[英]Three-way comparing strings in Python 3
Say you want to optimize a (byte) string compare intensive algorithm implemented in Python. 假设您要优化Python中实现的(字节)字符串比较密集算法。 Since a central code path contains this sequence of statements
由于中央代码路径包含此序列语句
if s < t:
# less than ...
elif t < s:
# greater than ...
else:
# equal ...
it would be great to optimize it to something like 将它优化为类似的东西会很棒
r = bytes_compare(s, t)
if r < 0:
# less than ...
elif r > 0:
# greater than ...
else:
# equal ...
where (the hypothetical) bytes_compare()
ideally would just call the three-way comparison C function memcmp()
which is usually quite well optimized. 其中(假设的)
bytes_compare()
理想情况下只调用三向比较 C函数memcmp()
,这通常是非常优化的。 This would reduce the number of string comparisons by half. 这会将字符串比较的数量减少一半。 A very feasible optimization unless the strings are ultra short.
一个非常可行的优化,除非字符串超短。
But how to get there with Python 3? 但是如何使用Python 3实现目标?
PS : PS :
Python 3 has removed the three way comparison global function cmp()
and the magic method __cmp__()
. Python 3删除了三路比较全局函数
cmp()
和魔术方法__cmp__()
。 And even with Python 2, the bytes
class doesn't had a __cmp__()
member. 即使使用Python 2,
bytes
类也没有__cmp__()
成员。
With the ctypes
package it's straight forward to call memcmp()
but the foreign function call overhead with ctypes
is prohibitively high. 使用
ctypes
包它可以直接调用memcmp()
但是使用ctypes
的外部函数调用开销非常高。
Python 3 (including 3.6) simply doesn't include any three-way comparison support for strings. Python 3(包括3.6)根本不包含对字符串的任何三向比较支持。 Although the internal implementation of the rich comparison operator
__lt__()
, __eq__()
etc. do call memcmp()
(in the C implementation of bytes
- cf. Objects/bytesobject.c
) there is no internal three-way comparison function that could be leveraged. 虽然富比较运算符
__lt__()
, __eq__()
__lt__()
等的内部实现确实调用了memcmp()
(在bytes
的C实现中 - 参见Objects/bytesobject.c
)但是没有内部的三向比较函数可以利用。
Thus, writing a C extension that provides a three-way comparison function by calling memcmp()
is the next best thing: 因此,编写一个通过调用
memcmp()
提供三向比较功能的C扩展是下一个最好的事情:
#include <Python.h>
static PyObject* cmp(PyObject* self, PyObject* args) {
PyObject *a = 0, *b = 0;
if (!PyArg_UnpackTuple(args, "cmp", 2, 2, &a, &b))
return 0;
if (!PyBytes_Check(a) || !PyBytes_Check(b)) {
PyErr_SetString(PyExc_TypeError, "only bytes() strings supported");
return 0;
}
Py_ssize_t n = PyBytes_GET_SIZE(a), m = PyBytes_GET_SIZE(b);
char *s = PyBytes_AsString(a), *t = PyBytes_AsString(b);
int r = 0;
if (n == m) {
r = memcmp(s, t, n);
} else if (n < m) {
r = memcmp(s, t, n);
if (!r)
r = -1;
} else {
r = memcmp(s, t, m);
if (!r)
r = 1;
}
return PyLong_FromLong(r);
}
static PyMethodDef bytes_util_methods[] = {
{ "cmp", cmp, METH_VARARGS, "Three way compare 2 bytes() objects." },
{0,0,0,0} };
static struct PyModuleDef bytes_util_def = {
PyModuleDef_HEAD_INIT, "bytes_util", "Three way comparison for strings.",
-1, bytes_util_methods };
PyMODINIT_FUNC PyInit_bytes_util(void) {
Py_Initialize();
return PyModule_Create(&bytes_util_def);
}
Compile with: 编译:
gcc -Wall -O3 -fPIC -shared bytes_util.c -o bytes_util.so -I/usr/include/python3.6m
Test: 测试:
>>> import bytes_util
>>> bytes_util.cmp(b'foo', b'barx')
265725
In contrast to calling memcmp
via the ctypes
package, this foreign call has the same overhead as the builtin bytes comparison operators (as they also are implemented as C extension with the standard Python version). 与通过
ctypes
包调用memcmp
相反,此外部调用具有与内置字节比较运算符相同的开销(因为它们也实现为带有标准Python版本的C扩展)。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.