Given that one wants to write an extension module that needs to pass a Python string to C library function. So, the question arises to properly handle Unicode. So, one of the main issues that arise is that existing C libraries won’t understand Python’s native representation of Unicode. Therefore, the main challenge is to convert the Python string into a form that can be more easily understood by C libraries.
To illustrate the solution – given below are two C functions that operate on string data and output it for debugging and experimentation.
Code #1 : Uses bytes provided in the form char *, int
void print_chars(char *s, int len) { int n = 0; while (n < len) { printf("%2x ", (unsigned char) s[n]); n++; } printf("\n"); } |
Code #2 : Uses wide characters in the form wchar_t *, int
void print_wchars(wchar_t *s, int len) { int n = 0; while (n < len) { printf("%x ", s[n]); n++; } printf("\n"); } |
Python strings need to be converted to a suitable byte encoding such as UTF-8 for the byte-oriented function print_chars(). The code given below a simple extension function solving the purpose.
Code #3 :
static PyObject *py_print_chars(PyObject *self, PyObject *args) { char *s; Py_ssize_t len; if (!PyArg_ParseTuple(args, "s#", &s, &len)) { return NULL; } print_chars(s, len); Py_RETURN_NONE; } |
For library functions that work with the machine native wchar_t type, C extension code can be written as –
Code #4 :
static PyObject * py_print_wchars(PyObject * self, PyObject * args) { wchar_t * s; Py_ssize_t len; if (! PyArg_ParseTuple(args, "u#", &s, &len)) { return NULL; } print_wchars(s, len); Py_RETURN_NONE; } |
Now the code below checks how the extension functions work.
It is to be observed the way the byte-oriented function print_chars() is receiving UTF-8 encoded data, whereas print_wchars() is receiving the Unicode code point values.
Code #5 :
s = 'Spicy Jalape\u00f1o'print (print_chars(s)) print ("\n", print_wchars(s)) |
Output :
53 70 69 63 79 20 4a 61 6c 61 70 65 c3 b1 6f 53 70 69 63 79 20 4a 61 6c 61 70 65 f1 6f
Let’s check the nature of C library that is been accessed. For many C libraries, it might make more sense to pass bytes instead of a string. Let’s use the conversion code given below to do so.
Code #6 :
static PyObject *py_print_chars(PyObject *self, PyObject *args) { char *s; Py_ssize_t len; // accepts bytes, bytearray, or other byte-like object if (!PyArg_ParseTuple(args, "y#", &s, &len)) { return NULL; } print_chars(s, len); Py_RETURN_NONE; } |
If still desire to pass strings, it is to be taken care that Python3 uses an adaptable string representation that is not entirely straightforward to map directly to C libraries using the standard types char * or wchar_t *. Thus, in order to present string data to C, some kind of conversion is almost always necessary. The s# and u# format codes to PyArg_ParseTuple() safely perform such conversions.
Whenever a conversion is made, a copy of the converted data is attached to the original string object so that it can be reused later as shown in the code below.
Code #7 :
import sys s = 'Spicy Jalape\u00f1o'print ("Size : ", sys.getsizeof(s)) print("\n", print_chars(s)) print ("\nSize : ", sys.getsizeof(s)) print ("\n", print_wchars(s)) print ("\nSize : ", sys.getsizeof(s)) |
Output :
Size : 87
53 70 69 63 79 20 4a 61 6c 61 70 65 c3 b1 6f
Size : 103
53 70 69 63 79 20 4a 61 6c 61 70 65 f1 6f
Size : 163
