6 ******************************
7 Extending Python with C or C++
8 ******************************
10 It is quite easy to add new built-in modules to Python, if you know how to
11 program in C. Such :dfn:`extension modules` can do two things that can't be
12 done directly in Python: they can implement new built-in object types, and they
13 can call C library functions and system calls.
15 To support extensions, the Python API (Application Programmers Interface)
16 defines a set of functions, macros and variables that provide access to most
17 aspects of the Python run-time system. The Python API is incorporated in a C
18 source file by including the header ``"Python.h"``.
20 The compilation of an extension module depends on its intended use as well as on
21 your system setup; details are given in later chapters.
23 Do note that if your use case is calling C library functions or system calls,
24 you should consider using the :mod:`ctypes` module rather than writing custom
25 C code. Not only does :mod:`ctypes` let you write Python code to interface
26 with C code, but it is more portable between implementations of Python than
27 writing and compiling an extension module which typically ties you to CPython.
31 .. _extending-simpleexample:
36 Let's create an extension module called ``spam`` (the favorite food of Monty
37 Python fans...) and let's say we want to create a Python interface to the C
38 library function :cfunc:`system`. [#]_ This function takes a null-terminated
39 character string as argument and returns an integer. We want this function to
40 be callable from Python as follows::
43 >>> status = spam.system("ls -l")
45 Begin by creating a file :file:`spammodule.c`. (Historically, if a module is
46 called ``spam``, the C file containing its implementation is called
47 :file:`spammodule.c`; if the module name is very long, like ``spammify``, the
48 module name can be just :file:`spammify.c`.)
50 The first line of our file can be::
54 which pulls in the Python API (you can add a comment describing the purpose of
55 the module and a copyright notice if you like).
59 Since Python may define some pre-processor definitions which affect the standard
60 headers on some systems, you *must* include :file:`Python.h` before any standard
63 All user-visible symbols defined by :file:`Python.h` have a prefix of ``Py`` or
64 ``PY``, except those defined in standard header files. For convenience, and
65 since they are used extensively by the Python interpreter, ``"Python.h"``
66 includes a few standard header files: ``<stdio.h>``, ``<string.h>``,
67 ``<errno.h>``, and ``<stdlib.h>``. If the latter header file does not exist on
68 your system, it declares the functions :cfunc:`malloc`, :cfunc:`free` and
69 :cfunc:`realloc` directly.
71 The next thing we add to our module file is the C function that will be called
72 when the Python expression ``spam.system(string)`` is evaluated (we'll see
73 shortly how it ends up being called)::
76 spam_system(PyObject *self, PyObject *args)
81 if (!PyArg_ParseTuple(args, "s", &command))
83 sts = system(command);
84 return Py_BuildValue("i", sts);
87 There is a straightforward translation from the argument list in Python (for
88 example, the single expression ``"ls -l"``) to the arguments passed to the C
89 function. The C function always has two arguments, conventionally named *self*
92 The *self* argument is only used when the C function implements a built-in
93 method, not a function. In the example, *self* will always be a *NULL* pointer,
94 since we are defining a function, not a method. (This is done so that the
95 interpreter doesn't have to understand two different types of C functions.)
97 The *args* argument will be a pointer to a Python tuple object containing the
98 arguments. Each item of the tuple corresponds to an argument in the call's
99 argument list. The arguments are Python objects --- in order to do anything
100 with them in our C function we have to convert them to C values. The function
101 :cfunc:`PyArg_ParseTuple` in the Python API checks the argument types and
102 converts them to C values. It uses a template string to determine the required
103 types of the arguments as well as the types of the C variables into which to
104 store the converted values. More about this later.
106 :cfunc:`PyArg_ParseTuple` returns true (nonzero) if all arguments have the right
107 type and its components have been stored in the variables whose addresses are
108 passed. It returns false (zero) if an invalid argument list was passed. In the
109 latter case it also raises an appropriate exception so the calling function can
110 return *NULL* immediately (as we saw in the example).
113 .. _extending-errors:
115 Intermezzo: Errors and Exceptions
116 =================================
118 An important convention throughout the Python interpreter is the following: when
119 a function fails, it should set an exception condition and return an error value
120 (usually a *NULL* pointer). Exceptions are stored in a static global variable
121 inside the interpreter; if this variable is *NULL* no exception has occurred. A
122 second global variable stores the "associated value" of the exception (the
123 second argument to :keyword:`raise`). A third variable contains the stack
124 traceback in case the error originated in Python code. These three variables
125 are the C equivalents of the Python variables ``sys.exc_type``,
126 ``sys.exc_value`` and ``sys.exc_traceback`` (see the section on module
127 :mod:`sys` in the Python Library Reference). It is important to know about them
128 to understand how errors are passed around.
130 The Python API defines a number of functions to set various types of exceptions.
132 The most common one is :cfunc:`PyErr_SetString`. Its arguments are an exception
133 object and a C string. The exception object is usually a predefined object like
134 :cdata:`PyExc_ZeroDivisionError`. The C string indicates the cause of the error
135 and is converted to a Python string object and stored as the "associated value"
138 Another useful function is :cfunc:`PyErr_SetFromErrno`, which only takes an
139 exception argument and constructs the associated value by inspection of the
140 global variable :cdata:`errno`. The most general function is
141 :cfunc:`PyErr_SetObject`, which takes two object arguments, the exception and
142 its associated value. You don't need to :cfunc:`Py_INCREF` the objects passed
143 to any of these functions.
145 You can test non-destructively whether an exception has been set with
146 :cfunc:`PyErr_Occurred`. This returns the current exception object, or *NULL*
147 if no exception has occurred. You normally don't need to call
148 :cfunc:`PyErr_Occurred` to see whether an error occurred in a function call,
149 since you should be able to tell from the return value.
151 When a function *f* that calls another function *g* detects that the latter
152 fails, *f* should itself return an error value (usually *NULL* or ``-1``). It
153 should *not* call one of the :cfunc:`PyErr_\*` functions --- one has already
154 been called by *g*. *f*'s caller is then supposed to also return an error
155 indication to *its* caller, again *without* calling :cfunc:`PyErr_\*`, and so on
156 --- the most detailed cause of the error was already reported by the function
157 that first detected it. Once the error reaches the Python interpreter's main
158 loop, this aborts the currently executing Python code and tries to find an
159 exception handler specified by the Python programmer.
161 (There are situations where a module can actually give a more detailed error
162 message by calling another :cfunc:`PyErr_\*` function, and in such cases it is
163 fine to do so. As a general rule, however, this is not necessary, and can cause
164 information about the cause of the error to be lost: most operations can fail
165 for a variety of reasons.)
167 To ignore an exception set by a function call that failed, the exception
168 condition must be cleared explicitly by calling :cfunc:`PyErr_Clear`. The only
169 time C code should call :cfunc:`PyErr_Clear` is if it doesn't want to pass the
170 error on to the interpreter but wants to handle it completely by itself
171 (possibly by trying something else, or pretending nothing went wrong).
173 Every failing :cfunc:`malloc` call must be turned into an exception --- the
174 direct caller of :cfunc:`malloc` (or :cfunc:`realloc`) must call
175 :cfunc:`PyErr_NoMemory` and return a failure indicator itself. All the
176 object-creating functions (for example, :cfunc:`PyInt_FromLong`) already do
177 this, so this note is only relevant to those who call :cfunc:`malloc` directly.
179 Also note that, with the important exception of :cfunc:`PyArg_ParseTuple` and
180 friends, functions that return an integer status usually return a positive value
181 or zero for success and ``-1`` for failure, like Unix system calls.
183 Finally, be careful to clean up garbage (by making :cfunc:`Py_XDECREF` or
184 :cfunc:`Py_DECREF` calls for objects you have already created) when you return
187 The choice of which exception to raise is entirely yours. There are predeclared
188 C objects corresponding to all built-in Python exceptions, such as
189 :cdata:`PyExc_ZeroDivisionError`, which you can use directly. Of course, you
190 should choose exceptions wisely --- don't use :cdata:`PyExc_TypeError` to mean
191 that a file couldn't be opened (that should probably be :cdata:`PyExc_IOError`).
192 If something's wrong with the argument list, the :cfunc:`PyArg_ParseTuple`
193 function usually raises :cdata:`PyExc_TypeError`. If you have an argument whose
194 value must be in a particular range or must satisfy other conditions,
195 :cdata:`PyExc_ValueError` is appropriate.
197 You can also define a new exception that is unique to your module. For this, you
198 usually declare a static object variable at the beginning of your file::
200 static PyObject *SpamError;
202 and initialize it in your module's initialization function (:cfunc:`initspam`)
203 with an exception object (leaving out the error checking for now)::
210 m = Py_InitModule("spam", SpamMethods);
214 SpamError = PyErr_NewException("spam.error", NULL, NULL);
215 Py_INCREF(SpamError);
216 PyModule_AddObject(m, "error", SpamError);
219 Note that the Python name for the exception object is :exc:`spam.error`. The
220 :cfunc:`PyErr_NewException` function may create a class with the base class
221 being :exc:`Exception` (unless another class is passed in instead of *NULL*),
222 described in :ref:`bltin-exceptions`.
224 Note also that the :cdata:`SpamError` variable retains a reference to the newly
225 created exception class; this is intentional! Since the exception could be
226 removed from the module by external code, an owned reference to the class is
227 needed to ensure that it will not be discarded, causing :cdata:`SpamError` to
228 become a dangling pointer. Should it become a dangling pointer, C code which
229 raises the exception could cause a core dump or other unintended side effects.
231 We discuss the use of PyMODINIT_FUNC as a function return type later in this
240 Going back to our example function, you should now be able to understand this
243 if (!PyArg_ParseTuple(args, "s", &command))
246 It returns *NULL* (the error indicator for functions returning object pointers)
247 if an error is detected in the argument list, relying on the exception set by
248 :cfunc:`PyArg_ParseTuple`. Otherwise the string value of the argument has been
249 copied to the local variable :cdata:`command`. This is a pointer assignment and
250 you are not supposed to modify the string to which it points (so in Standard C,
251 the variable :cdata:`command` should properly be declared as ``const char
254 The next statement is a call to the Unix function :cfunc:`system`, passing it
255 the string we just got from :cfunc:`PyArg_ParseTuple`::
257 sts = system(command);
259 Our :func:`spam.system` function must return the value of :cdata:`sts` as a
260 Python object. This is done using the function :cfunc:`Py_BuildValue`, which is
261 something like the inverse of :cfunc:`PyArg_ParseTuple`: it takes a format
262 string and an arbitrary number of C values, and returns a new Python object.
263 More info on :cfunc:`Py_BuildValue` is given later. ::
265 return Py_BuildValue("i", sts);
267 In this case, it will return an integer object. (Yes, even integers are objects
268 on the heap in Python!)
270 If you have a C function that returns no useful argument (a function returning
271 :ctype:`void`), the corresponding Python function must return ``None``. You
272 need this idiom to do so (which is implemented by the :cmacro:`Py_RETURN_NONE`
278 :cdata:`Py_None` is the C name for the special Python object ``None``. It is a
279 genuine Python object rather than a *NULL* pointer, which means "error" in most
280 contexts, as we have seen.
285 The Module's Method Table and Initialization Function
286 =====================================================
288 I promised to show how :cfunc:`spam_system` is called from Python programs.
289 First, we need to list its name and address in a "method table"::
291 static PyMethodDef SpamMethods[] = {
293 {"system", spam_system, METH_VARARGS,
294 "Execute a shell command."},
296 {NULL, NULL, 0, NULL} /* Sentinel */
299 Note the third entry (``METH_VARARGS``). This is a flag telling the interpreter
300 the calling convention to be used for the C function. It should normally always
301 be ``METH_VARARGS`` or ``METH_VARARGS | METH_KEYWORDS``; a value of ``0`` means
302 that an obsolete variant of :cfunc:`PyArg_ParseTuple` is used.
304 When using only ``METH_VARARGS``, the function should expect the Python-level
305 parameters to be passed in as a tuple acceptable for parsing via
306 :cfunc:`PyArg_ParseTuple`; more information on this function is provided below.
308 The :const:`METH_KEYWORDS` bit may be set in the third field if keyword
309 arguments should be passed to the function. In this case, the C function should
310 accept a third ``PyObject *`` parameter which will be a dictionary of keywords.
311 Use :cfunc:`PyArg_ParseTupleAndKeywords` to parse the arguments to such a
314 The method table must be passed to the interpreter in the module's
315 initialization function. The initialization function must be named
316 :cfunc:`initname`, where *name* is the name of the module, and should be the
317 only non-\ ``static`` item defined in the module file::
322 (void) Py_InitModule("spam", SpamMethods);
325 Note that PyMODINIT_FUNC declares the function as ``void`` return type,
326 declares any special linkage declarations required by the platform, and for C++
327 declares the function as ``extern "C"``.
329 When the Python program imports module :mod:`spam` for the first time,
330 :cfunc:`initspam` is called. (See below for comments about embedding Python.)
331 It calls :cfunc:`Py_InitModule`, which creates a "module object" (which is
332 inserted in the dictionary ``sys.modules`` under the key ``"spam"``), and
333 inserts built-in function objects into the newly created module based upon the
334 table (an array of :ctype:`PyMethodDef` structures) that was passed as its
335 second argument. :cfunc:`Py_InitModule` returns a pointer to the module object
336 that it creates (which is unused here). It may abort with a fatal error for
337 certain errors, or return *NULL* if the module could not be initialized
340 When embedding Python, the :cfunc:`initspam` function is not called
341 automatically unless there's an entry in the :cdata:`_PyImport_Inittab` table.
342 The easiest way to handle this is to statically initialize your
343 statically-linked modules by directly calling :cfunc:`initspam` after the call
344 to :cfunc:`Py_Initialize`::
347 main(int argc, char *argv[])
349 /* Pass argv[0] to the Python interpreter */
350 Py_SetProgramName(argv[0]);
352 /* Initialize the Python interpreter. Required. */
355 /* Add a static module */
358 An example may be found in the file :file:`Demo/embed/demo.c` in the Python
363 Removing entries from ``sys.modules`` or importing compiled modules into
364 multiple interpreters within a process (or following a :cfunc:`fork` without an
365 intervening :cfunc:`exec`) can create problems for some extension modules.
366 Extension module authors should exercise caution when initializing internal data
367 structures. Note also that the :func:`reload` function can be used with
368 extension modules, and will call the module initialization function
369 (:cfunc:`initspam` in the example), but will not load the module again if it was
370 loaded from a dynamically loadable object file (:file:`.so` on Unix,
371 :file:`.dll` on Windows).
373 A more substantial example module is included in the Python source distribution
374 as :file:`Modules/xxmodule.c`. This file may be used as a template or simply
375 read as an example. The :program:`modulator.py` script included in the source
376 distribution or Windows install provides a simple graphical user interface for
377 declaring the functions and objects which a module should implement, and can
378 generate a template which can be filled in. The script lives in the
379 :file:`Tools/modulator/` directory; see the :file:`README` file there for more
385 Compilation and Linkage
386 =======================
388 There are two more things to do before you can use your new extension: compiling
389 and linking it with the Python system. If you use dynamic loading, the details
390 may depend on the style of dynamic loading your system uses; see the chapters
391 about building extension modules (chapter :ref:`building`) and additional
392 information that pertains only to building on Windows (chapter
393 :ref:`building-on-windows`) for more information about this.
395 If you can't use dynamic loading, or if you want to make your module a permanent
396 part of the Python interpreter, you will have to change the configuration setup
397 and rebuild the interpreter. Luckily, this is very simple on Unix: just place
398 your file (:file:`spammodule.c` for example) in the :file:`Modules/` directory
399 of an unpacked source distribution, add a line to the file
400 :file:`Modules/Setup.local` describing your file::
404 and rebuild the interpreter by running :program:`make` in the toplevel
405 directory. You can also run :program:`make` in the :file:`Modules/`
406 subdirectory, but then you must first rebuild :file:`Makefile` there by running
407 ':program:`make` Makefile'. (This is necessary each time you change the
410 If your module requires additional libraries to link with, these can be listed
411 on the line in the configuration file as well, for instance::
413 spam spammodule.o -lX11
418 Calling Python Functions from C
419 ===============================
421 So far we have concentrated on making C functions callable from Python. The
422 reverse is also useful: calling Python functions from C. This is especially the
423 case for libraries that support so-called "callback" functions. If a C
424 interface makes use of callbacks, the equivalent Python often needs to provide a
425 callback mechanism to the Python programmer; the implementation will require
426 calling the Python callback functions from a C callback. Other uses are also
429 Fortunately, the Python interpreter is easily called recursively, and there is a
430 standard interface to call a Python function. (I won't dwell on how to call the
431 Python parser with a particular string as input --- if you're interested, have a
432 look at the implementation of the :option:`-c` command line option in
433 :file:`Modules/main.c` from the Python source code.)
435 Calling a Python function is easy. First, the Python program must somehow pass
436 you the Python function object. You should provide a function (or some other
437 interface) to do this. When this function is called, save a pointer to the
438 Python function object (be careful to :cfunc:`Py_INCREF` it!) in a global
439 variable --- or wherever you see fit. For example, the following function might
440 be part of a module definition::
442 static PyObject *my_callback = NULL;
445 my_set_callback(PyObject *dummy, PyObject *args)
447 PyObject *result = NULL;
450 if (PyArg_ParseTuple(args, "O:set_callback", &temp)) {
451 if (!PyCallable_Check(temp)) {
452 PyErr_SetString(PyExc_TypeError, "parameter must be callable");
455 Py_XINCREF(temp); /* Add a reference to new callback */
456 Py_XDECREF(my_callback); /* Dispose of previous callback */
457 my_callback = temp; /* Remember new callback */
458 /* Boilerplate to return "None" */
465 This function must be registered with the interpreter using the
466 :const:`METH_VARARGS` flag; this is described in section :ref:`methodtable`. The
467 :cfunc:`PyArg_ParseTuple` function and its arguments are documented in section
470 The macros :cfunc:`Py_XINCREF` and :cfunc:`Py_XDECREF` increment/decrement the
471 reference count of an object and are safe in the presence of *NULL* pointers
472 (but note that *temp* will not be *NULL* in this context). More info on them
473 in section :ref:`refcounts`.
475 .. index:: single: PyObject_CallObject()
477 Later, when it is time to call the function, you call the C function
478 :cfunc:`PyObject_CallObject`. This function has two arguments, both pointers to
479 arbitrary Python objects: the Python function, and the argument list. The
480 argument list must always be a tuple object, whose length is the number of
481 arguments. To call the Python function with no arguments, pass in NULL, or
482 an empty tuple; to call it with one argument, pass a singleton tuple.
483 :cfunc:`Py_BuildValue` returns a tuple when its format string consists of zero
484 or more format codes between parentheses. For example::
492 /* Time to call the callback */
493 arglist = Py_BuildValue("(i)", arg);
494 result = PyObject_CallObject(my_callback, arglist);
497 :cfunc:`PyObject_CallObject` returns a Python object pointer: this is the return
498 value of the Python function. :cfunc:`PyObject_CallObject` is
499 "reference-count-neutral" with respect to its arguments. In the example a new
500 tuple was created to serve as the argument list, which is :cfunc:`Py_DECREF`\
501 -ed immediately after the call.
503 The return value of :cfunc:`PyObject_CallObject` is "new": either it is a brand
504 new object, or it is an existing object whose reference count has been
505 incremented. So, unless you want to save it in a global variable, you should
506 somehow :cfunc:`Py_DECREF` the result, even (especially!) if you are not
507 interested in its value.
509 Before you do this, however, it is important to check that the return value
510 isn't *NULL*. If it is, the Python function terminated by raising an exception.
511 If the C code that called :cfunc:`PyObject_CallObject` is called from Python, it
512 should now return an error indication to its Python caller, so the interpreter
513 can print a stack trace, or the calling Python code can handle the exception.
514 If this is not possible or desirable, the exception should be cleared by calling
515 :cfunc:`PyErr_Clear`. For example::
518 return NULL; /* Pass error back */
522 Depending on the desired interface to the Python callback function, you may also
523 have to provide an argument list to :cfunc:`PyObject_CallObject`. In some cases
524 the argument list is also provided by the Python program, through the same
525 interface that specified the callback function. It can then be saved and used
526 in the same manner as the function object. In other cases, you may have to
527 construct a new tuple to pass as the argument list. The simplest way to do this
528 is to call :cfunc:`Py_BuildValue`. For example, if you want to pass an integral
529 event code, you might use the following code::
533 arglist = Py_BuildValue("(l)", eventcode);
534 result = PyObject_CallObject(my_callback, arglist);
537 return NULL; /* Pass error back */
538 /* Here maybe use the result */
541 Note the placement of ``Py_DECREF(arglist)`` immediately after the call, before
542 the error check! Also note that strictly speaking this code is not complete:
543 :cfunc:`Py_BuildValue` may run out of memory, and this should be checked.
545 You may also call a function with keyword arguments by using
546 :cfunc:`PyObject_Call`, which supports arguments and keyword arguments. As in
547 the above example, we use :cfunc:`Py_BuildValue` to construct the dictionary. ::
551 dict = Py_BuildValue("{s:i}", "name", val);
552 result = PyObject_Call(my_callback, NULL, dict);
555 return NULL; /* Pass error back */
556 /* Here maybe use the result */
562 Extracting Parameters in Extension Functions
563 ============================================
565 .. index:: single: PyArg_ParseTuple()
567 The :cfunc:`PyArg_ParseTuple` function is declared as follows::
569 int PyArg_ParseTuple(PyObject *arg, char *format, ...);
571 The *arg* argument must be a tuple object containing an argument list passed
572 from Python to a C function. The *format* argument must be a format string,
573 whose syntax is explained in :ref:`arg-parsing` in the Python/C API Reference
574 Manual. The remaining arguments must be addresses of variables whose type is
575 determined by the format string.
577 Note that while :cfunc:`PyArg_ParseTuple` checks that the Python arguments have
578 the required types, it cannot check the validity of the addresses of C variables
579 passed to the call: if you make mistakes there, your code will probably crash or
580 at least overwrite random bits in memory. So be careful!
582 Note that any Python object references which are provided to the caller are
583 *borrowed* references; do not decrement their reference count!
593 ok = PyArg_ParseTuple(args, ""); /* No arguments */
594 /* Python call: f() */
598 ok = PyArg_ParseTuple(args, "s", &s); /* A string */
599 /* Possible Python call: f('whoops!') */
603 ok = PyArg_ParseTuple(args, "lls", &k, &l, &s); /* Two longs and a string */
604 /* Possible Python call: f(1, 2, 'three') */
608 ok = PyArg_ParseTuple(args, "(ii)s#", &i, &j, &s, &size);
609 /* A pair of ints and a string, whose size is also returned */
610 /* Possible Python call: f((1, 2), 'three') */
616 const char *mode = "r";
618 ok = PyArg_ParseTuple(args, "s|si", &file, &mode, &bufsize);
619 /* A string, and optionally another string and an integer */
620 /* Possible Python calls:
623 f('spam', 'wb', 100000) */
629 int left, top, right, bottom, h, v;
630 ok = PyArg_ParseTuple(args, "((ii)(ii))(ii)",
631 &left, &top, &right, &bottom, &h, &v);
632 /* A rectangle and a point */
633 /* Possible Python call:
634 f(((0, 0), (400, 300)), (10, 10)) */
641 ok = PyArg_ParseTuple(args, "D:myfunction", &c);
642 /* a complex, also providing a function name for errors */
643 /* Possible Python call: myfunction(1+2j) */
647 .. _parsetupleandkeywords:
649 Keyword Parameters for Extension Functions
650 ==========================================
652 .. index:: single: PyArg_ParseTupleAndKeywords()
654 The :cfunc:`PyArg_ParseTupleAndKeywords` function is declared as follows::
656 int PyArg_ParseTupleAndKeywords(PyObject *arg, PyObject *kwdict,
657 char *format, char *kwlist[], ...);
659 The *arg* and *format* parameters are identical to those of the
660 :cfunc:`PyArg_ParseTuple` function. The *kwdict* parameter is the dictionary of
661 keywords received as the third parameter from the Python runtime. The *kwlist*
662 parameter is a *NULL*-terminated list of strings which identify the parameters;
663 the names are matched with the type information from *format* from left to
664 right. On success, :cfunc:`PyArg_ParseTupleAndKeywords` returns true, otherwise
665 it returns false and raises an appropriate exception.
669 Nested tuples cannot be parsed when using keyword arguments! Keyword parameters
670 passed in which are not present in the *kwlist* will cause :exc:`TypeError` to
673 .. index:: single: Philbrick, Geoff
675 Here is an example module which uses keywords, based on an example by Geoff
676 Philbrick (philbrick@hks.com)::
681 keywdarg_parrot(PyObject *self, PyObject *args, PyObject *keywds)
684 char *state = "a stiff";
685 char *action = "voom";
686 char *type = "Norwegian Blue";
688 static char *kwlist[] = {"voltage", "state", "action", "type", NULL};
690 if (!PyArg_ParseTupleAndKeywords(args, keywds, "i|sss", kwlist,
691 &voltage, &state, &action, &type))
694 printf("-- This parrot wouldn't %s if you put %i Volts through it.\n",
696 printf("-- Lovely plumage, the %s -- It's %s!\n", type, state);
703 static PyMethodDef keywdarg_methods[] = {
704 /* The cast of the function is necessary since PyCFunction values
705 * only take two PyObject* parameters, and keywdarg_parrot() takes
708 {"parrot", (PyCFunction)keywdarg_parrot, METH_VARARGS | METH_KEYWORDS,
709 "Print a lovely skit to standard output."},
710 {NULL, NULL, 0, NULL} /* sentinel */
718 /* Create the module and add the functions */
719 Py_InitModule("keywdarg", keywdarg_methods);
725 Building Arbitrary Values
726 =========================
728 This function is the counterpart to :cfunc:`PyArg_ParseTuple`. It is declared
731 PyObject *Py_BuildValue(char *format, ...);
733 It recognizes a set of format units similar to the ones recognized by
734 :cfunc:`PyArg_ParseTuple`, but the arguments (which are input to the function,
735 not output) must not be pointers, just values. It returns a new Python object,
736 suitable for returning from a C function called from Python.
738 One difference with :cfunc:`PyArg_ParseTuple`: while the latter requires its
739 first argument to be a tuple (since Python argument lists are always represented
740 as tuples internally), :cfunc:`Py_BuildValue` does not always build a tuple. It
741 builds a tuple only if its format string contains two or more format units. If
742 the format string is empty, it returns ``None``; if it contains exactly one
743 format unit, it returns whatever object is described by that format unit. To
744 force it to return a tuple of size 0 or one, parenthesize the format string.
746 Examples (to the left the call, to the right the resulting Python value)::
748 Py_BuildValue("") None
749 Py_BuildValue("i", 123) 123
750 Py_BuildValue("iii", 123, 456, 789) (123, 456, 789)
751 Py_BuildValue("s", "hello") 'hello'
752 Py_BuildValue("ss", "hello", "world") ('hello', 'world')
753 Py_BuildValue("s#", "hello", 4) 'hell'
754 Py_BuildValue("()") ()
755 Py_BuildValue("(i)", 123) (123,)
756 Py_BuildValue("(ii)", 123, 456) (123, 456)
757 Py_BuildValue("(i,i)", 123, 456) (123, 456)
758 Py_BuildValue("[i,i]", 123, 456) [123, 456]
759 Py_BuildValue("{s:i,s:i}",
760 "abc", 123, "def", 456) {'abc': 123, 'def': 456}
761 Py_BuildValue("((ii)(ii)) (ii)",
762 1, 2, 3, 4, 5, 6) (((1, 2), (3, 4)), (5, 6))
770 In languages like C or C++, the programmer is responsible for dynamic allocation
771 and deallocation of memory on the heap. In C, this is done using the functions
772 :cfunc:`malloc` and :cfunc:`free`. In C++, the operators ``new`` and
773 ``delete`` are used with essentially the same meaning and we'll restrict
774 the following discussion to the C case.
776 Every block of memory allocated with :cfunc:`malloc` should eventually be
777 returned to the pool of available memory by exactly one call to :cfunc:`free`.
778 It is important to call :cfunc:`free` at the right time. If a block's address
779 is forgotten but :cfunc:`free` is not called for it, the memory it occupies
780 cannot be reused until the program terminates. This is called a :dfn:`memory
781 leak`. On the other hand, if a program calls :cfunc:`free` for a block and then
782 continues to use the block, it creates a conflict with re-use of the block
783 through another :cfunc:`malloc` call. This is called :dfn:`using freed memory`.
784 It has the same bad consequences as referencing uninitialized data --- core
785 dumps, wrong results, mysterious crashes.
787 Common causes of memory leaks are unusual paths through the code. For instance,
788 a function may allocate a block of memory, do some calculation, and then free
789 the block again. Now a change in the requirements for the function may add a
790 test to the calculation that detects an error condition and can return
791 prematurely from the function. It's easy to forget to free the allocated memory
792 block when taking this premature exit, especially when it is added later to the
793 code. Such leaks, once introduced, often go undetected for a long time: the
794 error exit is taken only in a small fraction of all calls, and most modern
795 machines have plenty of virtual memory, so the leak only becomes apparent in a
796 long-running process that uses the leaking function frequently. Therefore, it's
797 important to prevent leaks from happening by having a coding convention or
798 strategy that minimizes this kind of errors.
800 Since Python makes heavy use of :cfunc:`malloc` and :cfunc:`free`, it needs a
801 strategy to avoid memory leaks as well as the use of freed memory. The chosen
802 method is called :dfn:`reference counting`. The principle is simple: every
803 object contains a counter, which is incremented when a reference to the object
804 is stored somewhere, and which is decremented when a reference to it is deleted.
805 When the counter reaches zero, the last reference to the object has been deleted
806 and the object is freed.
808 An alternative strategy is called :dfn:`automatic garbage collection`.
809 (Sometimes, reference counting is also referred to as a garbage collection
810 strategy, hence my use of "automatic" to distinguish the two.) The big
811 advantage of automatic garbage collection is that the user doesn't need to call
812 :cfunc:`free` explicitly. (Another claimed advantage is an improvement in speed
813 or memory usage --- this is no hard fact however.) The disadvantage is that for
814 C, there is no truly portable automatic garbage collector, while reference
815 counting can be implemented portably (as long as the functions :cfunc:`malloc`
816 and :cfunc:`free` are available --- which the C Standard guarantees). Maybe some
817 day a sufficiently portable automatic garbage collector will be available for C.
818 Until then, we'll have to live with reference counts.
820 While Python uses the traditional reference counting implementation, it also
821 offers a cycle detector that works to detect reference cycles. This allows
822 applications to not worry about creating direct or indirect circular references;
823 these are the weakness of garbage collection implemented using only reference
824 counting. Reference cycles consist of objects which contain (possibly indirect)
825 references to themselves, so that each object in the cycle has a reference count
826 which is non-zero. Typical reference counting implementations are not able to
827 reclaim the memory belonging to any objects in a reference cycle, or referenced
828 from the objects in the cycle, even though there are no further references to
831 The cycle detector is able to detect garbage cycles and can reclaim them so long
832 as there are no finalizers implemented in Python (:meth:`__del__` methods).
833 When there are such finalizers, the detector exposes the cycles through the
834 :mod:`gc` module (specifically, the
835 ``garbage`` variable in that module). The :mod:`gc` module also exposes a way
836 to run the detector (the :func:`collect` function), as well as configuration
837 interfaces and the ability to disable the detector at runtime. The cycle
838 detector is considered an optional component; though it is included by default,
839 it can be disabled at build time using the :option:`--without-cycle-gc` option
840 to the :program:`configure` script on Unix platforms (including Mac OS X) or by
841 removing the definition of ``WITH_CYCLE_GC`` in the :file:`pyconfig.h` header on
842 other platforms. If the cycle detector is disabled in this way, the :mod:`gc`
843 module will not be available.
846 .. _refcountsinpython:
848 Reference Counting in Python
849 ----------------------------
851 There are two macros, ``Py_INCREF(x)`` and ``Py_DECREF(x)``, which handle the
852 incrementing and decrementing of the reference count. :cfunc:`Py_DECREF` also
853 frees the object when the count reaches zero. For flexibility, it doesn't call
854 :cfunc:`free` directly --- rather, it makes a call through a function pointer in
855 the object's :dfn:`type object`. For this purpose (and others), every object
856 also contains a pointer to its type object.
858 The big question now remains: when to use ``Py_INCREF(x)`` and ``Py_DECREF(x)``?
859 Let's first introduce some terms. Nobody "owns" an object; however, you can
860 :dfn:`own a reference` to an object. An object's reference count is now defined
861 as the number of owned references to it. The owner of a reference is
862 responsible for calling :cfunc:`Py_DECREF` when the reference is no longer
863 needed. Ownership of a reference can be transferred. There are three ways to
864 dispose of an owned reference: pass it on, store it, or call :cfunc:`Py_DECREF`.
865 Forgetting to dispose of an owned reference creates a memory leak.
867 It is also possible to :dfn:`borrow` [#]_ a reference to an object. The
868 borrower of a reference should not call :cfunc:`Py_DECREF`. The borrower must
869 not hold on to the object longer than the owner from which it was borrowed.
870 Using a borrowed reference after the owner has disposed of it risks using freed
871 memory and should be avoided completely. [#]_
873 The advantage of borrowing over owning a reference is that you don't need to
874 take care of disposing of the reference on all possible paths through the code
875 --- in other words, with a borrowed reference you don't run the risk of leaking
876 when a premature exit is taken. The disadvantage of borrowing over owning is
877 that there are some subtle situations where in seemingly correct code a borrowed
878 reference can be used after the owner from which it was borrowed has in fact
881 A borrowed reference can be changed into an owned reference by calling
882 :cfunc:`Py_INCREF`. This does not affect the status of the owner from which the
883 reference was borrowed --- it creates a new owned reference, and gives full
884 owner responsibilities (the new owner must dispose of the reference properly, as
885 well as the previous owner).
893 Whenever an object reference is passed into or out of a function, it is part of
894 the function's interface specification whether ownership is transferred with the
897 Most functions that return a reference to an object pass on ownership with the
898 reference. In particular, all functions whose function it is to create a new
899 object, such as :cfunc:`PyInt_FromLong` and :cfunc:`Py_BuildValue`, pass
900 ownership to the receiver. Even if the object is not actually new, you still
901 receive ownership of a new reference to that object. For instance,
902 :cfunc:`PyInt_FromLong` maintains a cache of popular values and can return a
903 reference to a cached item.
905 Many functions that extract objects from other objects also transfer ownership
906 with the reference, for instance :cfunc:`PyObject_GetAttrString`. The picture
907 is less clear, here, however, since a few common routines are exceptions:
908 :cfunc:`PyTuple_GetItem`, :cfunc:`PyList_GetItem`, :cfunc:`PyDict_GetItem`, and
909 :cfunc:`PyDict_GetItemString` all return references that you borrow from the
910 tuple, list or dictionary.
912 The function :cfunc:`PyImport_AddModule` also returns a borrowed reference, even
913 though it may actually create the object it returns: this is possible because an
914 owned reference to the object is stored in ``sys.modules``.
916 When you pass an object reference into another function, in general, the
917 function borrows the reference from you --- if it needs to store it, it will use
918 :cfunc:`Py_INCREF` to become an independent owner. There are exactly two
919 important exceptions to this rule: :cfunc:`PyTuple_SetItem` and
920 :cfunc:`PyList_SetItem`. These functions take over ownership of the item passed
921 to them --- even if they fail! (Note that :cfunc:`PyDict_SetItem` and friends
922 don't take over ownership --- they are "normal.")
924 When a C function is called from Python, it borrows references to its arguments
925 from the caller. The caller owns a reference to the object, so the borrowed
926 reference's lifetime is guaranteed until the function returns. Only when such a
927 borrowed reference must be stored or passed on, it must be turned into an owned
928 reference by calling :cfunc:`Py_INCREF`.
930 The object reference returned from a C function that is called from Python must
931 be an owned reference --- ownership is transferred from the function to its
940 There are a few situations where seemingly harmless use of a borrowed reference
941 can lead to problems. These all have to do with implicit invocations of the
942 interpreter, which can cause the owner of a reference to dispose of it.
944 The first and most important case to know about is using :cfunc:`Py_DECREF` on
945 an unrelated object while borrowing a reference to a list item. For instance::
950 PyObject *item = PyList_GetItem(list, 0);
952 PyList_SetItem(list, 1, PyInt_FromLong(0L));
953 PyObject_Print(item, stdout, 0); /* BUG! */
956 This function first borrows a reference to ``list[0]``, then replaces
957 ``list[1]`` with the value ``0``, and finally prints the borrowed reference.
958 Looks harmless, right? But it's not!
960 Let's follow the control flow into :cfunc:`PyList_SetItem`. The list owns
961 references to all its items, so when item 1 is replaced, it has to dispose of
962 the original item 1. Now let's suppose the original item 1 was an instance of a
963 user-defined class, and let's further suppose that the class defined a
964 :meth:`__del__` method. If this class instance has a reference count of 1,
965 disposing of it will call its :meth:`__del__` method.
967 Since it is written in Python, the :meth:`__del__` method can execute arbitrary
968 Python code. Could it perhaps do something to invalidate the reference to
969 ``item`` in :cfunc:`bug`? You bet! Assuming that the list passed into
970 :cfunc:`bug` is accessible to the :meth:`__del__` method, it could execute a
971 statement to the effect of ``del list[0]``, and assuming this was the last
972 reference to that object, it would free the memory associated with it, thereby
973 invalidating ``item``.
975 The solution, once you know the source of the problem, is easy: temporarily
976 increment the reference count. The correct version of the function reads::
979 no_bug(PyObject *list)
981 PyObject *item = PyList_GetItem(list, 0);
984 PyList_SetItem(list, 1, PyInt_FromLong(0L));
985 PyObject_Print(item, stdout, 0);
989 This is a true story. An older version of Python contained variants of this bug
990 and someone spent a considerable amount of time in a C debugger to figure out
991 why his :meth:`__del__` methods would fail...
993 The second case of problems with a borrowed reference is a variant involving
994 threads. Normally, multiple threads in the Python interpreter can't get in each
995 other's way, because there is a global lock protecting Python's entire object
996 space. However, it is possible to temporarily release this lock using the macro
997 :cmacro:`Py_BEGIN_ALLOW_THREADS`, and to re-acquire it using
998 :cmacro:`Py_END_ALLOW_THREADS`. This is common around blocking I/O calls, to
999 let other threads use the processor while waiting for the I/O to complete.
1000 Obviously, the following function has the same problem as the previous one::
1005 PyObject *item = PyList_GetItem(list, 0);
1006 Py_BEGIN_ALLOW_THREADS
1007 ...some blocking I/O call...
1008 Py_END_ALLOW_THREADS
1009 PyObject_Print(item, stdout, 0); /* BUG! */
1018 In general, functions that take object references as arguments do not expect you
1019 to pass them *NULL* pointers, and will dump core (or cause later core dumps) if
1020 you do so. Functions that return object references generally return *NULL* only
1021 to indicate that an exception occurred. The reason for not testing for *NULL*
1022 arguments is that functions often pass the objects they receive on to other
1023 function --- if each function were to test for *NULL*, there would be a lot of
1024 redundant tests and the code would run more slowly.
1026 It is better to test for *NULL* only at the "source:" when a pointer that may be
1027 *NULL* is received, for example, from :cfunc:`malloc` or from a function that
1028 may raise an exception.
1030 The macros :cfunc:`Py_INCREF` and :cfunc:`Py_DECREF` do not check for *NULL*
1031 pointers --- however, their variants :cfunc:`Py_XINCREF` and :cfunc:`Py_XDECREF`
1034 The macros for checking for a particular object type (``Pytype_Check()``) don't
1035 check for *NULL* pointers --- again, there is much code that calls several of
1036 these in a row to test an object against various different expected types, and
1037 this would generate redundant tests. There are no variants with *NULL*
1040 The C function calling mechanism guarantees that the argument list passed to C
1041 functions (``args`` in the examples) is never *NULL* --- in fact it guarantees
1042 that it is always a tuple. [#]_
1044 It is a severe error to ever let a *NULL* pointer "escape" to the Python user.
1047 A pedagogically buggy example, along the lines of the previous listing, would
1048 be helpful here -- showing in more concrete terms what sort of actions could
1049 cause the problem. I can't very well imagine it from the description.
1054 Writing Extensions in C++
1055 =========================
1057 It is possible to write extension modules in C++. Some restrictions apply. If
1058 the main program (the Python interpreter) is compiled and linked by the C
1059 compiler, global or static objects with constructors cannot be used. This is
1060 not a problem if the main program is linked by the C++ compiler. Functions that
1061 will be called by the Python interpreter (in particular, module initialization
1062 functions) have to be declared using ``extern "C"``. It is unnecessary to
1063 enclose the Python header files in ``extern "C" {...}`` --- they use this form
1064 already if the symbol ``__cplusplus`` is defined (all recent C++ compilers
1065 define this symbol).
1070 Providing a C API for an Extension Module
1071 =========================================
1073 .. sectionauthor:: Konrad Hinsen <hinsen@cnrs-orleans.fr>
1076 Many extension modules just provide new functions and types to be used from
1077 Python, but sometimes the code in an extension module can be useful for other
1078 extension modules. For example, an extension module could implement a type
1079 "collection" which works like lists without order. Just like the standard Python
1080 list type has a C API which permits extension modules to create and manipulate
1081 lists, this new collection type should have a set of C functions for direct
1082 manipulation from other extension modules.
1084 At first sight this seems easy: just write the functions (without declaring them
1085 ``static``, of course), provide an appropriate header file, and document
1086 the C API. And in fact this would work if all extension modules were always
1087 linked statically with the Python interpreter. When modules are used as shared
1088 libraries, however, the symbols defined in one module may not be visible to
1089 another module. The details of visibility depend on the operating system; some
1090 systems use one global namespace for the Python interpreter and all extension
1091 modules (Windows, for example), whereas others require an explicit list of
1092 imported symbols at module link time (AIX is one example), or offer a choice of
1093 different strategies (most Unices). And even if symbols are globally visible,
1094 the module whose functions one wishes to call might not have been loaded yet!
1096 Portability therefore requires not to make any assumptions about symbol
1097 visibility. This means that all symbols in extension modules should be declared
1098 ``static``, except for the module's initialization function, in order to
1099 avoid name clashes with other extension modules (as discussed in section
1100 :ref:`methodtable`). And it means that symbols that *should* be accessible from
1101 other extension modules must be exported in a different way.
1103 Python provides a special mechanism to pass C-level information (pointers) from
1104 one extension module to another one: CObjects. A CObject is a Python data type
1105 which stores a pointer (:ctype:`void \*`). CObjects can only be created and
1106 accessed via their C API, but they can be passed around like any other Python
1107 object. In particular, they can be assigned to a name in an extension module's
1108 namespace. Other extension modules can then import this module, retrieve the
1109 value of this name, and then retrieve the pointer from the CObject.
1111 There are many ways in which CObjects can be used to export the C API of an
1112 extension module. Each name could get its own CObject, or all C API pointers
1113 could be stored in an array whose address is published in a CObject. And the
1114 various tasks of storing and retrieving the pointers can be distributed in
1115 different ways between the module providing the code and the client modules.
1117 The following example demonstrates an approach that puts most of the burden on
1118 the writer of the exporting module, which is appropriate for commonly used
1119 library modules. It stores all C API pointers (just one in the example!) in an
1120 array of :ctype:`void` pointers which becomes the value of a CObject. The header
1121 file corresponding to the module provides a macro that takes care of importing
1122 the module and retrieving its C API pointers; client modules only have to call
1123 this macro before accessing the C API.
1125 The exporting module is a modification of the :mod:`spam` module from section
1126 :ref:`extending-simpleexample`. The function :func:`spam.system` does not call
1127 the C library function :cfunc:`system` directly, but a function
1128 :cfunc:`PySpam_System`, which would of course do something more complicated in
1129 reality (such as adding "spam" to every command). This function
1130 :cfunc:`PySpam_System` is also exported to other extension modules.
1132 The function :cfunc:`PySpam_System` is a plain C function, declared
1133 ``static`` like everything else::
1136 PySpam_System(const char *command)
1138 return system(command);
1141 The function :cfunc:`spam_system` is modified in a trivial way::
1144 spam_system(PyObject *self, PyObject *args)
1146 const char *command;
1149 if (!PyArg_ParseTuple(args, "s", &command))
1151 sts = PySpam_System(command);
1152 return Py_BuildValue("i", sts);
1155 In the beginning of the module, right after the line ::
1159 two more lines must be added::
1162 #include "spammodule.h"
1164 The ``#define`` is used to tell the header file that it is being included in the
1165 exporting module, not a client module. Finally, the module's initialization
1166 function must take care of initializing the C API pointer array::
1172 static void *PySpam_API[PySpam_API_pointers];
1173 PyObject *c_api_object;
1175 m = Py_InitModule("spam", SpamMethods);
1179 /* Initialize the C API pointer array */
1180 PySpam_API[PySpam_System_NUM] = (void *)PySpam_System;
1182 /* Create a CObject containing the API pointer array's address */
1183 c_api_object = PyCObject_FromVoidPtr((void *)PySpam_API, NULL);
1185 if (c_api_object != NULL)
1186 PyModule_AddObject(m, "_C_API", c_api_object);
1189 Note that ``PySpam_API`` is declared ``static``; otherwise the pointer
1190 array would disappear when :func:`initspam` terminates!
1192 The bulk of the work is in the header file :file:`spammodule.h`, which looks
1195 #ifndef Py_SPAMMODULE_H
1196 #define Py_SPAMMODULE_H
1201 /* Header file for spammodule */
1203 /* C API functions */
1204 #define PySpam_System_NUM 0
1205 #define PySpam_System_RETURN int
1206 #define PySpam_System_PROTO (const char *command)
1208 /* Total number of C API pointers */
1209 #define PySpam_API_pointers 1
1213 /* This section is used when compiling spammodule.c */
1215 static PySpam_System_RETURN PySpam_System PySpam_System_PROTO;
1218 /* This section is used in modules that use spammodule's API */
1220 static void **PySpam_API;
1222 #define PySpam_System \
1223 (*(PySpam_System_RETURN (*)PySpam_System_PROTO) PySpam_API[PySpam_System_NUM])
1225 /* Return -1 and set exception on error, 0 on success. */
1229 PyObject *c_api_object;
1232 module = PyImport_ImportModule("spam");
1236 c_api_object = PyObject_GetAttrString(module, "_C_API");
1237 if (c_api_object == NULL) {
1241 if (PyCObject_Check(c_api_object))
1242 PySpam_API = (void **)PyCObject_AsVoidPtr(c_api_object);
1244 Py_DECREF(c_api_object);
1255 #endif /* !defined(Py_SPAMMODULE_H) */
1257 All that a client module must do in order to have access to the function
1258 :cfunc:`PySpam_System` is to call the function (or rather macro)
1259 :cfunc:`import_spam` in its initialization function::
1266 m = Py_InitModule("client", ClientMethods);
1269 if (import_spam() < 0)
1271 /* additional initialization can happen here */
1274 The main disadvantage of this approach is that the file :file:`spammodule.h` is
1275 rather complicated. However, the basic structure is the same for each function
1276 that is exported, so it has to be learned only once.
1278 Finally it should be mentioned that CObjects offer additional functionality,
1279 which is especially useful for memory allocation and deallocation of the pointer
1280 stored in a CObject. The details are described in the Python/C API Reference
1281 Manual in the section :ref:`cobjects` and in the implementation of CObjects (files
1282 :file:`Include/cobject.h` and :file:`Objects/cobject.c` in the Python source
1285 .. rubric:: Footnotes
1287 .. [#] An interface for this function already exists in the standard module :mod:`os`
1288 --- it was chosen as a simple and straightforward example.
1290 .. [#] The metaphor of "borrowing" a reference is not completely correct: the owner
1291 still has a copy of the reference.
1293 .. [#] Checking that the reference count is at least 1 **does not work** --- the
1294 reference count itself could be in freed memory and may thus be reused for
1297 .. [#] These guarantees don't hold when you use the "old" style calling convention ---
1298 this is still found in much existing code.