两个多字节二进制数据变量之间的最快按位xor

Question

实现以下逻辑的最快方法是什么：

def xor(data, key):
    l = len(key)

    buff = ""
    for i in range(0, len(data)):
        buff += chr(ord(data[i]) ^ ord(key[i % l]))
    return buff

在我的情况下， key 是20字节sha1摘要，而数据是20字节和几（1,2,3）兆字节之间的一些二进制数据

更新

好的伙计们。这是一个快3.5倍的实现，它将数据和密钥分为4个，2个或1个字节（在我的情况下，大部分时间是4字节长整数）：

def xor(data, key):
    index = len(data) % 4
    size = (4, 1, 2, 1)[index]
    type = ('L', 'B', 'H', 'B')[index]
    key_len = len(key)/size
    data_len = len(data)/size
    key_fmt = "<" + str(key_len) + type;
    data_fmt = "<" + str(data_len) + type;

    key_list = struct.unpack(key_fmt, key)
    data_list = struct.unpack(data_fmt, data)

    result = []
    for i in range(data_len):
        result.append (key_list[i % key_len] ^ data_list[i])

    return struct.pack(data_fmt, *result)

使用大量内存，但就我而言，这并不是什么大问题。

任何想法如何提高速度几次？ ： - ）

最终更新：

好的，好的... numpy完成了这项工作。那真是太快了：

def xor(data, key):
    import numpy, math

    # key multiplication in order to match the data length
    key = (key*int(math.ceil(float(len(data))/float(len(key)))))[:len(data)]

    # Select the type size in bytes       
    for i in (8,4,2,1):
        if not len(data) % i: break

    if i == 8: dt = numpy.dtype('<Q8');
    elif i == 4: dt = numpy.dtype('<L4');
    elif i == 2: dt = numpy.dtype('<H2');
    else: dt = numpy.dtype('B');

    return numpy.bitwise_xor(numpy.fromstring(key, dtype=dt), numpy.fromstring(data, dtype=dt)).tostring()

初始实现需要8分50秒来处理一个千兆字节，第二个 - 大约2分30秒，最后一个只需要...... 0分10秒。

感谢任何贡献想法和代码的人。你们太棒了！

Answer 1

如果len(data)很大，您可能会发现xrange有显着改善。实际上，您可以使用enumerate完全替换范围函数。您也可以使用列表而不是附加到字符串中受益。

def xor(data, key):
    l = len(key)
    buff = []
    for idx, val in enumerate(data):
        buff.append(chr(ord(val) ^ ord(key[idx % l]))
    return ''.join(buff)

我还没有计时，但是我认为对于大量数据来说，我会更快一点。确保你衡量每一个变化。

如果分析表明对ord()的调用实际上需要时间，您可以提前在key中的所有值上运行它以在循环中保存调用。

您也可以将循环转换为普通的旧列表理解，但它会对可读性产生负面影响。无论如何，试一试，看看它是否更快。

Answer 2

未经测试

不知道它是否更快

假设len（mystring）是4的倍数

def xor(hash,mystring):
    s = struct.Struct("<L")

    v1 = memoryview(hash)

    tab1 = []
    for i in range(5):
        tab1.append(s.unpack_from(v1,i*4)

    v2 = memoryview(mystring)
    tab2=[]
    for i in range(len(mystring)/4):
        tab2.append(s.unpack_from(v1,i*4))
    tab3 = []
    try:
        for i in range(len(mystring)/20):
            for j in range(5):
               tab3.append(s.pack(tab1[j]^tab2[5*i+j]))
    expect IndexError:
        pass
    return "".join(tab3)

Answer 3

此代码应适用于Python 2.6+，包括Py3k。

from binascii import hexlify as _hexlify
from binascii import unhexlify as _unhexlify


def packl(lnum, padmultiple=0):
    """Packs the lnum (which must be convertable to a long) into a
    byte string 0 padded to a multiple of padmultiple bytes in size. 0
    means no padding whatsoever, so that packing 0 result in an empty
    string.  The resulting byte string is the big-endian two's
    complement representation of the passed in long."""

    if lnum == 0:
        return b'\0' * padmultiple
    elif lnum < 0:
        raise ValueError("Can only convert non-negative numbers.")
    s = hex(lnum)[2:]
    s = s.rstrip('L')
    if len(s) & 1:
        s = '0' + s
    s = _unhexlify(s)
    if (padmultiple != 1) and (padmultiple != 0):
        filled_so_far = len(s) % padmultiple
        if filled_so_far != 0:
            s = b'\0' * (padmultiple - filled_so_far) + s
    return s

def unpackl(bytestr):
    """Treats a byte string as a sequence of base 256 digits
    representing an unsigned integer in big-endian format and converts
    that representation into a Python integer."""

    return int(_hexlify(bytestr), 16) if len(bytestr) > 0 else 0

def xor(data, key):
    dlen = len(data)
    klen = len(key)
    if dlen > klen:
        key = key * ((dlen + klen - 1) // klen)
    key = key[:dlen]
    result = packl(unpackl(data) ^ unpackl(key))
    if len(result) < dlen:
         result = b'\0' * (dlen - len(result)) + result
    return result

这也适用于Python 2.7和3.x.它具有比前一个更简单的优点，同时在大约相同的时间内完成相同的操作：

from binascii import hexlify as _hexlify
from binascii import unhexlify as _unhexlify

def xor(data, key):
    dlen = len(data)
    klen = len(key)
    if dlen > klen:
        key = key * ((dlen + klen - 1) // klen)
    key = key[:dlen]
    data = int(_hexlify(data), 16)
    key = int(_hexlify(key), 16)
    result = (data ^ key) | (1 << (dlen * 8 + 7))
    # Python 2.6/2.7 only lines (comment out in Python 3.x)
    result = memoryview(hex(result))
    result = (result[4:-1] if result[-1] == 'L' else result[4:])
    # Python 3.x line
    #result = memoryview(hex(result).encode('ascii'))[4:]
    result = _unhexlify(result)
    return result

Answer 4

免责声明：正如其他海报所说，这是加密文件的一种非常糟糕的方式。 This article演示了如何简单地扭转这种混淆。

首先，一个简单的xor算法：

def xor(a,b,_xor8k=lambda a,b:struct.pack("!1000Q",*map(operator.xor,
                    struct.unpack("!1000Q",a),
                    struct.unpack("!1000Q",b)))
        ):
    if len(a)<=8000:
        s="!%iQ%iB"%divmod(len(a),8)
        return struct.pack(s,*map(operator.xor,
            struct.unpack(s,a),
            struct.unpack(s,b)))
    a=bytearray(a)
    for i in range(8000,len(a),8000):
        a[i-8000:i]=_xor8k(
            a[i-8000:i],
            b[i-8000:i])
    a[i:]=xor(a[i:],b[i:])
    return str(a)

其次是包装xor算法：

def xor_wrap(data,key,_struct8k=struct.Struct("!1000Q")):
    l=len(key)
    if len(data)>=8000:
        keyrpt=key*((7999+2*l)//l)#this buffer is accessed with whatever offset is required for a given 8k block
        #this expression should create at most 1 more copy of the key than is needed
        data=bytearray(data)
        offset=-8000#initial offset, set to zero on first loop iteration
        modulo=0#offset used to access the repeated key
        for offset in range(0,len(data)-7999,8000):
            _struct8k.pack_into(data,offset,*map(operator.xor,
                _struct8k.unpack_from(data,offset),
                _struct8k.unpack_from(keyrpt,modulo)))
            modulo+=8000;modulo%=l
        offset+=8000
    else:offset=0;keyrpt=key*(len(data)//l+1)#simple calculation guaranteed to be enough
    rest=len(data)-offset
    srest=struct.Struct("!%iQ%iB"%divmod(len(data)-offset,8))
    srest.pack_into(data,offset,*map(operator.xor,
        srest.unpack_from(data,offset),
        srest.unpack_from(keyrpt,modulo)))
    return data

Answer 5

这是一个只使用Python内置和标准模块的版本，看起来非常快 - 虽然我还没有将它与你的numpy版本进行比较。它使用了Python Cryptography Toolkit中的一些优化转换函数，如图所示。

# Part of the Python Cryptography Toolkit
# found here:
# http://www.google.com/codesearch/p?hl=en#Y_gnTlD6ECg/trunk/src/gdata/Crypto/Util/number.py&q=lang:python%20%22def%20long_to_bytes%22&sa=N&cd=1&ct=rc

# Improved conversion functions contributed by Barry Warsaw, after
# careful benchmarking

import struct

def long_to_bytes(n, blocksize=0):
    """long_to_bytes(n:long, blocksize:int) : string
    Convert a long integer to a byte string.

    If optional blocksize is given and greater than zero, pad the front of the
    byte string with binary zeros so that the length is a multiple of
    blocksize.
    """
    # after much testing, this algorithm was deemed to be the fastest
    s = ''
    n = long(n)
    pack = struct.pack
    while n > 0:
        s = pack('>I', n & 0xffffffffL) + s
        n = n >> 32
    # strip off leading zeros
    for i in range(len(s)):
        if s[i] != '\000':
            break
    else:
        # only happens when n == 0
        s = '\000'
        i = 0
    s = s[i:]
    # add back some pad bytes.  this could be done more efficiently w.r.t. the
    # de-padding being done above, but sigh...
    if blocksize > 0 and len(s) % blocksize:
        s = (blocksize - len(s) % blocksize) * '\000' + s
    return s

def bytes_to_long(s):
    """bytes_to_long(string) : long
    Convert a byte string to a long integer.

    This is (essentially) the inverse of long_to_bytes().
    """
    acc = 0L
    unpack = struct.unpack
    length = len(s)
    if length % 4:
        extra = (4 - length % 4)
        s = '\000' * extra + s
        length = length + extra
    for i in range(0, length, 4):
        acc = (acc << 32) + unpack('>I', s[i:i+4])[0]
    return acc


# original code in SO question
def xor_orig(data, key):
    l = len(key)

    buff = ""
    for i in range(0, len(data)):
        buff += chr(ord(data[i]) ^ ord(key[i % l]))
    return buff

# faster pure python version
def xor_new(data, key):
    import math

    # key multiplication in order to match the data length
    key = (key*int( math.ceil(float(len(data))/float(len(key)))))[:len(data)]

    # convert key and data to long integers
    key_as_long = bytes_to_long(key)
    data_as_long = bytes_to_long(data)

    # xor the numbers together and convert the result back to a byte string
    return long_to_bytes(data_as_long ^ key_as_long)

if __name__=='__main__':
    import random
    import sha

    TEST_DATA_LEN = 100000

    data = ''.join(chr(random.randint(0, 255)) for i in xrange(TEST_DATA_LEN))
    key = sha.new(data).digest()

    assert xor_new(data, key) == xor_orig(data, key)
    print 'done'

Answer 6

根据我在第一篇文章中的评论，如果坚持使用(CASE WHEN SUBSTRING_INDEX(SUBSTRING_INDEX(cal_participants.name, ' ', 1), ' ', -1) = 7591 THEN "Europe" END) as memberfirst 进行键填充和按位XOR运算，则可以相当快地处理大型文件，如下所示：

numpy

Answer 7

你拥有的东西已经达到了Python的速度。

如果您确实需要更快，请在C。

中实施