Python 和 C 的混合編程工具有很多，這里介紹 Python 標(biāo)準(zhǔn)庫自帶的 ctypes 模塊的使用方法。

初識(shí)

Python 的 ctypes 要使用 C 函數(shù)，需要先將 C 編譯成動(dòng)態(tài)鏈接庫的形式，即 Windows 下的 .dll 文件，或者 Linux 下的 .so 文件。先來看一下 ctypes 怎么使用 C 標(biāo)準(zhǔn)庫。

Windows 系統(tǒng)下的 C 標(biāo)準(zhǔn)庫動(dòng)態(tài)鏈接文件為 msvcrt.dll (一般在目錄 C:\Windows\System32 和 C:\Windows\SysWOW64 下分別對應(yīng) 32-bit 和 64-bit，使用時(shí)不用刻意區(qū)分，Python 會(huì)選擇合適的)

Linux 系統(tǒng)下的 C 標(biāo)準(zhǔn)庫動(dòng)態(tài)鏈接文件為 libc.so.6 (以 64-bit Ubuntu 系統(tǒng)為例， 在目錄 /lib/x86_64-linux-gnu 下)

例如，以下代碼片段導(dǎo)入 C 標(biāo)準(zhǔn)庫，并使用 printf 函數(shù)打印一條消息，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() =='Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    
libc.printf('Hello ctypes!\n')

另外導(dǎo)入dll文件，還有其它方式如下，詳細(xì)解釋請參閱 ctypes module 相關(guān)文檔，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
    #libc = windll.LoadLibrary('msvcrt.dll')  # Windows only
    #libc = oledll.LoadLibrary('msvcrt.dll')  # Windows only
    #libc = pydll.LoadLibrary('msvcrt.dll')
  
    #libc = CDLL('msvcrt.dll')
    #libc = WinDLL('msvcrt.dll')  # Windows only
    #libc = OleDLL('msvcrt.dll')  # Windows only
    #libc = PyDLL('msvcrt.dll')
elif platform.system() =='Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    #libc = pydll.LoadLibrary('libc.so.6')

    #libc = CDLL('libc.so.6')
    #libc = PyDLL('libc.so.6')
    
libc.printf('Hello ctypes!\n')

ctypes 數(shù)據(jù)類型

ctypes 作為 Python 和 C 聯(lián)系的橋梁，它定義了專有的數(shù)據(jù)類型來銜接這兩種編程語言。如下表，

注：Python 中的類型，除了 None，int， long， Byte String，Unicode String 作為 C 函數(shù)的參數(shù)默認(rèn)提供轉(zhuǎn)換外，其它類型都必須顯式提供轉(zhuǎn)換。

None：對應(yīng) C 中的 NULL

int, long： 對應(yīng) C 中的 int，具體實(shí)現(xiàn)時(shí)會(huì)根據(jù)機(jī)器字長自動(dòng)適配。

Byte String：對應(yīng) C 中的一個(gè)字符串指針 char * ，指向一塊內(nèi)存區(qū)域。

Unicode String ：對應(yīng) C 中一個(gè)寬字符串指針 wchar_t *，指向一塊內(nèi)存區(qū)域。

 例如，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')

libc.printf('%s\n', 'here!')        # here!
libc.printf('%S\n', u'there!')      # there!
libc.printf('%d\n', 42)             # 42
libc.printf('%ld\n', 60000000)      # 60000000

#libc.printf('%f\n', 3.14)          #>>> ctypes.ArgumentError
#libc.printf('%f\n', c_float(3.14)) #>>> dont know why 0.000000
libc.printf('%f\n', c_double(3.14)) # 3.140000

創(chuàng)建可變的 string buffer

Python 默認(rèn)的 string 是不可變的，所以不能傳遞 string 到一個(gè) C 函數(shù)去改變它的內(nèi)容，所以需要使用 create_string_buffer，對應(yīng) Unicode 字符串，要使用 create_unicode_buffer,

定義和用法如下，

>>> help(create_string_buffer)
Help on function create_string_buffer in module ctypes:

create_string_buffer(init, size=None)
    create_string_buffer(aString) -> character array
    create_string_buffer(anInteger) -> character array
    create_string_buffer(aString, anInteger) -> character array

from ctypes import *

p = create_string_buffer(5)  
print sizeof(p)     # 5
print repr(p.raw)   # '\x00\x00\x00\x00\x00'
p.raw = 'Hi'
print repr(p.raw)   # 'Hi\x00\x00\x00'
print repr(p.value) # 'Hi'

傳遞自定義參數(shù)類型到 C 函數(shù)

ctypes 允許你創(chuàng)建自定義參數(shù)類型，它會(huì)自動(dòng)去搜索自定義數(shù)據(jù)的 _as_parameter 屬性，將其作為 C 函數(shù)的參數(shù)，例如，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    
class Bottles(object):
    def __init__(self, number):
        self._as_parameter_ = number  # here only accept integer, string, unicode string
bottles = Bottles(42)
libc.printf('%d bottles of beer\n', bottles)

輸出，

42 bottles of beer

也可以為你的數(shù)據(jù)定義 _as_parameter 屬性，如下，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    
class Bottles(object):
    def __init__(self):
        self._as_parameter_ = None  # only accept integer, string, unicode string
     
    @property
    def aspram(self):
        return self._as_parameter_
     
    @aspram.setter
    def aspram(self, number):
        self._as_parameter_ = number
         
bottles = Bottles()
bottles.aspram = 63
libc.printf('%d bottles of beer\n', bottles)

輸出，

63 bottles of beer

指定 C 函數(shù)的參數(shù)類型

可以指定要調(diào)用 C 函數(shù)的參數(shù)類型，如果傳入?yún)?shù)不符合指定的類型，則 ctypes 會(huì)嘗試轉(zhuǎn)換，如果轉(zhuǎn)換不成功，則拋 ArgumentError，例如，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    
libc.printf.argtypes = [c_char_p, c_char_p, c_int, c_double]
    
libc.printf('String is "%s", Int is %d, Double is %f\n', 'Hi', 10, 2.2)
libc.printf('%s, %d, %f\n', 'X', 2, 3)
try:
    libc.printf("%d %d %d", 1, 2, 3)
except ArgumentError, e:
    print "*** ERROR: %s" % str(e)

輸出，

String is "Hi", Int is 10, Double is 2.200000 X, 2, 3.000000
*** ERROR: argument 2: 'exceptions.TypeError'>: wrong type

指定 C 函數(shù)的返回值類型

如果不指定 C 函數(shù)的返回值， ctypes 默認(rèn)返回 int 類型，如果要返回特定類型，需要指定返回類型 restype，

例如，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    
print '1->', libc.strchr('abcdefghij', c_char('d'))
     
libc.strchr.restype = c_char_p

print '2->', libc.strchr('abcdefghij', c_char('d'))
print '3->', libc.strchr('abcdefghij', 'd')  # Note, here C function strchr not know what 'd' mean, so rerurn None

libc.strchr.argtypes = [c_char_p, c_char]
print '4->', libc.strchr('abcdefghij', 'd')  # Note, here not use c_char('w')

輸出：

1-> 40291315
2-> defghij
3-> None
4-> defghij

按引用傳遞參數(shù)

有些情況下，需要 C 函數(shù)修改傳入的參數(shù)，或者參數(shù)過大不適合傳值，需要按引用傳遞，ctypes 提供關(guān)鍵字 byref() 處理這種情況，

例如，

import platform
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')
i = c_int()
f = c_float()
s = create_string_buffer('\000' * 32)
print 'i.val =', i.value
print 'f.val =', f.value
print 'repr(s.value) =', repr(s.value)
libc.sscanf('1 3.14 Hello', '%d %f %s', byref(i), byref(f), s)
print 'after, i.val =', i.value
print 'after, f.val =', f.value
print 'after, repr(s.value) =', repr(s.value)

輸出，

i.val = 0
f.val = 0.0
repr(s.value) = ''
after, i.val = 1
after, f.val = 3.1400001049
after, repr(s.value) = 'Hello'

使用結(jié)構(gòu)體

ctypes 支持結(jié)構(gòu)體的使用，從 Structure 類派生，數(shù)據(jù)放在 _fields_ 中，

例如，

class Point(Structure):
    _fields_ = [('x', c_int), ('y', c_int)]
        
point = Point(10, 20)
print 'point.x =', point.x 
print 'point.y =', point.y point = Point(y=5)
print 'after, point.x =', point.x 
print 'after, point.y =', point.y
print
class Rect(Structure):
    _fields_ = [('upperleft', Point), ('lowerright', Point)]
        
rc = Rect(point)
print 'rc.upperleft.x = %d, rc.upperleft.y = %d' % (rc.upperleft.x, rc.upperleft.y)
print 'rc.lowerright.x = %d, rc.lowerright.y = %d' % (rc.lowerright.x, rc.lowerright.y)
        
r = Rect(Point(1, 2), Point(3, 4))
print 'r.upperleft.x = %d, r.upperleft.y = %d' % (r.upperleft.x, r.upperleft.y)
print 'r.lowerright.x = %d, r.lowerright.y = %d' % (r.lowerright.x, r.lowerright.y)

輸出，

point.x = 10
point.y = 20
after, point.x = 0
after, point.y = 5

rc.upperleft.x = 0, rc.upperleft.y = 5
rc.lowerright.x = 0, rc.lowerright.y = 0
r.upperleft.x = 1, r.upperleft.y = 2
r.lowerright.x = 3, r.lowerright.y = 4

位域

ctypes 提供了對位域的支持，

例如，

class IntBit(Structure):
    _fields_ = [('x', c_uint, 2), ('y', c_uint, 4)]
    
IB = IntBit(1, 15)
print 'IB.x = %d' % IB.x 
print 'IB.y = %d' % IB.y
    
IB2 = IntBit(4, 16)
print '-> IB2.x = %d' % IB2.x 
print '-> IB2.y = %d' % IB2.y

輸出，

IB.x = 1
IB.y = 15
-> IB2.x = 0
-> IB2.y = 0

數(shù)組

ctypes 提供了對 Array 的支持，例如，

TenIntArrayType = c_int * 10
ta = TenIntArrayType(1, 2, 3, 4, 5, 6, 7, 8, 9, 10)
for item in ta:
    print item,
print
    
class PointEx(Structure):
    _fields_ = [('x', c_int), ('y', c_int)]
        
class MyStruct(Structure):
    _fields_ = [('a', c_int), ('b', c_int), ('pointex_array', PointEx * 4)]
        
ms = MyStruct(4, 5, ((1,1), (2,2), (3,3), (4,4)))
for item in ms.pointex_array:
    print '(item.x, item.y) = (%d, %d)' % (item.x, item.y) 
print

輸出，

1 2 3 4 5 6 7 8 9 10
(item.x, item.y) = (1, 1)
(item.x, item.y) = (2, 2)
(item.x, item.y) = (3, 3)
(item.x, item.y) = (4, 4)

指針

ctypes 使用關(guān)鍵字 pointer 提供了對指針的支持，注意指針解引用使用 [0]，例如，

i = c_int(42)
print 'before, i.value =', i.value
pi = pointer(i)
pi[0] = 57
print 'after, i.value =', i.value
    
# create NULL pointer, also can use this way, but recommend use 'pointer' not 'POINTER'
null_ptr = POINTER(c_int)()
print 'bool(null_ptr) =', bool(null_ptr)

輸出，

before, i.value = 42
after, i.value = 57
bool(null_ptr) = False

類型轉(zhuǎn)換

ctypes 提供了類型轉(zhuǎn)換方法 cast(),

例如，

class Bar(Structure):
    _fields_ = [('count', c_int), ('value', POINTER(c_int))]
        
bar = Bar()
bar.count = 3
bar.value = (c_int * 3)(1, 2, 3)
for idx in range(bar.count):
    print 'bar.value[%d] = %d' % (idx, bar.value[idx])
    
## use cast to convert 
try:
    bar.value = (c_byte * 4)()
except TypeError, e:
    print '*** ERROR: %s' % str(e)
        
bar.value = cast((c_byte * 4)(), POINTER(c_int))
for idx in range(4):
    print 'now, bar.value[%d] = %d' % (idx, bar.value[idx])
print

輸出，

bar.value[0] = 1
bar.value[1] = 2
bar.value[2] = 3
*** ERROR: incompatible types, c_byte_Array_4 instance instead of LP_c_long instance
now, bar.value[0] = 0
now, bar.value[1] = 0
now, bar.value[2] = 0
now, bar.value[3] = 0

回調(diào)函數(shù)

ctypes 通過 CFUNCTYPE 支持回調(diào)函數(shù)，

例如，

import platform 
from ctypes import *

if platform.system() == 'Windows':
    libc = cdll.LoadLibrary('msvcrt.dll')
elif platform.system() == 'Linux':
    libc = cdll.LoadLibrary('libc.so.6')
    
IntArray5 = c_int * 5
ia = IntArray5(5, 1, 7, 33, 99)
    
# CFUNCTYPE(restype, *argtypes, **kw)
CmpFuncType = CFUNCTYPE(c_int, POINTER(c_int), POINTER(c_int))
    
def py_cmp_func(a, b):
    if a[0] > b[0]:
        return 1
    elif a[0] < b[0]:
        return -1
    else:
        return 0
        
cmpfunc = CmpFuncType(py_cmp_func)
    
print 'before sort, the ia list is: ',
for item in ia:
    print item,
        
# void qsort (void* base, size_t num, size_t size, int (*compar)(const void*,const void*));
libc.qsort(ia, len(ia), sizeof(c_int), cmpfunc)
    
print '\nafter sort, the ia list is: ',
for item in ia:
    print item,
print

輸出，

before sort, the ia list is:  5 1 7 33 99 
after sort, the ia list is:  1 5 7 33 99

Resize Space

ctypes 提供了 resize 變量占用空間的方法 resize()，注意，只能增大，不能小于原始空間，

例如，

short_array = (c_short * 4)(1, 2, 3, 4)
print 'sizeof(short_array) =', sizeof(short_array)  # get 8, means short_array take 8-byte memory
print 'len(short_array) =', len(short_array)
print 'before resize, short_array is: ',
for idx in range(len(short_array)):
    print short_array[idx],
print
    
try:
    resize(short_array, 4)  # resize short_array to 4-byte, raise error, due to cannot resize smaller than original
except ValueError, e:
    print 'ERROR: %s' % str(e)
        
resize(short_array, 32)
print 'after succeed resize to 32-byte, now sizeof(short_array) =', sizeof(short_array)
print 'after succeed resize to 32-byte, now len(short_array) =', len(short_array)
print 'after reszie, short_array is: ',
for idx in range(len(short_array)):
    print short_array[idx],

輸出，

sizeof(short_array) = 8
len(short_array) = 4
before resize, short_array is:  1 2 3 4
ERROR: minimum size is 8
after succeed resize to 32-byte, now sizeof(short_array) = 32
after succeed resize to 32-byte, now len(short_array) = 4
after reszie, short_array is:  1 2 3 4

Other

class cell(Structure):
    pass
cell._fields_ = [('name', c_char_p), ('next', POINTER(cell))]
    
c1 = cell()
c2 = cell()
c1.name = 'foo' c2.name = 'bar' c1.next = pointer(c2) c2.next = pointer(c1) p = c1 for i in range(10): print p.name, p = p.next[0]

輸出，

foo bar foo bar foo bar foo bar foo bar

坑

ctypes 相對于其它工具，使用起來有點(diǎn)繁瑣，而且有很多坑，需要小心謹(jǐn)慎，

例如，

class POINT(Structure):
    _fields_ = [('x', c_int), ('y', c_int)]
        
class RECT(Structure):
    _fields_ = [('a', POINT), ('b', POINT)]
        
p1 = POINT(1, 2)
p2 = POINT(3, 4)
rc = RECT(p1, p2)
    
print 'rc.a.x =', rc.a.x
print 'rc.a.y =', rc.a.y
print 'rc.b.x =', rc.b.x
print 'rc.b.y =', rc.b.y
    
rc.a, rc.b = rc.b, rc.a
    
print 'after swap, bad result due to this is the pointer,'
print 'rc.a.x =', rc.a.x
print 'rc.a.y =', rc.a.y
print 'rc.b.x =', rc.b.x
print 'rc.b.y =', rc.b.y
print 

輸出，

rc.a.x = 1
rc.a.y = 2
rc.b.x = 3
rc.b.y = 4
after swap, bad result due to this is the pointer,
rc.a.x = 3
rc.a.y = 4
rc.b.x = 3
rc.b.y = 4

以 C 函數(shù)文件 needforspeed.c 為例，

//----------------------------------------------------------------------------
// Purpose: this c module is used to speed up the Python program, should be 
//          compiled into dll, and then load into Python module with ctypes
//          method.
//
// Compile Methods:
//
//    ======================
//    Windows: use MSVC, x64
//    ======================
//
//    C:\Program Files (x86)\Microsoft Visual Studio 12.0\VC\bin\amd64> cl /LD needforspeed.c /o nfs.dll
//
//    ======
//    Linux:
//    ======
//     
//    $ gcc -fPIC -shared needforspeed.c -o nfs.so
//----------------------------------------------------------------------------
#include 
// Windows need this compile direction for dll compilation, Linux no need 
#ifdef _MSC_VER
    #define DLL_EXPORT __declspec( dllexport ) 
#else
    #define DLL_EXPORT
#endif
DLL_EXPORT void hello_world(void) {
    printf("Hello world!\n");
}
DLL_EXPORT int mod(int m, int n) {
    return m % n;
}
DLL_EXPORT int get_array_elem(int arr[], int idx) {
    return arr[idx];
}
DLL_EXPORT int get_array2D_elem(int arr[][3], int row, int col) {
    return arr[row][col];
}

在 Windows 下編譯為 nfs.dll, 在 Linux 下編譯為 nfs.so，Python 中調(diào)用如下，

import platform 
from ctypes import *

if platform.system() == 'Windows':
    mylib = cdll.LoadLibrary('./nfs.dll')
elif platform.system() == 'Linux':
    mylib = cdll.LoadLibrary('./nfs.so')
 
mylib.hello_world()
print
 
mod_rtv = mylib.mod(c_int(10), c_int(4))
print 'mod_rtv(10 % 4) =', mod_rtv
print

#####################
# 1D array get elem #
#####################
 
IntArray10Type = c_int * 10
intArray = IntArray10Type()
for idx in range(10):
    intArray[idx] = idx**2
 
for idx in range(10):
    print 'intArray[%d] = %d' % (idx, mylib.get_array_elem(intArray, idx))
print

#####################
# 2D array get elem #
#####################

IntArray3Col = c_int * 3
IntArray3Row3Col = IntArray3Col * 3
arr2d = IntArray3Row3Col(IntArray3Col(1, 2, 3), IntArray3Col(8, 9, 4), IntArray3Col(7, 6, 5))

print 'arr2d is:'
for r in range(3):
    for c in range(3):
        print '%d ' % mylib.get_array2D_elem(arr2d, r, c),
    print

輸出，

Hello world!

mod_rtv(10 % 4) = 2

intArray[0] = 0
intArray[1] = 1
intArray[2] = 4
intArray[3] = 9
intArray[4] = 16
intArray[5] = 25
intArray[6] = 36
intArray[7] = 49
intArray[8] = 64
intArray[9] = 81

arr2d is:
1  2  3 
8  9  4 
7  6  5 

完。