How to use Cython typed memoryviews to accept strings from Python?

问题

How can I write a Cython function that takes a byte string object (a normal string, a bytearray, or another object that follows the buffer protocol) as a typed memoryview?

According to the Unicode and Passing Strings Cython tutorial page, the following should work:

cpdef object printbuf(unsigned char[:] buf):
    chars = [chr(x) for x in buf]
    print repr(''.join(chars))

It does work for bytearrays and other writable buffers:

$ python -c 'import test; test.printbuf(bytearray("test\0ing"))'
'test\x00ing'

But it doesn't work for normal strings and other read-only buffer objects:

$ python -c 'import test; test.printbuf("test\0ing")'
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "test.pyx", line 1, in test.printbuf (test.c:1417)
  File "stringsource", line 614, in View.MemoryView.memoryview_cwrapper (test.c:6795)
  File "stringsource", line 321, in View.MemoryView.memoryview.__cinit__ (test.c:3341)
BufferError: Object is not writable.

Looking at the generated C code, Cython is always passing the PyBUF_WRITABLE flag to PyObject_GetBuffer(), which explains the exception.

I can manually get a view into the buffer object myself, but it's not as convenient:

from cpython.buffer cimport \
    PyBUF_SIMPLE, PyBUF_WRITABLE, \
    PyObject_CheckBuffer, PyObject_GetBuffer, PyBuffer_Release

cpdef object printbuf(object buf):
    if not PyObject_CheckBuffer(buf):
        raise TypeError("argument must follow the buffer protocol")
    cdef Py_buffer view
    PyObject_GetBuffer(buf, &view, PyBUF_SIMPLE)
    try:
        chars = [chr((<unsigned char *>view.buf)[i])
                 for i in range(view.len)]
        print repr(''.join(chars))
    finally:
        PyBuffer_Release(&view)

$ python -c 'import test; test.printbuf(bytearray("test\0ing"))'
'test\x00ing'
$ python -c 'import test; test.printbuf("test\0ing")'
'test\x00ing'

Am I doing something wrong, or does Cython not support coercing read-only buffer objects (such as normal strings) into typed memoryview objects?

回答1:

Despite the documentation suggesting otherwise, Cython (at least up to version 0.22) does not support coercing read-only buffer objects into typed memoryview objects. Cython always passes the PyBUF_WRITABLE flag to PyObject_GetBuffer(), even when it doesn't need write access. This causes read-only buffer objects to raise an exception.

I raised this issue on the Cython developer mailing list, and even included a (very rough) patch. I never got a reply, so I assume the Cython developers are not interested in fixing this bug.

回答2:

This issue was fixed in Cython 0.28, released 2018-03-13 (PR #1869). The changelog says:

The const modifier can be applied to memoryview declarations to allow read-only buffers as input.

There is also a new section in the documentation.

The example you gave will work in Cython 0.28 if you write your function like this:

cpdef object printbuf(const unsigned char[:] buf):
    chars = [chr(x) for x in buf]
    print repr(''.join(chars))

来源：https://stackoverflow.com/questions/28203670/how-to-use-cython-typed-memoryviews-to-accept-strings-from-python