• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1.. highlight:: c
2
3
4.. _memory:
5
6*****************
7Memory Management
8*****************
9
10.. sectionauthor:: Vladimir Marangozov <Vladimir.Marangozov@inrialpes.fr>
11
12
13
14.. _memoryoverview:
15
16Overview
17========
18
19Memory management in Python involves a private heap containing all Python
20objects and data structures. The management of this private heap is ensured
21internally by the *Python memory manager*.  The Python memory manager has
22different components which deal with various dynamic storage management aspects,
23like sharing, segmentation, preallocation or caching.
24
25At the lowest level, a raw memory allocator ensures that there is enough room in
26the private heap for storing all Python-related data by interacting with the
27memory manager of the operating system. On top of the raw memory allocator,
28several object-specific allocators operate on the same heap and implement
29distinct memory management policies adapted to the peculiarities of every object
30type. For example, integer objects are managed differently within the heap than
31strings, tuples or dictionaries because integers imply different storage
32requirements and speed/space tradeoffs. The Python memory manager thus delegates
33some of the work to the object-specific allocators, but ensures that the latter
34operate within the bounds of the private heap.
35
36It is important to understand that the management of the Python heap is
37performed by the interpreter itself and that the user has no control over it,
38even if they regularly manipulate object pointers to memory blocks inside that
39heap.  The allocation of heap space for Python objects and other internal
40buffers is performed on demand by the Python memory manager through the Python/C
41API functions listed in this document.
42
43.. index::
44   single: malloc()
45   single: calloc()
46   single: realloc()
47   single: free()
48
49To avoid memory corruption, extension writers should never try to operate on
50Python objects with the functions exported by the C library: :c:func:`malloc`,
51:c:func:`calloc`, :c:func:`realloc` and :c:func:`free`.  This will result in  mixed
52calls between the C allocator and the Python memory manager with fatal
53consequences, because they implement different algorithms and operate on
54different heaps.  However, one may safely allocate and release memory blocks
55with the C library allocator for individual purposes, as shown in the following
56example::
57
58   PyObject *res;
59   char *buf = (char *) malloc(BUFSIZ); /* for I/O */
60
61   if (buf == NULL)
62       return PyErr_NoMemory();
63   ...Do some I/O operation involving buf...
64   res = PyBytes_FromString(buf);
65   free(buf); /* malloc'ed */
66   return res;
67
68In this example, the memory request for the I/O buffer is handled by the C
69library allocator. The Python memory manager is involved only in the allocation
70of the bytes object returned as a result.
71
72In most situations, however, it is recommended to allocate memory from the
73Python heap specifically because the latter is under control of the Python
74memory manager. For example, this is required when the interpreter is extended
75with new object types written in C. Another reason for using the Python heap is
76the desire to *inform* the Python memory manager about the memory needs of the
77extension module. Even when the requested memory is used exclusively for
78internal, highly-specific purposes, delegating all memory requests to the Python
79memory manager causes the interpreter to have a more accurate image of its
80memory footprint as a whole. Consequently, under certain circumstances, the
81Python memory manager may or may not trigger appropriate actions, like garbage
82collection, memory compaction or other preventive procedures. Note that by using
83the C library allocator as shown in the previous example, the allocated memory
84for the I/O buffer escapes completely the Python memory manager.
85
86.. seealso::
87
88   The :envvar:`PYTHONMALLOC` environment variable can be used to configure
89   the memory allocators used by Python.
90
91   The :envvar:`PYTHONMALLOCSTATS` environment variable can be used to print
92   statistics of the :ref:`pymalloc memory allocator <pymalloc>` every time a
93   new pymalloc object arena is created, and on shutdown.
94
95Allocator Domains
96=================
97
98All allocating functions belong to one of three different "domains" (see also
99:c:type:`PyMemAllocatorDomain`). These domains represent different allocation
100strategies and are optimized for different purposes. The specific details on
101how every domain allocates memory or what internal functions each domain calls
102is considered an implementation detail, but for debugging purposes a simplified
103table can be found at :ref:`here <default-memory-allocators>`. There is no hard
104requirement to use the memory returned by the allocation functions belonging to
105a given domain for only the purposes hinted by that domain (although this is the
106recommended practice). For example, one could use the memory returned by
107:c:func:`PyMem_RawMalloc` for allocating Python objects or the memory returned
108by :c:func:`PyObject_Malloc` for allocating memory for buffers.
109
110The three allocation domains are:
111
112* Raw domain: intended for allocating memory for general-purpose memory
113  buffers where the allocation *must* go to the system allocator or where the
114  allocator can operate without the :term:`GIL`. The memory is requested directly
115  to the system.
116
117* "Mem" domain: intended for allocating memory for Python buffers and
118  general-purpose memory buffers where the allocation must be performed with
119  the :term:`GIL` held. The memory is taken from the Python private heap.
120
121* Object domain: intended for allocating memory belonging to Python objects. The
122  memory is taken from the Python private heap.
123
124When freeing memory previously allocated by the allocating functions belonging to a
125given domain,the matching specific deallocating functions must be used. For example,
126:c:func:`PyMem_Free` must be used to free memory allocated using :c:func:`PyMem_Malloc`.
127
128Raw Memory Interface
129====================
130
131The following function sets are wrappers to the system allocator. These
132functions are thread-safe, the :term:`GIL <global interpreter lock>` does not
133need to be held.
134
135The :ref:`default raw memory allocator <default-memory-allocators>` uses
136the following functions: :c:func:`malloc`, :c:func:`calloc`, :c:func:`realloc`
137and :c:func:`free`; call ``malloc(1)`` (or ``calloc(1, 1)``) when requesting
138zero bytes.
139
140.. versionadded:: 3.4
141
142.. c:function:: void* PyMem_RawMalloc(size_t n)
143
144   Allocates *n* bytes and returns a pointer of type :c:type:`void*` to the
145   allocated memory, or ``NULL`` if the request fails.
146
147   Requesting zero bytes returns a distinct non-``NULL`` pointer if possible, as
148   if ``PyMem_RawMalloc(1)`` had been called instead. The memory will not have
149   been initialized in any way.
150
151
152.. c:function:: void* PyMem_RawCalloc(size_t nelem, size_t elsize)
153
154   Allocates *nelem* elements each whose size in bytes is *elsize* and returns
155   a pointer of type :c:type:`void*` to the allocated memory, or ``NULL`` if the
156   request fails. The memory is initialized to zeros.
157
158   Requesting zero elements or elements of size zero bytes returns a distinct
159   non-``NULL`` pointer if possible, as if ``PyMem_RawCalloc(1, 1)`` had been
160   called instead.
161
162   .. versionadded:: 3.5
163
164
165.. c:function:: void* PyMem_RawRealloc(void *p, size_t n)
166
167   Resizes the memory block pointed to by *p* to *n* bytes. The contents will
168   be unchanged to the minimum of the old and the new sizes.
169
170   If *p* is ``NULL``, the call is equivalent to ``PyMem_RawMalloc(n)``; else if
171   *n* is equal to zero, the memory block is resized but is not freed, and the
172   returned pointer is non-``NULL``.
173
174   Unless *p* is ``NULL``, it must have been returned by a previous call to
175   :c:func:`PyMem_RawMalloc`, :c:func:`PyMem_RawRealloc` or
176   :c:func:`PyMem_RawCalloc`.
177
178   If the request fails, :c:func:`PyMem_RawRealloc` returns ``NULL`` and *p*
179   remains a valid pointer to the previous memory area.
180
181
182.. c:function:: void PyMem_RawFree(void *p)
183
184   Frees the memory block pointed to by *p*, which must have been returned by a
185   previous call to :c:func:`PyMem_RawMalloc`, :c:func:`PyMem_RawRealloc` or
186   :c:func:`PyMem_RawCalloc`.  Otherwise, or if ``PyMem_RawFree(p)`` has been
187   called before, undefined behavior occurs.
188
189   If *p* is ``NULL``, no operation is performed.
190
191
192.. _memoryinterface:
193
194Memory Interface
195================
196
197The following function sets, modeled after the ANSI C standard, but specifying
198behavior when requesting zero bytes, are available for allocating and releasing
199memory from the Python heap.
200
201The :ref:`default memory allocator <default-memory-allocators>` uses the
202:ref:`pymalloc memory allocator <pymalloc>`.
203
204.. warning::
205
206   The :term:`GIL <global interpreter lock>` must be held when using these
207   functions.
208
209.. versionchanged:: 3.6
210
211   The default allocator is now pymalloc instead of system :c:func:`malloc`.
212
213.. c:function:: void* PyMem_Malloc(size_t n)
214
215   Allocates *n* bytes and returns a pointer of type :c:type:`void*` to the
216   allocated memory, or ``NULL`` if the request fails.
217
218   Requesting zero bytes returns a distinct non-``NULL`` pointer if possible, as
219   if ``PyMem_Malloc(1)`` had been called instead. The memory will not have
220   been initialized in any way.
221
222
223.. c:function:: void* PyMem_Calloc(size_t nelem, size_t elsize)
224
225   Allocates *nelem* elements each whose size in bytes is *elsize* and returns
226   a pointer of type :c:type:`void*` to the allocated memory, or ``NULL`` if the
227   request fails. The memory is initialized to zeros.
228
229   Requesting zero elements or elements of size zero bytes returns a distinct
230   non-``NULL`` pointer if possible, as if ``PyMem_Calloc(1, 1)`` had been called
231   instead.
232
233   .. versionadded:: 3.5
234
235
236.. c:function:: void* PyMem_Realloc(void *p, size_t n)
237
238   Resizes the memory block pointed to by *p* to *n* bytes. The contents will be
239   unchanged to the minimum of the old and the new sizes.
240
241   If *p* is ``NULL``, the call is equivalent to ``PyMem_Malloc(n)``; else if *n*
242   is equal to zero, the memory block is resized but is not freed, and the
243   returned pointer is non-``NULL``.
244
245   Unless *p* is ``NULL``, it must have been returned by a previous call to
246   :c:func:`PyMem_Malloc`, :c:func:`PyMem_Realloc` or :c:func:`PyMem_Calloc`.
247
248   If the request fails, :c:func:`PyMem_Realloc` returns ``NULL`` and *p* remains
249   a valid pointer to the previous memory area.
250
251
252.. c:function:: void PyMem_Free(void *p)
253
254   Frees the memory block pointed to by *p*, which must have been returned by a
255   previous call to :c:func:`PyMem_Malloc`, :c:func:`PyMem_Realloc` or
256   :c:func:`PyMem_Calloc`.  Otherwise, or if ``PyMem_Free(p)`` has been called
257   before, undefined behavior occurs.
258
259   If *p* is ``NULL``, no operation is performed.
260
261The following type-oriented macros are provided for convenience.  Note  that
262*TYPE* refers to any C type.
263
264
265.. c:function:: TYPE* PyMem_New(TYPE, size_t n)
266
267   Same as :c:func:`PyMem_Malloc`, but allocates ``(n * sizeof(TYPE))`` bytes of
268   memory.  Returns a pointer cast to :c:type:`TYPE*`.  The memory will not have
269   been initialized in any way.
270
271
272.. c:function:: TYPE* PyMem_Resize(void *p, TYPE, size_t n)
273
274   Same as :c:func:`PyMem_Realloc`, but the memory block is resized to ``(n *
275   sizeof(TYPE))`` bytes.  Returns a pointer cast to :c:type:`TYPE*`. On return,
276   *p* will be a pointer to the new memory area, or ``NULL`` in the event of
277   failure.
278
279   This is a C preprocessor macro; *p* is always reassigned.  Save the original
280   value of *p* to avoid losing memory when handling errors.
281
282
283.. c:function:: void PyMem_Del(void *p)
284
285   Same as :c:func:`PyMem_Free`.
286
287In addition, the following macro sets are provided for calling the Python memory
288allocator directly, without involving the C API functions listed above. However,
289note that their use does not preserve binary compatibility across Python
290versions and is therefore deprecated in extension modules.
291
292* ``PyMem_MALLOC(size)``
293* ``PyMem_NEW(type, size)``
294* ``PyMem_REALLOC(ptr, size)``
295* ``PyMem_RESIZE(ptr, type, size)``
296* ``PyMem_FREE(ptr)``
297* ``PyMem_DEL(ptr)``
298
299
300Object allocators
301=================
302
303The following function sets, modeled after the ANSI C standard, but specifying
304behavior when requesting zero bytes, are available for allocating and releasing
305memory from the Python heap.
306
307.. note::
308    There is no guarantee that the memory returned by these allocators can be
309    successfully casted to a Python object when intercepting the allocating
310    functions in this domain by the methods described in
311    the :ref:`Customize Memory Allocators <customize-memory-allocators>` section.
312
313The :ref:`default object allocator <default-memory-allocators>` uses the
314:ref:`pymalloc memory allocator <pymalloc>`.
315
316.. warning::
317
318   The :term:`GIL <global interpreter lock>` must be held when using these
319   functions.
320
321.. c:function:: void* PyObject_Malloc(size_t n)
322
323   Allocates *n* bytes and returns a pointer of type :c:type:`void*` to the
324   allocated memory, or ``NULL`` if the request fails.
325
326   Requesting zero bytes returns a distinct non-``NULL`` pointer if possible, as
327   if ``PyObject_Malloc(1)`` had been called instead. The memory will not have
328   been initialized in any way.
329
330
331.. c:function:: void* PyObject_Calloc(size_t nelem, size_t elsize)
332
333   Allocates *nelem* elements each whose size in bytes is *elsize* and returns
334   a pointer of type :c:type:`void*` to the allocated memory, or ``NULL`` if the
335   request fails. The memory is initialized to zeros.
336
337   Requesting zero elements or elements of size zero bytes returns a distinct
338   non-``NULL`` pointer if possible, as if ``PyObject_Calloc(1, 1)`` had been called
339   instead.
340
341   .. versionadded:: 3.5
342
343
344.. c:function:: void* PyObject_Realloc(void *p, size_t n)
345
346   Resizes the memory block pointed to by *p* to *n* bytes. The contents will be
347   unchanged to the minimum of the old and the new sizes.
348
349   If *p* is ``NULL``, the call is equivalent to ``PyObject_Malloc(n)``; else if *n*
350   is equal to zero, the memory block is resized but is not freed, and the
351   returned pointer is non-``NULL``.
352
353   Unless *p* is ``NULL``, it must have been returned by a previous call to
354   :c:func:`PyObject_Malloc`, :c:func:`PyObject_Realloc` or :c:func:`PyObject_Calloc`.
355
356   If the request fails, :c:func:`PyObject_Realloc` returns ``NULL`` and *p* remains
357   a valid pointer to the previous memory area.
358
359
360.. c:function:: void PyObject_Free(void *p)
361
362   Frees the memory block pointed to by *p*, which must have been returned by a
363   previous call to :c:func:`PyObject_Malloc`, :c:func:`PyObject_Realloc` or
364   :c:func:`PyObject_Calloc`.  Otherwise, or if ``PyObject_Free(p)`` has been called
365   before, undefined behavior occurs.
366
367   If *p* is ``NULL``, no operation is performed.
368
369
370.. _default-memory-allocators:
371
372Default Memory Allocators
373=========================
374
375Default memory allocators:
376
377===============================  ====================  ==================  =====================  ====================
378Configuration                    Name                  PyMem_RawMalloc     PyMem_Malloc           PyObject_Malloc
379===============================  ====================  ==================  =====================  ====================
380Release build                    ``"pymalloc"``        ``malloc``          ``pymalloc``           ``pymalloc``
381Debug build                      ``"pymalloc_debug"``  ``malloc`` + debug  ``pymalloc`` + debug   ``pymalloc`` + debug
382Release build, without pymalloc  ``"malloc"``          ``malloc``          ``malloc``             ``malloc``
383Debug build, without pymalloc    ``"malloc_debug"``    ``malloc`` + debug  ``malloc`` + debug     ``malloc`` + debug
384===============================  ====================  ==================  =====================  ====================
385
386Legend:
387
388* Name: value for :envvar:`PYTHONMALLOC` environment variable.
389* ``malloc``: system allocators from the standard C library, C functions:
390  :c:func:`malloc`, :c:func:`calloc`, :c:func:`realloc` and :c:func:`free`.
391* ``pymalloc``: :ref:`pymalloc memory allocator <pymalloc>`.
392* "+ debug": with :ref:`debug hooks on the Python memory allocators
393  <pymem-debug-hooks>`.
394* "Debug build": :ref:`Python build in debug mode <debug-build>`.
395
396.. _customize-memory-allocators:
397
398Customize Memory Allocators
399===========================
400
401.. versionadded:: 3.4
402
403.. c:type:: PyMemAllocatorEx
404
405   Structure used to describe a memory block allocator. The structure has
406   four fields:
407
408   +----------------------------------------------------------+---------------------------------------+
409   | Field                                                    | Meaning                               |
410   +==========================================================+=======================================+
411   | ``void *ctx``                                            | user context passed as first argument |
412   +----------------------------------------------------------+---------------------------------------+
413   | ``void* malloc(void *ctx, size_t size)``                 | allocate a memory block               |
414   +----------------------------------------------------------+---------------------------------------+
415   | ``void* calloc(void *ctx, size_t nelem, size_t elsize)`` | allocate a memory block initialized   |
416   |                                                          | with zeros                            |
417   +----------------------------------------------------------+---------------------------------------+
418   | ``void* realloc(void *ctx, void *ptr, size_t new_size)`` | allocate or resize a memory block     |
419   +----------------------------------------------------------+---------------------------------------+
420   | ``void free(void *ctx, void *ptr)``                      | free a memory block                   |
421   +----------------------------------------------------------+---------------------------------------+
422
423   .. versionchanged:: 3.5
424      The :c:type:`PyMemAllocator` structure was renamed to
425      :c:type:`PyMemAllocatorEx` and a new ``calloc`` field was added.
426
427
428.. c:type:: PyMemAllocatorDomain
429
430   Enum used to identify an allocator domain. Domains:
431
432   .. c:macro:: PYMEM_DOMAIN_RAW
433
434      Functions:
435
436      * :c:func:`PyMem_RawMalloc`
437      * :c:func:`PyMem_RawRealloc`
438      * :c:func:`PyMem_RawCalloc`
439      * :c:func:`PyMem_RawFree`
440
441   .. c:macro:: PYMEM_DOMAIN_MEM
442
443      Functions:
444
445      * :c:func:`PyMem_Malloc`,
446      * :c:func:`PyMem_Realloc`
447      * :c:func:`PyMem_Calloc`
448      * :c:func:`PyMem_Free`
449
450   .. c:macro:: PYMEM_DOMAIN_OBJ
451
452      Functions:
453
454      * :c:func:`PyObject_Malloc`
455      * :c:func:`PyObject_Realloc`
456      * :c:func:`PyObject_Calloc`
457      * :c:func:`PyObject_Free`
458
459.. c:function:: void PyMem_GetAllocator(PyMemAllocatorDomain domain, PyMemAllocatorEx *allocator)
460
461   Get the memory block allocator of the specified domain.
462
463
464.. c:function:: void PyMem_SetAllocator(PyMemAllocatorDomain domain, PyMemAllocatorEx *allocator)
465
466   Set the memory block allocator of the specified domain.
467
468   The new allocator must return a distinct non-``NULL`` pointer when requesting
469   zero bytes.
470
471   For the :c:data:`PYMEM_DOMAIN_RAW` domain, the allocator must be
472   thread-safe: the :term:`GIL <global interpreter lock>` is not held when the
473   allocator is called.
474
475   If the new allocator is not a hook (does not call the previous allocator),
476   the :c:func:`PyMem_SetupDebugHooks` function must be called to reinstall the
477   debug hooks on top on the new allocator.
478
479
480.. c:function:: void PyMem_SetupDebugHooks(void)
481
482   Setup :ref:`debug hooks in the Python memory allocators <pymem-debug-hooks>`
483   to detect memory errors.
484
485
486.. _pymem-debug-hooks:
487
488Debug hooks on the Python memory allocators
489===========================================
490
491When :ref:`Python is built in debug mode <debug-build>`, the
492:c:func:`PyMem_SetupDebugHooks` function is called at the :ref:`Python
493preinitialization <c-preinit>` to setup debug hooks on Python memory allocators
494to detect memory errors.
495
496The :envvar:`PYTHONMALLOC` environment variable can be used to install debug
497hooks on a Python compiled in release mode (ex: ``PYTHONMALLOC=debug``).
498
499The :c:func:`PyMem_SetupDebugHooks` function can be used to set debug hooks
500after calling :c:func:`PyMem_SetAllocator`.
501
502These debug hooks fill dynamically allocated memory blocks with special,
503recognizable bit patterns. Newly allocated memory is filled with the byte
504``0xCD`` (``PYMEM_CLEANBYTE``), freed memory is filled with the byte ``0xDD``
505(``PYMEM_DEADBYTE``). Memory blocks are surrounded by "forbidden bytes"
506filled with the byte ``0xFD`` (``PYMEM_FORBIDDENBYTE``). Strings of these bytes
507are unlikely to be valid addresses, floats, or ASCII strings.
508
509Runtime checks:
510
511- Detect API violations. For example, detect if :c:func:`PyObject_Free` is
512  called on a memory block allocated by :c:func:`PyMem_Malloc`.
513- Detect write before the start of the buffer (buffer underflow).
514- Detect write after the end of the buffer (buffer overflow).
515- Check that the :term:`GIL <global interpreter lock>` is held when
516  allocator functions of :c:data:`PYMEM_DOMAIN_OBJ` (ex:
517  :c:func:`PyObject_Malloc`) and :c:data:`PYMEM_DOMAIN_MEM` (ex:
518  :c:func:`PyMem_Malloc`) domains are called.
519
520On error, the debug hooks use the :mod:`tracemalloc` module to get the
521traceback where a memory block was allocated. The traceback is only displayed
522if :mod:`tracemalloc` is tracing Python memory allocations and the memory block
523was traced.
524
525Let *S* = ``sizeof(size_t)``. ``2*S`` bytes are added at each end of each block
526of *N* bytes requested.  The memory layout is like so, where p represents the
527address returned by a malloc-like or realloc-like function (``p[i:j]`` means
528the slice of bytes from ``*(p+i)`` inclusive up to ``*(p+j)`` exclusive; note
529that the treatment of negative indices differs from a Python slice):
530
531``p[-2*S:-S]``
532    Number of bytes originally asked for.  This is a size_t, big-endian (easier
533    to read in a memory dump).
534``p[-S]``
535    API identifier (ASCII character):
536
537    * ``'r'`` for :c:data:`PYMEM_DOMAIN_RAW`.
538    * ``'m'`` for :c:data:`PYMEM_DOMAIN_MEM`.
539    * ``'o'`` for :c:data:`PYMEM_DOMAIN_OBJ`.
540
541``p[-S+1:0]``
542    Copies of PYMEM_FORBIDDENBYTE.  Used to catch under- writes and reads.
543
544``p[0:N]``
545    The requested memory, filled with copies of PYMEM_CLEANBYTE, used to catch
546    reference to uninitialized memory.  When a realloc-like function is called
547    requesting a larger memory block, the new excess bytes are also filled with
548    PYMEM_CLEANBYTE.  When a free-like function is called, these are
549    overwritten with PYMEM_DEADBYTE, to catch reference to freed memory.  When
550    a realloc- like function is called requesting a smaller memory block, the
551    excess old bytes are also filled with PYMEM_DEADBYTE.
552
553``p[N:N+S]``
554    Copies of PYMEM_FORBIDDENBYTE.  Used to catch over- writes and reads.
555
556``p[N+S:N+2*S]``
557    Only used if the ``PYMEM_DEBUG_SERIALNO`` macro is defined (not defined by
558    default).
559
560    A serial number, incremented by 1 on each call to a malloc-like or
561    realloc-like function.  Big-endian ``size_t``.  If "bad memory" is detected
562    later, the serial number gives an excellent way to set a breakpoint on the
563    next run, to capture the instant at which this block was passed out.  The
564    static function bumpserialno() in obmalloc.c is the only place the serial
565    number is incremented, and exists so you can set such a breakpoint easily.
566
567A realloc-like or free-like function first checks that the PYMEM_FORBIDDENBYTE
568bytes at each end are intact.  If they've been altered, diagnostic output is
569written to stderr, and the program is aborted via Py_FatalError().  The other
570main failure mode is provoking a memory error when a program reads up one of
571the special bit patterns and tries to use it as an address.  If you get in a
572debugger then and look at the object, you're likely to see that it's entirely
573filled with PYMEM_DEADBYTE (meaning freed memory is getting used) or
574PYMEM_CLEANBYTE (meaning uninitialized memory is getting used).
575
576.. versionchanged:: 3.6
577   The :c:func:`PyMem_SetupDebugHooks` function now also works on Python
578   compiled in release mode.  On error, the debug hooks now use
579   :mod:`tracemalloc` to get the traceback where a memory block was allocated.
580   The debug hooks now also check if the GIL is held when functions of
581   :c:data:`PYMEM_DOMAIN_OBJ` and :c:data:`PYMEM_DOMAIN_MEM` domains are
582   called.
583
584.. versionchanged:: 3.8
585   Byte patterns ``0xCB`` (``PYMEM_CLEANBYTE``), ``0xDB`` (``PYMEM_DEADBYTE``)
586   and ``0xFB`` (``PYMEM_FORBIDDENBYTE``) have been replaced with ``0xCD``,
587   ``0xDD`` and ``0xFD`` to use the same values than Windows CRT debug
588   ``malloc()`` and ``free()``.
589
590
591.. _pymalloc:
592
593The pymalloc allocator
594======================
595
596Python has a *pymalloc* allocator optimized for small objects (smaller or equal
597to 512 bytes) with a short lifetime. It uses memory mappings called "arenas"
598with a fixed size of 256 KiB. It falls back to :c:func:`PyMem_RawMalloc` and
599:c:func:`PyMem_RawRealloc` for allocations larger than 512 bytes.
600
601*pymalloc* is the :ref:`default allocator <default-memory-allocators>` of the
602:c:data:`PYMEM_DOMAIN_MEM` (ex: :c:func:`PyMem_Malloc`) and
603:c:data:`PYMEM_DOMAIN_OBJ` (ex: :c:func:`PyObject_Malloc`) domains.
604
605The arena allocator uses the following functions:
606
607* :c:func:`VirtualAlloc` and :c:func:`VirtualFree` on Windows,
608* :c:func:`mmap` and :c:func:`munmap` if available,
609* :c:func:`malloc` and :c:func:`free` otherwise.
610
611This allocator is disabled if Python is configured with the
612:option:`--without-pymalloc` option. It can also be disabled at runtime using
613the :envvar:`PYTHONMALLOC` environment variable (ex: ``PYTHONMALLOC=malloc``).
614
615Customize pymalloc Arena Allocator
616----------------------------------
617
618.. versionadded:: 3.4
619
620.. c:type:: PyObjectArenaAllocator
621
622   Structure used to describe an arena allocator. The structure has
623   three fields:
624
625   +--------------------------------------------------+---------------------------------------+
626   | Field                                            | Meaning                               |
627   +==================================================+=======================================+
628   | ``void *ctx``                                    | user context passed as first argument |
629   +--------------------------------------------------+---------------------------------------+
630   | ``void* alloc(void *ctx, size_t size)``          | allocate an arena of size bytes       |
631   +--------------------------------------------------+---------------------------------------+
632   | ``void free(void *ctx, void *ptr, size_t size)`` | free an arena                         |
633   +--------------------------------------------------+---------------------------------------+
634
635.. c:function:: void PyObject_GetArenaAllocator(PyObjectArenaAllocator *allocator)
636
637   Get the arena allocator.
638
639.. c:function:: void PyObject_SetArenaAllocator(PyObjectArenaAllocator *allocator)
640
641   Set the arena allocator.
642
643
644tracemalloc C API
645=================
646
647.. versionadded:: 3.7
648
649.. c:function:: int PyTraceMalloc_Track(unsigned int domain, uintptr_t ptr, size_t size)
650
651   Track an allocated memory block in the :mod:`tracemalloc` module.
652
653   Return ``0`` on success, return ``-1`` on error (failed to allocate memory to
654   store the trace). Return ``-2`` if tracemalloc is disabled.
655
656   If memory block is already tracked, update the existing trace.
657
658.. c:function:: int PyTraceMalloc_Untrack(unsigned int domain, uintptr_t ptr)
659
660   Untrack an allocated memory block in the :mod:`tracemalloc` module.
661   Do nothing if the block was not tracked.
662
663   Return ``-2`` if tracemalloc is disabled, otherwise return ``0``.
664
665
666.. _memoryexamples:
667
668Examples
669========
670
671Here is the example from section :ref:`memoryoverview`, rewritten so that the
672I/O buffer is allocated from the Python heap by using the first function set::
673
674   PyObject *res;
675   char *buf = (char *) PyMem_Malloc(BUFSIZ); /* for I/O */
676
677   if (buf == NULL)
678       return PyErr_NoMemory();
679   /* ...Do some I/O operation involving buf... */
680   res = PyBytes_FromString(buf);
681   PyMem_Free(buf); /* allocated with PyMem_Malloc */
682   return res;
683
684The same code using the type-oriented function set::
685
686   PyObject *res;
687   char *buf = PyMem_New(char, BUFSIZ); /* for I/O */
688
689   if (buf == NULL)
690       return PyErr_NoMemory();
691   /* ...Do some I/O operation involving buf... */
692   res = PyBytes_FromString(buf);
693   PyMem_Del(buf); /* allocated with PyMem_New */
694   return res;
695
696Note that in the two examples above, the buffer is always manipulated via
697functions belonging to the same set. Indeed, it is required to use the same
698memory API family for a given memory block, so that the risk of mixing different
699allocators is reduced to a minimum. The following code sequence contains two
700errors, one of which is labeled as *fatal* because it mixes two different
701allocators operating on different heaps. ::
702
703   char *buf1 = PyMem_New(char, BUFSIZ);
704   char *buf2 = (char *) malloc(BUFSIZ);
705   char *buf3 = (char *) PyMem_Malloc(BUFSIZ);
706   ...
707   PyMem_Del(buf3);  /* Wrong -- should be PyMem_Free() */
708   free(buf2);       /* Right -- allocated via malloc() */
709   free(buf1);       /* Fatal -- should be PyMem_Del()  */
710
711In addition to the functions aimed at handling raw memory blocks from the Python
712heap, objects in Python are allocated and released with :c:func:`PyObject_New`,
713:c:func:`PyObject_NewVar` and :c:func:`PyObject_Del`.
714
715These will be explained in the next chapter on defining and implementing new
716object types in C.
717