5 Buffers and Memoryview Objects
6 ------------------------------
8 .. sectionauthor:: Greg Stein <gstein@lyra.org>
9 .. sectionauthor:: Benjamin Peterson
14 single: buffer interface
16 Python objects implemented in C can export a group of functions called the
17 "buffer interface." These functions can be used by an object to expose its
18 data in a raw, byte-oriented format. Clients of the object can use the buffer
19 interface to access the object data directly, without needing to copy it
22 Two examples of objects that support the buffer interface are strings and
23 arrays. The string object exposes the character contents in the buffer
24 interface's byte-oriented form. An array can also expose its contents, but it
25 should be noted that array elements may be multi-byte values.
27 An example user of the buffer interface is the file object's :meth:`write`
28 method. Any object that can export a series of bytes through the buffer
29 interface can be written to a file. There are a number of format codes to
30 :cfunc:`PyArg_ParseTuple` that operate against an object's buffer interface,
31 returning data from the target object.
33 Starting from version 1.6, Python has been providing Python-level buffer
34 objects and a C-level buffer API so that any built-in or used-defined type can
35 expose its characteristics. Both, however, have been deprecated because of
36 various shortcomings, and have been officially removed in Python 3.0 in favour
37 of a new C-level buffer API and a new Python-level object named
40 The new buffer API has been backported to Python 2.6, and the
41 :class:`memoryview` object has been backported to Python 2.7. It is strongly
42 advised to use them rather than the old APIs, unless you are blocked from
43 doing so for compatibility reasons.
46 The new-style Py_buffer struct
47 ==============================
52 .. cmember:: void *buf
54 A pointer to the start of the memory for the object.
56 .. cmember:: Py_ssize_t len
59 The total length of the memory in bytes.
61 .. cmember:: int readonly
63 An indicator of whether the buffer is read only.
65 .. cmember:: const char *format
68 A *NULL* terminated string in :mod:`struct` module style syntax giving
69 the contents of the elements available through the buffer. If this is
70 *NULL*, ``"B"`` (unsigned bytes) is assumed.
74 The number of dimensions the memory represents as a multi-dimensional
75 array. If it is 0, :cdata:`strides` and :cdata:`suboffsets` must be
78 .. cmember:: Py_ssize_t *shape
80 An array of :ctype:`Py_ssize_t`\s the length of :cdata:`ndim` giving the
81 shape of the memory as a multi-dimensional array. Note that
82 ``((*shape)[0] * ... * (*shape)[ndims-1])*itemsize`` should be equal to
85 .. cmember:: Py_ssize_t *strides
87 An array of :ctype:`Py_ssize_t`\s the length of :cdata:`ndim` giving the
88 number of bytes to skip to get to a new element in each dimension.
90 .. cmember:: Py_ssize_t *suboffsets
92 An array of :ctype:`Py_ssize_t`\s the length of :cdata:`ndim`. If these
93 suboffset numbers are greater than or equal to 0, then the value stored
94 along the indicated dimension is a pointer and the suboffset value
95 dictates how many bytes to add to the pointer after de-referencing. A
96 suboffset value that it negative indicates that no de-referencing should
97 occur (striding in a contiguous memory block).
99 Here is a function that returns a pointer to the element in an N-D array
100 pointed to by an N-dimesional index when there are both non-NULL strides
103 void *get_item_pointer(int ndim, void *buf, Py_ssize_t *strides,
104 Py_ssize_t *suboffsets, Py_ssize_t *indices) {
105 char *pointer = (char*)buf;
107 for (i = 0; i < ndim; i++) {
108 pointer += strides[i] * indices[i];
109 if (suboffsets[i] >=0 ) {
110 pointer = *((char**)pointer) + suboffsets[i];
113 return (void*)pointer;
117 .. cmember:: Py_ssize_t itemsize
119 This is a storage for the itemsize (in bytes) of each element of the
120 shared memory. It is technically un-necessary as it can be obtained
121 using :cfunc:`PyBuffer_SizeFromFormat`, however an exporter may know
122 this information without parsing the format string and it is necessary
123 to know the itemsize for proper interpretation of striding. Therefore,
124 storing it is more convenient and faster.
126 .. cmember:: void *internal
128 This is for use internally by the exporting object. For example, this
129 might be re-cast as an integer by the exporter and used to store flags
130 about whether or not the shape, strides, and suboffsets arrays must be
131 freed when the buffer is released. The consumer should never alter this
135 Buffer related functions
136 ========================
139 .. cfunction:: int PyObject_CheckBuffer(PyObject *obj)
141 Return 1 if *obj* supports the buffer interface otherwise 0.
144 .. cfunction:: int PyObject_GetBuffer(PyObject *obj, Py_buffer *view, int flags)
146 Export *obj* into a :ctype:`Py_buffer`, *view*. These arguments must
147 never be *NULL*. The *flags* argument is a bit field indicating what
148 kind of buffer the caller is prepared to deal with and therefore what
149 kind of buffer the exporter is allowed to return. The buffer interface
150 allows for complicated memory sharing possibilities, but some caller may
151 not be able to handle all the complexity but may want to see if the
152 exporter will let them take a simpler view to its memory.
154 Some exporters may not be able to share memory in every possible way and
155 may need to raise errors to signal to some consumers that something is
156 just not possible. These errors should be a :exc:`BufferError` unless
157 there is another error that is actually causing the problem. The
158 exporter can use flags information to simplify how much of the
159 :cdata:`Py_buffer` structure is filled in with non-default values and/or
160 raise an error if the object can't support a simpler view of its memory.
162 0 is returned on success and -1 on error.
164 The following table gives possible values to the *flags* arguments.
166 +------------------------------+---------------------------------------------------+
167 | Flag | Description |
168 +==============================+===================================================+
169 | :cmacro:`PyBUF_SIMPLE` | This is the default flag state. The returned |
170 | | buffer may or may not have writable memory. The |
171 | | format of the data will be assumed to be unsigned |
172 | | bytes. This is a "stand-alone" flag constant. It |
173 | | never needs to be '|'d to the others. The exporter|
174 | | will raise an error if it cannot provide such a |
175 | | contiguous buffer of bytes. |
177 +------------------------------+---------------------------------------------------+
178 | :cmacro:`PyBUF_WRITABLE` | The returned buffer must be writable. If it is |
179 | | not writable, then raise an error. |
180 +------------------------------+---------------------------------------------------+
181 | :cmacro:`PyBUF_STRIDES` | This implies :cmacro:`PyBUF_ND`. The returned |
182 | | buffer must provide strides information (i.e. the |
183 | | strides cannot be NULL). This would be used when |
184 | | the consumer can handle strided, discontiguous |
185 | | arrays. Handling strides automatically assumes |
186 | | you can handle shape. The exporter can raise an |
187 | | error if a strided representation of the data is |
188 | | not possible (i.e. without the suboffsets). |
190 +------------------------------+---------------------------------------------------+
191 | :cmacro:`PyBUF_ND` | The returned buffer must provide shape |
192 | | information. The memory will be assumed C-style |
193 | | contiguous (last dimension varies the |
194 | | fastest). The exporter may raise an error if it |
195 | | cannot provide this kind of contiguous buffer. If |
196 | | this is not given then shape will be *NULL*. |
200 +------------------------------+---------------------------------------------------+
201 |:cmacro:`PyBUF_C_CONTIGUOUS` | These flags indicate that the contiguity returned |
202 |:cmacro:`PyBUF_F_CONTIGUOUS` | buffer must be respectively, C-contiguous (last |
203 |:cmacro:`PyBUF_ANY_CONTIGUOUS`| dimension varies the fastest), Fortran contiguous |
204 | | (first dimension varies the fastest) or either |
205 | | one. All of these flags imply |
206 | | :cmacro:`PyBUF_STRIDES` and guarantee that the |
207 | | strides buffer info structure will be filled in |
210 +------------------------------+---------------------------------------------------+
211 | :cmacro:`PyBUF_INDIRECT` | This flag indicates the returned buffer must have |
212 | | suboffsets information (which can be NULL if no |
213 | | suboffsets are needed). This can be used when |
214 | | the consumer can handle indirect array |
215 | | referencing implied by these suboffsets. This |
216 | | implies :cmacro:`PyBUF_STRIDES`. |
220 +------------------------------+---------------------------------------------------+
221 | :cmacro:`PyBUF_FORMAT` | The returned buffer must have true format |
222 | | information if this flag is provided. This would |
223 | | be used when the consumer is going to be checking |
224 | | for what 'kind' of data is actually stored. An |
225 | | exporter should always be able to provide this |
226 | | information if requested. If format is not |
227 | | explicitly requested then the format must be |
228 | | returned as *NULL* (which means ``'B'``, or |
229 | | unsigned bytes) |
230 +------------------------------+---------------------------------------------------+
231 | :cmacro:`PyBUF_STRIDED` | This is equivalent to ``(PyBUF_STRIDES | |
232 | | PyBUF_WRITABLE)``. |
233 +------------------------------+---------------------------------------------------+
234 | :cmacro:`PyBUF_STRIDED_RO` | This is equivalent to ``(PyBUF_STRIDES)``. |
236 +------------------------------+---------------------------------------------------+
237 | :cmacro:`PyBUF_RECORDS` | This is equivalent to ``(PyBUF_STRIDES | |
238 | | PyBUF_FORMAT | PyBUF_WRITABLE)``. |
239 +------------------------------+---------------------------------------------------+
240 | :cmacro:`PyBUF_RECORDS_RO` | This is equivalent to ``(PyBUF_STRIDES | |
241 | | PyBUF_FORMAT)``. |
242 +------------------------------+---------------------------------------------------+
243 | :cmacro:`PyBUF_FULL` | This is equivalent to ``(PyBUF_INDIRECT | |
244 | | PyBUF_FORMAT | PyBUF_WRITABLE)``. |
245 +------------------------------+---------------------------------------------------+
246 | :cmacro:`PyBUF_FULL_RO` | This is equivalent to ``(PyBUF_INDIRECT | |
247 | | PyBUF_FORMAT)``. |
248 +------------------------------+---------------------------------------------------+
249 | :cmacro:`PyBUF_CONTIG` | This is equivalent to ``(PyBUF_ND | |
250 | | PyBUF_WRITABLE)``. |
251 +------------------------------+---------------------------------------------------+
252 | :cmacro:`PyBUF_CONTIG_RO` | This is equivalent to ``(PyBUF_ND)``. |
254 +------------------------------+---------------------------------------------------+
257 .. cfunction:: void PyBuffer_Release(Py_buffer *view)
259 Release the buffer *view*. This should be called when the buffer
260 is no longer being used as it may free memory from it.
263 .. cfunction:: Py_ssize_t PyBuffer_SizeFromFormat(const char *)
265 Return the implied :cdata:`~Py_buffer.itemsize` from the struct-stype
266 :cdata:`~Py_buffer.format`.
269 .. cfunction:: int PyObject_CopyToObject(PyObject *obj, void *buf, Py_ssize_t len, char fortran)
271 Copy *len* bytes of data pointed to by the contiguous chunk of memory
272 pointed to by *buf* into the buffer exported by obj. The buffer must of
273 course be writable. Return 0 on success and return -1 and raise an error
274 on failure. If the object does not have a writable buffer, then an error
275 is raised. If *fortran* is ``'F'``, then if the object is
276 multi-dimensional, then the data will be copied into the array in
277 Fortran-style (first dimension varies the fastest). If *fortran* is
278 ``'C'``, then the data will be copied into the array in C-style (last
279 dimension varies the fastest). If *fortran* is ``'A'``, then it does not
280 matter and the copy will be made in whatever way is more efficient.
283 .. cfunction:: int PyBuffer_IsContiguous(Py_buffer *view, char fortran)
285 Return 1 if the memory defined by the *view* is C-style (*fortran* is
286 ``'C'``) or Fortran-style (*fortran* is ``'F'``) contiguous or either one
287 (*fortran* is ``'A'``). Return 0 otherwise.
290 .. cfunction:: void PyBuffer_FillContiguousStrides(int ndim, Py_ssize_t *shape, Py_ssize_t *strides, Py_ssize_t itemsize, char fortran)
292 Fill the *strides* array with byte-strides of a contiguous (C-style if
293 *fortran* is ``'C'`` or Fortran-style if *fortran* is ``'F'`` array of the
294 given shape with the given number of bytes per element.
297 .. cfunction:: int PyBuffer_FillInfo(Py_buffer *view, void *buf, Py_ssize_t len, int readonly, int infoflags)
299 Fill in a buffer-info structure, *view*, correctly for an exporter that can
300 only share a contiguous chunk of memory of "unsigned bytes" of the given
301 length. Return 0 on success and -1 (with raising an error) on error.
307 A memoryview object is an extended buffer object that could replace the buffer
308 object (but doesn't have to as that could be kept as a simple 1-d memoryview
309 object). It, unlike :ctype:`Py_buffer`, is a Python object (exposed as
310 :class:`memoryview` in :mod:`builtins`), so it can be used with Python code.
312 .. cfunction:: PyObject* PyMemoryView_FromObject(PyObject *obj)
314 Return a memoryview object from an object that defines the buffer interface.
317 Old-style buffer objects
318 ========================
320 .. index:: single: PyBufferProcs
322 More information on the old buffer interface is provided in the section
323 :ref:`buffer-structs`, under the description for :ctype:`PyBufferProcs`.
325 A "buffer object" is defined in the :file:`bufferobject.h` header (included by
326 :file:`Python.h`). These objects look very similar to string objects at the
327 Python programming level: they support slicing, indexing, concatenation, and
328 some other standard string operations. However, their data can come from one
329 of two sources: from a block of memory, or from another object which exports
330 the buffer interface.
332 Buffer objects are useful as a way to expose the data from another object's
333 buffer interface to the Python programmer. They can also be used as a
334 zero-copy slicing mechanism. Using their ability to reference a block of
335 memory, it is possible to expose any data to the Python programmer quite
336 easily. The memory could be a large, constant array in a C extension, it could
337 be a raw block of memory for manipulation before passing to an operating
338 system library, or it could be used to pass around structured data in its
339 native, in-memory format.
342 .. ctype:: PyBufferObject
344 This subtype of :ctype:`PyObject` represents a buffer object.
347 .. cvar:: PyTypeObject PyBuffer_Type
349 .. index:: single: BufferType (in module types)
351 The instance of :ctype:`PyTypeObject` which represents the Python buffer type;
352 it is the same object as ``buffer`` and ``types.BufferType`` in the Python
356 .. cvar:: int Py_END_OF_BUFFER
358 This constant may be passed as the *size* parameter to
359 :cfunc:`PyBuffer_FromObject` or :cfunc:`PyBuffer_FromReadWriteObject`. It
360 indicates that the new :ctype:`PyBufferObject` should refer to *base*
361 object from the specified *offset* to the end of its exported buffer.
362 Using this enables the caller to avoid querying the *base* object for its
366 .. cfunction:: int PyBuffer_Check(PyObject *p)
368 Return true if the argument has type :cdata:`PyBuffer_Type`.
371 .. cfunction:: PyObject* PyBuffer_FromObject(PyObject *base, Py_ssize_t offset, Py_ssize_t size)
373 Return a new read-only buffer object. This raises :exc:`TypeError` if
374 *base* doesn't support the read-only buffer protocol or doesn't provide
375 exactly one buffer segment, or it raises :exc:`ValueError` if *offset* is
376 less than zero. The buffer will hold a reference to the *base* object, and
377 the buffer's contents will refer to the *base* object's buffer interface,
378 starting as position *offset* and extending for *size* bytes. If *size* is
379 :const:`Py_END_OF_BUFFER`, then the new buffer's contents extend to the
380 length of the *base* object's exported buffer data.
382 .. versionchanged:: 2.5
383 This function used an :ctype:`int` type for *offset* and *size*. This
384 might require changes in your code for properly supporting 64-bit
388 .. cfunction:: PyObject* PyBuffer_FromReadWriteObject(PyObject *base, Py_ssize_t offset, Py_ssize_t size)
390 Return a new writable buffer object. Parameters and exceptions are similar
391 to those for :cfunc:`PyBuffer_FromObject`. If the *base* object does not
392 export the writeable buffer protocol, then :exc:`TypeError` is raised.
394 .. versionchanged:: 2.5
395 This function used an :ctype:`int` type for *offset* and *size*. This
396 might require changes in your code for properly supporting 64-bit
400 .. cfunction:: PyObject* PyBuffer_FromMemory(void *ptr, Py_ssize_t size)
402 Return a new read-only buffer object that reads from a specified location
403 in memory, with a specified size. The caller is responsible for ensuring
404 that the memory buffer, passed in as *ptr*, is not deallocated while the
405 returned buffer object exists. Raises :exc:`ValueError` if *size* is less
406 than zero. Note that :const:`Py_END_OF_BUFFER` may *not* be passed for the
407 *size* parameter; :exc:`ValueError` will be raised in that case.
409 .. versionchanged:: 2.5
410 This function used an :ctype:`int` type for *size*. This might require
411 changes in your code for properly supporting 64-bit systems.
414 .. cfunction:: PyObject* PyBuffer_FromReadWriteMemory(void *ptr, Py_ssize_t size)
416 Similar to :cfunc:`PyBuffer_FromMemory`, but the returned buffer is
419 .. versionchanged:: 2.5
420 This function used an :ctype:`int` type for *size*. This might require
421 changes in your code for properly supporting 64-bit systems.
424 .. cfunction:: PyObject* PyBuffer_New(Py_ssize_t size)
426 Return a new writable buffer object that maintains its own memory buffer of
427 *size* bytes. :exc:`ValueError` is returned if *size* is not zero or
428 positive. Note that the memory buffer (as returned by
429 :cfunc:`PyObject_AsWriteBuffer`) is not specifically aligned.
431 .. versionchanged:: 2.5
432 This function used an :ctype:`int` type for *size*. This might require
433 changes in your code for properly supporting 64-bit systems.