</code>
<br><code>cpp_left % cpp_right</code>
<tr>
<td>
<code>__pow__, __rpow__</code>
<td>
<code>pow(left, right)</code><br>
(binary power)
<td>
<code>op_pow</code>
<td>
<code>pow(cpp_left, cpp_right)</code>
<tr>
<td>
<code>__rrpow__</code>
<td>
<code>pow(left, right, modulo)</code><br>
(ternary power modulo)
<td colspan="2">
no automatic wrapping, <a href="#ternary_pow">special treatment</a>
required
<tr>
<td>
<code>__lshift__, __rlshift__</code>
<td>
<code>left &lt;&lt; right</code>
<td>
<code>op_lshift</code>
<td>
<code>cpp_left &lt;&lt; cpp_right</code>
<tr>
<td>
<code>__rshift__, __rrshift__</code>
<td>
<code>left &gt;&gt; right</code>
<td>
<code>op_rshift</code>
<td>
<code>cpp_left &gt;&gt; cpp_right</code>
<tr>
<td>
<code>__and__, __rand__</code>
<td>
<code>left &amp; right</code>
<td>
<code>op_and</code>
<td>
<code>cpp_left &amp; cpp_right</code>
<tr>
<td>
<code>__xor__, __rxor__</code>
<td>
<code>left ^ right</code>
<td>
<code>op_xor</code>
<td>
<code>cpp_left ^ cpp_right</code>
<tr>
<td>
<code>__or__, __ror__</code>
<td>
<code>left | right</code>
<td>
<code>op_or</code>
<td>
<code>cpp_left | cpp_right</code>
<tr>
<td>
<code>__cmp__, __rcmp__</code>
<td>
<code>cmp(left, right)</code><br>
<code>left &lt; right</code><br>
<code>left &lt;= right</code><br>
<code>left &gt; right</code><br>
<code>left &gt;= right</code><br>
<code>left == right</code><br>
<code>left != right</code>
<td>
<code>op_cmp</code>
<td>
<code>cpp_left &lt; cpp_right<68></code>
<br><code>cpp_right &lt; cpp_left</code>
<tr>
<td>
<code>__neg__</code>
<td>
<code>-oper<65></code> (unary negation)
<td>
<code>op_neg</code>
<td>
<code>-cpp_oper</code>
<tr>
<td>
<code>__pos__</code>
<td>
<code>+oper<65></code> (identity)
<td>
<code>op_pos</code>
<td>
<code>+cpp_oper</code>
<tr>
<td>
<code>__abs__</code>
<td>
<code>abs(oper)<29></code> (absolute value)
<td>
<code>op_abs</code>
<td>
<code>abs(cpp_oper)</code>
<tr>
<td>
<code>__invert__</code>
<td>
<code>~oper<65></code> (bitwise inversion)
<td>
<code>op_invert</code>
<td>
<code>~cpp_oper</code>
<tr>
<td>
<code>__int__</code>
<td>
<code>int(oper)<29></code> (integer conversion)
<td>
<code>op_int</code>
<td>
<code>long(cpp_oper)</code>
<tr>
<td>
<code>__long__</code>
<td>
<code>long(oper)<29></code><br>
(infinite precision integer conversion)
<td>
<code>op_long</code>
<td>
<code>PyLong_FromLong(cpp_oper)</code>
<tr>
<td>
<code>__float__</code>
<td>
<code>float(oper)<29></code> (float conversion)
<td>
<code>op_float</code>
<td>
<code>double(cpp_oper)</code>
<tr>
<td>
<code>__str__</code>
<td>
<code>str(oper)<29></code> (string conversion)
<td>
<code>op_str</code>
<td>
<code>std::ostringstream s; s &lt;&lt; oper;</code>
<tr>
<td>
<code>__coerce__</code>
<td>
<code>coerce(left, right)</code>
<td colspan="2">
usually defined automatically, otherwise <a href="#coercion">
special treatment</a> required
</table>
<h2><a name="sequence_and_mapping">Sequence and Mapping Operators</a></h2>
<p>
Sequence and mapping operators let wrapped objects behave in accordance
to Python's iteration and access protocols. These protocols differ
considerably from the ones found in C++. For example, Python's typical
iteration idiom looks like
<blockquote><pre>
for i in S:
</blockquote></pre>
while in C++ one writes
<blockquote><pre>
for (iterator i = S.begin(), end = S.end(); i != end)
</blockquote></pre>
<p>One could try to wrap C++ iterators in order to carry the C++ idiom into
Python. However, this does not work very well because
<ol>
<li>It leads to
non-uniform Python code (wrapped sequences support a usage different from
Python built-in sequences) and
<li>Iterators (e.g. <code>std::vector::iterator</code>) are often implemented as plain C++
pointers which are <a href="pointers.html#problem">problematic</a> for any automatic
wrapping system.
</ol>
<p>
It is a better idea to support the standard <a
href="http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/sequence-types.html">Python
sequence and mapping protocols</a> for your wrapped containers. These
operators have to be wrapped manually because there are no corresponding
C++ operators that could be used for automatic wrapping. The Python
documentation lists the relevant <a href=
"http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/sequence-types.html">
container operators</a>. In particular, expose __getitem__, __setitem__
and remember to raise the appropriate Python exceptions
(<code>PyExc_IndexError</code> for sequences,
<code>PyExc_KeyError</code> for mappings) when the requested item is not
present.
<p>
In the following example, we expose <code>std::map&lt;std::size_t,std::string&gt;</code>:
<blockquote>
<pre>
typedef std::map&lt;std::size_t, std::string&gt; StringMap;
// A helper function for dealing with errors. Throw a Python exception
// if p == m.end().
void throw_key_error_if_end(
const StringMap&amp; m,
StringMap::const_iterator p,
std::size_t key)
{
if (p == m.end())
{
PyErr_SetObject(PyExc_KeyError, boost::python::converters::to_python(key));
throw boost::python::error_already_set();
}
}
// Define some simple wrapper functions which match the Python protocol
// for __getitem__, __setitem__, and __delitem__. Just as in Python, a
// free function with a &ldquo;self&rdquo; first parameter makes a fine class method.
const std::string&amp; get_item(const StringMap&amp; self, std::size_t key)
{
const StringMap::const_iterator p = self.find(key);
throw_key_error_if_end(self, p, key);
return p-&gt;second;
}
// Sets the item corresponding to key in the map.
void StringMapPythonClass::set_item(StringMap&amp; self, std::size_t key, const std::string&amp; value)
{
self[key] = value;
}
// Deletes the item corresponding to key from the map.
void StringMapPythonClass::del_item(StringMap&amp; self, std::size_t key)
{
const StringMap::iterator p = self.find(key);
throw_key_error_if_end(self, p, key);
self.erase(p);
}
class_builder&lt;StringMap&gt; string_map(my_module, "StringMap");
string_map.def(boost::python::constructor&lt;&gt;());
string_map.def(&amp;StringMap::size, "__len__");
string_map.def(get_item, "__getitem__");
string_map.def(set_item, "__setitem__");
string_map.def(del_item, "__delitem__");
</pre>
</blockquote>
<p>
Then in Python:
<blockquote>
<pre>
&gt;&gt;&gt; m = StringMap()
&gt;&gt;&gt; m[1]
Traceback (innermost last):
File "&lt;stdin&gt;", line 1, in ?
KeyError: 1
&gt;&gt;&gt; m[1] = 'hello'
&gt;&gt;&gt; m[1]
'hello'
&gt;&gt;&gt; del m[1]
&gt;&gt;&gt; m[1] # prove that it's gone
Traceback (innermost last):
File "&lt;stdin&gt;", line 1, in ?
KeyError: 1
&gt;&gt;&gt; del m[2]
Traceback (innermost last):
File "&lt;stdin&gt;", line 1, in ?
KeyError: 2
&gt;&gt;&gt; len(m)
0
&gt;&gt;&gt; m[0] = 'zero'
&gt;&gt;&gt; m[1] = 'one'
&gt;&gt;&gt; m[2] = 'two'
&gt;&gt;&gt; m[3] = 'three'
&gt;&gt;&gt; len(m)
4
</pre>
</blockquote>
<h2><a name="getter_setter">Customized Attribute Access</a></h2>
<p>
Just like built-in Python classes, py_cpp extension classes support <a
href="http://www.pythonlabs.com/pub/www.python.org/doc/current/ref/attribute-access.html">special
the usual attribute access methods</a> <code>__getattr__</code>,
<code>__setattr__</code>, and <code>__delattr__</code>.
Because writing these functions can
be tedious in the common case where the attributes being accessed are
known statically, py_cpp checks the special names
<ul>
<li>
<code>__getattr__<em>&lt;name&gt;</em>__</code>
<li>
<code>__setattr__<em>&lt;name&gt;</em>__</code>
<li>
<code>__delattr__<em>&lt;name&gt;</em>__</code>
</ul>
to provide functional access to the attribute <em>&lt;name&gt;</em>. This
facility can be used from C++ or entirely from Python. For example, the
following shows how we can implement a &ldquo;computed attribute&rdquo; in Python:
<blockquote>
<pre>
&gt;&gt;&gt; class Range(AnyPy_cppExtensionClass):
... def __init__(self, start, end):
... self.start = start
... self.end = end
... def __getattr__length__(self):
... return self.end - self.start
...
&gt;&gt;&gt; x = Range(3, 9)
&gt;&gt;&gt; x.length
6
</pre>
</blockquote>
<h4>
Direct Access to Data Members
</h4>
<p>
Py_cpp uses the special <code>
__xxxattr__<em>&lt;name&gt;</em>__</code> functionality described above
to allow direct access to data members through the following special
functions on <code>class_builder&lt;&gt;</code> and <code>
extension_class&lt;&gt;</code>:
<ul>
<li>
<code>def_getter(<em>pointer-to-member</em>, <em>name</em>)</code> //
read access to the member via attribute <em>name</em>
<li>
<code>def_setter(<em>pointer-to-member</em>, <em>name</em>)</code> //
write access to the member via attribute <em>name</em>
<li>
<code>def_readonly(<em>pointer-to-member</em>, <em>name</em>)</code>
// read-only access to the member via attribute <em>name</em>
<li>
<code>def_read_write(<em>pointer-to-member</em>, <em>
name</em>)</code> // read/write access to the member via attribute
<em>name</em>
</ul>
<p>
Note that the first two functions, used alone, may produce surprising
behavior. For example, when <code>def_getter()</code> is used, the
default functionality for <code>setattr()</code> and <code>
delattr()</code> remains in effect, operating on items in the extension
instance's name-space (i.e., its <code>__dict__</code>). For that
reason, you'll usually want to stick with <code>def_readonly</code> and
<code>def_read_write</code>.
<p>
For example, to expose a <code>std::pair&lt;int,long&gt;</code> we
might write:
<blockquote>
<pre>
typedef std::pair&lt;int,long&gt; Pil;
int first(const Pil&amp; x) { return x.first; }
long second(const Pil&amp; x) { return x.second; }
...
my_module.def(first, "first");
my_module.def(second, "second");
class_builder&lt;Pil&gt; pair_int_long(my_module, "Pair");
pair_int_long.def(boost::python::constructor&lt;&gt;());
pair_int_long.def(boost::python::constructor&lt;int,long&gt;());
pair_int_long.def_read_write(&amp;Pil::first, "first");
pair_int_long.def_read_write(&amp;Pil::second, "second");
</pre>
</blockquote>
<p>
Now your Python class has attributes <code>first</code> and <code>
second</code> which, when accessed, actually modify or reflect the
values of corresponding data members of the underlying C++ object. Now
in Python:
<blockquote>
<pre>
&gt;&gt;&gt; x = Pair(3,5)
&gt;&gt;&gt; x.first
3
&gt;&gt;&gt; x.second
5
&gt;&gt;&gt; x.second = 8
&gt;&gt;&gt; x.second
8
&gt;&gt;&gt; second(x) # Prove that we're not just changing the instance __dict__
8
</pre>
</blockquote>
<h2>
<a name="reasons">And what about <code>__complex__</code>?</a>
</h2>
<p>
That, dear reader, is one problem we don't know how to solve. The
Python source contains the following fragment, indicating the
special-case code really is hardwired:
<blockquote>
<pre>
/* XXX Hack to support classes with __complex__ method */
if (PyInstance_Check(r)) { ...
</pre>
</blockquote>
<p>
Previous: <a href="inheritance.html">Inheritance</a> Next: <a href=
"under-the-hood.html">A Peek Under the Hood</a> Up: <a href=
"index.html">Top</a>
<p>
&copy; Copyright David Abrahams and Ullrich K&ouml;the 2000.
Permission to copy, use, modify, sell and distribute this document is
granted provided this copyright notice appears in all copies. This
document is provided &ldquo;as is&rdquo; without express or implied
warranty, and with no claim as to its suitability for any purpose.
<p>
Updated: Nov 26, 2000
</div>