acefile.py

#!/usr/bin/env python3
# vim: set list et ts=8 sts=4 sw=4 ft=python:

# acefile - read/test/extract ACE 1.0 and 2.0 archives in pure python
# Copyright (C) 2017-2019, Daniel Roethlisberger <daniel@roe.ch>
# All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# 1. Redistributions of source code must retain the above copyright
#    notice, this list of conditions, and the following disclaimer.
# 2. Redistributions in binary form must reproduce the above copyright
#    notice, this list of conditions and the following disclaimer in the
#    documentation and/or other materials provided with the distribution.
#
# THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
# IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
# OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
# IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
# INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
# NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
# THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

# NOTE:  The ACE archive format and ACE compression and decompression
# algorithms have been designed by Marcel Lemke.  The above copyright
# notice and license does not constitute a claim of intellectual property
# over ACE technology beyond the copyright of this python implementation.

"""
Read/test/extract ACE 1.0 and 2.0 archives in pure python.

This single-file, pure python 3, no-dependencies implementation is intended
to be used as a library, but also provides a stand-alone unace utility.
As mostly pure-python implementation, it is significantly slower than
native implementations, but more robust against vulnerabilities.

This implementation supports up to version 2.0 of the ACE archive format,
including the EXE, DELTA, PIC and SOUND modes of ACE 2.0, password protected
archives and multi-volume archives.  It does not support writing to archives.
It is an implementation from scratch, based on the 1998 document titled
"Technical information of the archiver ACE v1.2" by Marcel Lemke, using
unace 2.5 and WinAce 2.69 by Marcel Lemke as reference implementations.

For more information, API documentation, source code, packages and release
notifications, refer to:

- https://www.roe.ch/acefile
- https://apidoc.roe.ch/acefile
- https://github.com/droe/acefile
- https://pypi.python.org/pypi/acefile
- https://twitter.com/droethlisberger
"""

__version__     = '0.6.13-dev'
__author__      = 'Daniel Roethlisberger'
__email__       = 'daniel@roe.ch'
__copyright__   = 'Copyright 2017-2019, Daniel Roethlisberger'
__credits__     = ['Marcel Lemke']
__license__     = 'BSD'
__url__         = 'https://www.roe.ch/acefile'



import array
import builtins
import ctypes
import datetime
import errno
import io
import math
import os
import platform
import re
import stat
import struct
import sys
import zlib

try:
    import acebitstream
except:
    acebitstream = None



# Very basic debugging facility; if set to True, exceptions raised during
# testing of archives will be raised and a minimal set of state information
# will be printed to stderr.
DEBUG = False



# Arbitrarily chosen buffer size to use for buffered file operations that
# have no obvious natural block size.
FILE_BLOCKSIZE = 131072
assert FILE_BLOCKSIZE % 4 == 0



if platform.system() == 'Windows':
    # BOOL WINAPI SetFileAttributes(
    #   _In_ LPCTSTR lpFileName,
    #   _In_ DWORD   dwFileAttributes
    # );
    try:
        SetFileAttributes = ctypes.windll.kernel32.SetFileAttributesW
    except:
        SetFileAttributes = None
    # BOOL WINAPI SetFileSecurity(
    #  _In_ LPCTSTR              lpFileName,
    #  _In_ SECURITY_INFORMATION SecurityInformation,
    #  _In_ PSECURITY_DESCRIPTOR pSecurityDescriptor
    # );
    try:
        SetFileSecurity = ctypes.windll.advapi32.SetFileSecurityW
    except:
        SetFileSecurity = None
else:
    SetFileAttributes = None
    SetFileSecurity = None



def eprint(*args, **kwargs):
    """
    Print to stderr.
    """
    print(*args, file=sys.stderr, **kwargs)



# haklib.dt
def _dt_fromdos(dosdt):
    """
    Convert DOS format 32bit timestamp to datetime object.
    Timestamps with illegal values out of the allowed range are ignored and a
    datetime object representing 1980-01-01 00:00:00 is returned instead.
    https://msdn.microsoft.com/en-us/library/9kkf9tah.aspx

    >>> _dt_fromdos(0x4a5c48fd)
    datetime.datetime(2017, 2, 28, 9, 7, 58)
    >>> _dt_fromdos(0)
    datetime.datetime(1980, 1, 1, 0, 0)
    >>> _dt_fromdos(-1)
    datetime.datetime(1980, 1, 1, 0, 0)
    """
    try:
        return datetime.datetime(
                ((dosdt >> 25) & 0x7F) + 1980,
                 (dosdt >> 21) & 0x0F,
                 (dosdt >> 16) & 0x1F,
                 (dosdt >> 11) & 0x1F,
                 (dosdt >>  5) & 0x3F,
                ((dosdt      ) & 0x1F) * 2)
    except ValueError:
        return datetime.datetime(1980, 1, 1, 0, 0, 0)



# haklib.c
def c_div(q, d):
    """
    Arbitrary signed integer division with c behaviour.

    >>> (c_div(10, 3), c_div(-10, -3), c_div(-10, 3), c_div(10, -3))
    (3, 3, -3, -3)
    >>> c_div(-11, 0)
    Traceback (most recent call last):
        ...
    ZeroDivisionError
    """
    s = int(math.copysign(1, q) * math.copysign(1, d))
    return s * int(abs(q) / abs(d))

def c_schar(i):
    """
    Convert arbitrary integer to c signed char type range as if casted in c.

    >>> c_schar(0x12345678)
    120
    >>> (c_schar(-128), c_schar(-129), c_schar(127), c_schar(128))
    (-128, 127, 127, -128)
    """
    return ((i + 128) % 256) - 128

def c_uchar(i):
    """
    Convert arbitrary integer to c unsigned char type range as if casted in c.

    >>> c_uchar(0x12345678)
    120
    >>> (c_uchar(-123), c_uchar(-1), c_uchar(255), c_uchar(256))
    (133, 255, 255, 0)
    """
    return i & 0xFF

def c_rot32(i, n):
    """
    Rotate *i* left by *n* bits within the uint32 value range.

    >>> c_rot32(0xF0000000, 4)
    15
    >>> c_rot32(0xF0, -4)
    15
    """
    if n < 0:
        n = 32 + n
    return (((i << n) & 0xFFFFFFFF) | (i >> (32 - n)))

def c_add32(a, b):
    """
    Add *a* and *b* within the uint32 value range.

    >>> c_add32(0xFFFFFFFF, 1)
    0
    >>> c_add32(0xFFFFFFFF, 0xFFFFFFFF)
    4294967294
    """
    return (a + b) & 0xFFFFFFFF

def c_sum32(*args):
    """
    Add all elements of *args* within the uint32 value range.

    >>> c_sum32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF)
    4294967293
    """
    return sum(args) & 0xFFFFFFFF



def asciibox(msg, title=None, minwidth=None):
    """
    Returns message string *msg* wrapped in a plain ASCII box.
    If *title* is given, add *title* in the top horizontal bar.
    Lines will be padded to the longest out of *minwidth* characters, the
    length of the longest line, or the length of the title plus six.
    Caller is responsible for ensuring a sensible line length in *msg*.
    """
    out = []
    lines = msg.splitlines()
    width = 0
    for line in lines:
        width = max(width, len(line))
    if minwidth != None:
        width = max(width, minwidth)
    if title != None:
        width = max(width, len(title) + 6)
    ftr = "+" + ("-" * (width + 2)) + "+"
    if title != None:
        hdr = ("+--[ %s ]--" % title) + ("-" * (width - 6 - len(title))) + "+"
    else:
        hdr = ftr
    fmt = "| %%-%is |" % width
    out.append(hdr)
    for line in msg.splitlines():
        out.append(fmt % line)
    out.append(ftr)
    return '\n'.join(out)



class FileSegmentIO:
    """
    Seekable file-like object that wraps and reads from seekable file-like
    object and fakes EOF when a read would extend beyond a defined boundary.

    >>> FileSegmentIO(io.BytesIO(b'0123456789'), 3, 4).read()
    b'3456'
    >>> f = FileSegmentIO(io.BytesIO(b'0123456789'), 3, 4)
    >>> f.seekable()
    True
    >>> f.seek(0)
    0
    >>> f.read(1)
    b'3'
    >>> f.seek(1, os.SEEK_CUR)
    2
    >>> f.read(1)
    b'5'
    >>> f.seek(0, os.SEEK_END)
    4
    >>> f.read(1)
    b''
    >>> f.seek(-1, os.SEEK_SET)
    Traceback (most recent call last):
        ...
    OSError
    >>> f.seek(1, os.SEEK_END)
    Traceback (most recent call last):
        ...
    OSError
    """
    def __init__(self, f, base, size):
        assert f.seekable()
        self.__file = f
        self.__base = base
        self.__eof = base + size
        self.__file.seek(self.__base)

    def seekable(self):
        return True

    def _tell(self):
        """
        Returns the current absolute position in the file and asserts that it
        lies within the defined file segment.
        """
        pos = self.__file.tell()
        assert pos >= self.__base and pos <= self.__eof
        return pos

    def tell(self):
        return self._tell() - self.__base

    def seek(self, offset, whence=os.SEEK_SET):
        if whence == os.SEEK_SET:
            newpos = self.__base + offset
        elif whence == os.SEEK_CUR:
            newpos = self._tell() + offset
        elif whence == os.SEEK_END:
            newpos = self.__eof + offset
        if not (newpos >= self.__base and newpos <= self.__eof):
            raise OSError(errno.EINVAL, os.strerror(errno.EINVAL))
        self.__file.seek(newpos)
        return newpos - self.__base

    def read(self, n=None):
        pos = self._tell()
        if n == None:
            amount = self.__eof - pos
        else:
            amount = min(n, self.__eof - pos)
        if amount == 0:
            return b''
        return self.__file.read(amount)



class MultipleFilesIO:
    """
    Seekable file-like object that wraps and reads from multiple
    seekable lower-level file-like objects.

    >>> MultipleFilesIO((io.BytesIO(b'01234'), io.BytesIO(b'56789'))).read()
    b'0123456789'
    >>> f = MultipleFilesIO((io.BytesIO(b'01234'), io.BytesIO(b'56789')))
    >>> f.seekable()
    True
    >>> f.seek(0)
    0
    >>> f.read(1)
    b'0'
    >>> f.seek(1, os.SEEK_CUR)
    2
    >>> f.read(1)
    b'2'
    >>> f.seek(0, os.SEEK_END)
    10
    >>> f.read(1)
    b''
    >>> f.seek(-1, os.SEEK_SET)
    Traceback (most recent call last):
        ...
    OSError
    >>> f.seek(1, os.SEEK_END)
    Traceback (most recent call last):
        ...
    OSError
    """
    def __init__(self, files):
        assert len(files) > 0
        self.__files = files
        self.__sizes = []
        for f in files:
            f.seek(0, os.SEEK_END)
            self.__sizes.append(f.tell())
        self.__files[0].seek(0)
        self.__idx = 0
        self.__pos = 0
        self.__eof = sum(self.__sizes)

    def seekable(self):
        return True

    def tell(self):
        return self.__pos

    def seek(self, offset, whence=os.SEEK_SET):
        if whence == os.SEEK_SET:
            newpos = offset
        elif whence == os.SEEK_CUR:
            newpos = self.__pos + offset
        elif whence == os.SEEK_END:
            newpos = self.__eof + offset
        if not (newpos >= 0 and newpos <= self.__eof):
            raise OSError(errno.EINVAL, os.strerror(errno.EINVAL))
        idx = 0
        relpos = newpos
        while relpos > self.__sizes[idx]:
            relpos -= self.__sizes[idx]
            idx += 1
        self.__files[idx].seek(relpos)
        self.__idx = idx
        return newpos

    def read(self, n=None):
        if n == None:
            n = self.__eof - self.__pos
        out = []
        have_size = 0
        while have_size < n:
            if self.__idx >= len(self.__files):
                break
            chunk = self.__files[self.__idx].read(n - have_size)
            if len(chunk) == 0:
                self.__idx += 1
                if self.__idx < len(self.__files):
                    self.__files[self.__idx].seek(0)
                continue
            out.append(chunk)
            self.__pos += len(chunk)
            have_size += len(chunk)
        return b''.join(out)



class EncryptedFileIO:
    """
    Non-seekable file-like object that reads from a lower-level seekable
    file-like object, and transparently decrypts the data stream using a
    decryption engine.  The decryption engine is assumed to support a
    decrypt() method and a blocksize property.  The underlying file-like
    object is expected to contain a multiple of blocksize bytes, if not,
    CorruptedArchiveError is raised.

    >>> EncryptedFileIO(io.BytesIO(b'7'*16), AceBlowfish(b'123456789')).read()
    b'\\t_\\xd0a}\\x1dh\\xdd>h\\xe7VJ*_\\xea'
    >>> EncryptedFileIO(io.BytesIO(b'7'*17), AceBlowfish(b'123456789')).read()
    Traceback (most recent call last):
        ...
    CorruptedArchiveError
    >>> f = EncryptedFileIO(io.BytesIO(b'7'*16), AceBlowfish(b'123456789'))
    >>> f.seekable()
    False
    """
    def __init__(self, f, engine):
        self.__file = f
        self.__file.seek(0, os.SEEK_END)
        self.__eof = self.__file.tell()
        self.__file.seek(0)
        self.__engine = engine
        self.__buffer = b''

    def seekable(self):
        return False

    def read(self, n=None):
        if n == None:
            n = self.__eof - (self.__file.tell() - len(self.__buffer))
        if n < len(self.__buffer):
            rbuf = self.__buffer[:n]
            self.__buffer = self.__buffer[n:]
            return rbuf
        want_bytes = n - len(self.__buffer)
        read_bytes = want_bytes
        blocksize = self.__engine.blocksize
        if want_bytes % blocksize:
            read_bytes += blocksize - (want_bytes % blocksize)
        buf = self.__file.read(read_bytes)
        if len(buf) % blocksize:
            raise CorruptedArchiveError("Truncated ciphertext block")
        buf = self.__engine.decrypt(buf)
        rbuf = self.__buffer + buf[:n]
        self.__buffer = buf[n:]
        return rbuf



class AceBlowfish:
    """
    Decryption engine for ACE Blowfish.

    >>> bf = AceBlowfish(b'123456789')
    >>> bf.blocksize
    8
    >>> bf.decrypt(b'\\xFF'*8)
    b'\\xb7wF@5.er'
    >>> bf.decrypt(b'\\xC7'*8)
    b'eE\\x05\\xc4\\xa5\\x85)\\xbc'
    >>> bf.decrypt(b'123')
    Traceback (most recent call last):
        ...
    AssertionError
    """

    SHA1_A = 0x67452301
    SHA1_B = 0xefcdab89
    SHA1_C = 0x98badcfe
    SHA1_D = 0x10325476
    SHA1_E = 0xc3d2e1f0

    BF_P = (
        0x243F6A88, 0x85A308D3, 0x13198A2E, 0x03707344,
        0xA4093822, 0x299F31D0, 0x082EFA98, 0xEC4E6C89,
        0x452821E6, 0x38D01377, 0xBE5466CF, 0x34E90C6C,
        0xC0AC29B7, 0xC97C50DD, 0x3F84D5B5, 0xB5470917,
        0x9216D5D9, 0x8979FB1B)

    BF_S0 = (
        0xD1310BA6, 0x98DFB5AC, 0x2FFD72DB, 0xD01ADFB7,
        0xB8E1AFED, 0x6A267E96, 0xBA7C9045, 0xF12C7F99,
        0x24A19947, 0xB3916CF7, 0x0801F2E2, 0x858EFC16,
        0x636920D8, 0x71574E69, 0xA458FEA3, 0xF4933D7E,
        0x0D95748F, 0x728EB658, 0x718BCD58, 0x82154AEE,
        0x7B54A41D, 0xC25A59B5, 0x9C30D539, 0x2AF26013,
        0xC5D1B023, 0x286085F0, 0xCA417918, 0xB8DB38EF,
        0x8E79DCB0, 0x603A180E, 0x6C9E0E8B, 0xB01E8A3E,
        0xD71577C1, 0xBD314B27, 0x78AF2FDA, 0x55605C60,
        0xE65525F3, 0xAA55AB94, 0x57489862, 0x63E81440,
        0x55CA396A, 0x2AAB10B6, 0xB4CC5C34, 0x1141E8CE,
        0xA15486AF, 0x7C72E993, 0xB3EE1411, 0x636FBC2A,
        0x2DA9C55D, 0x741831F6, 0xCE5C3E16, 0x9B87901E,
        0xAFD6BA33, 0x6C24CF5C, 0x7A325381, 0x28958677,
        0x3B8F4898, 0x6B4BB9AF, 0xC4BFE81B, 0x66282193,
        0x61D809CC, 0xFB21A991, 0x487CAC60, 0x5DEC8032,
        0xEF845D5D, 0xE98575B1, 0xDC262302, 0xEB651B88,
        0x23893E81, 0xD396ACC5, 0x0F6D6FF3, 0x83F44239,
        0x2E0B4482, 0xA4842004, 0x69C8F04A, 0x9E1F9B5E,
        0x21C66842, 0xF6E96C9A, 0x670C9C61, 0xABD388F0,
        0x6A51A0D2, 0xD8542F68, 0x960FA728, 0xAB5133A3,
        0x6EEF0B6C, 0x137A3BE4, 0xBA3BF050, 0x7EFB2A98,
        0xA1F1651D, 0x39AF0176, 0x66CA593E, 0x82430E88,
        0x8CEE8619, 0x456F9FB4, 0x7D84A5C3, 0x3B8B5EBE,
        0xE06F75D8, 0x85C12073, 0x401A449F, 0x56C16AA6,
        0x4ED3AA62, 0x363F7706, 0x1BFEDF72, 0x429B023D,
        0x37D0D724, 0xD00A1248, 0xDB0FEAD3, 0x49F1C09B,
        0x075372C9, 0x80991B7B, 0x25D479D8, 0xF6E8DEF7,
        0xE3FE501A, 0xB6794C3B, 0x976CE0BD, 0x04C006BA,
        0xC1A94FB6, 0x409F60C4, 0x5E5C9EC2, 0x196A2463,
        0x68FB6FAF, 0x3E6C53B5, 0x1339B2EB, 0x3B52EC6F,
        0x6DFC511F, 0x9B30952C, 0xCC814544, 0xAF5EBD09,
        0xBEE3D004, 0xDE334AFD, 0x660F2807, 0x192E4BB3,
        0xC0CBA857, 0x45C8740F, 0xD20B5F39, 0xB9D3FBDB,
        0x5579C0BD, 0x1A60320A, 0xD6A100C6, 0x412C7279,
        0x679F25FE, 0xFB1FA3CC, 0x8EA5E9F8, 0xDB3222F8,
        0x3C7516DF, 0xFD616B15, 0x2F501EC8, 0xAD0552AB,
        0x323DB5FA, 0xFD238760, 0x53317B48, 0x3E00DF82,
        0x9E5C57BB, 0xCA6F8CA0, 0x1A87562E, 0xDF1769DB,
        0xD542A8F6, 0x287EFFC3, 0xAC6732C6, 0x8C4F5573,
        0x695B27B0, 0xBBCA58C8, 0xE1FFA35D, 0xB8F011A0,
        0x10FA3D98, 0xFD2183B8, 0x4AFCB56C, 0x2DD1D35B,
        0x9A53E479, 0xB6F84565, 0xD28E49BC, 0x4BFB9790,
        0xE1DDF2DA, 0xA4CB7E33, 0x62FB1341, 0xCEE4C6E8,
        0xEF20CADA, 0x36774C01, 0xD07E9EFE, 0x2BF11FB4,
        0x95DBDA4D, 0xAE909198, 0xEAAD8E71, 0x6B93D5A0,
        0xD08ED1D0, 0xAFC725E0, 0x8E3C5B2F, 0x8E7594B7,
        0x8FF6E2FB, 0xF2122B64, 0x8888B812, 0x900DF01C,
        0x4FAD5EA0, 0x688FC31C, 0xD1CFF191, 0xB3A8C1AD,
        0x2F2F2218, 0xBE0E1777, 0xEA752DFE, 0x8B021FA1,
        0xE5A0CC0F, 0xB56F74E8, 0x18ACF3D6, 0xCE89E299,
        0xB4A84FE0, 0xFD13E0B7, 0x7CC43B81, 0xD2ADA8D9,
        0x165FA266, 0x80957705, 0x93CC7314, 0x211A1477,
        0xE6AD2065, 0x77B5FA86, 0xC75442F5, 0xFB9D35CF,
        0xEBCDAF0C, 0x7B3E89A0, 0xD6411BD3, 0xAE1E7E49,
        0x00250E2D, 0x2071B35E, 0x226800BB, 0x57B8E0AF,
        0x2464369B, 0xF009B91E, 0x5563911D, 0x59DFA6AA,
        0x78C14389, 0xD95A537F, 0x207D5BA2, 0x02E5B9C5,
        0x83260376, 0x6295CFA9, 0x11C81968, 0x4E734A41,
        0xB3472DCA, 0x7B14A94A, 0x1B510052, 0x9A532915,
        0xD60F573F, 0xBC9BC6E4, 0x2B60A476, 0x81E67400,
        0x08BA6FB5, 0x571BE91F, 0xF296EC6B, 0x2A0DD915,
        0xB6636521, 0xE7B9F9B6, 0xFF34052E, 0xC5855664,
        0x53B02D5D, 0xA99F8FA1, 0x08BA4799, 0x6E85076A)

    BF_S1 = (
        0x4B7A70E9, 0xB5B32944, 0xDB75092E, 0xC4192623,
        0xAD6EA6B0, 0x49A7DF7D, 0x9CEE60B8, 0x8FEDB266,
        0xECAA8C71, 0x699A17FF, 0x5664526C, 0xC2B19EE1,
        0x193602A5, 0x75094C29, 0xA0591340, 0xE4183A3E,
        0x3F54989A, 0x5B429D65, 0x6B8FE4D6, 0x99F73FD6,
        0xA1D29C07, 0xEFE830F5, 0x4D2D38E6, 0xF0255DC1,
        0x4CDD2086, 0x8470EB26, 0x6382E9C6, 0x021ECC5E,
        0x09686B3F, 0x3EBAEFC9, 0x3C971814, 0x6B6A70A1,
        0x687F3584, 0x52A0E286, 0xB79C5305, 0xAA500737,
        0x3E07841C, 0x7FDEAE5C, 0x8E7D44EC, 0x5716F2B8,
        0xB03ADA37, 0xF0500C0D, 0xF01C1F04, 0x0200B3FF,
        0xAE0CF51A, 0x3CB574B2, 0x25837A58, 0xDC0921BD,
        0xD19113F9, 0x7CA92FF6, 0x94324773, 0x22F54701,
        0x3AE5E581, 0x37C2DADC, 0xC8B57634, 0x9AF3DDA7,
        0xA9446146, 0x0FD0030E, 0xECC8C73E, 0xA4751E41,
        0xE238CD99, 0x3BEA0E2F, 0x3280BBA1, 0x183EB331,
        0x4E548B38, 0x4F6DB908, 0x6F420D03, 0xF60A04BF,
        0x2CB81290, 0x24977C79, 0x5679B072, 0xBCAF89AF,
        0xDE9A771F, 0xD9930810, 0xB38BAE12, 0xDCCF3F2E,
        0x5512721F, 0x2E6B7124, 0x501ADDE6, 0x9F84CD87,
        0x7A584718, 0x7408DA17, 0xBC9F9ABC, 0xE94B7D8C,
        0xEC7AEC3A, 0xDB851DFA, 0x63094366, 0xC464C3D2,
        0xEF1C1847, 0x3215D908, 0xDD433B37, 0x24C2BA16,
        0x12A14D43, 0x2A65C451, 0x50940002, 0x133AE4DD,
        0x71DFF89E, 0x10314E55, 0x81AC77D6, 0x5F11199B,
        0x043556F1, 0xD7A3C76B, 0x3C11183B, 0x5924A509,
        0xF28FE6ED, 0x97F1FBFA, 0x9EBABF2C, 0x1E153C6E,
        0x86E34570, 0xEAE96FB1, 0x860E5E0A, 0x5A3E2AB3,
        0x771FE71C, 0x4E3D06FA, 0x2965DCB9, 0x99E71D0F,
        0x803E89D6, 0x5266C825, 0x2E4CC978, 0x9C10B36A,
        0xC6150EBA, 0x94E2EA78, 0xA5FC3C53, 0x1E0A2DF4,
        0xF2F74EA7, 0x361D2B3D, 0x1939260F, 0x19C27960,
        0x5223A708, 0xF71312B6, 0xEBADFE6E, 0xEAC31F66,
        0xE3BC4595, 0xA67BC883, 0xB17F37D1, 0x018CFF28,
        0xC332DDEF, 0xBE6C5AA5, 0x65582185, 0x68AB9802,
        0xEECEA50F, 0xDB2F953B, 0x2AEF7DAD, 0x5B6E2F84,
        0x1521B628, 0x29076170, 0xECDD4775, 0x619F1510,
        0x13CCA830, 0xEB61BD96, 0x0334FE1E, 0xAA0363CF,
        0xB5735C90, 0x4C70A239, 0xD59E9E0B, 0xCBAADE14,
        0xEECC86BC, 0x60622CA7, 0x9CAB5CAB, 0xB2F3846E,
        0x648B1EAF, 0x19BDF0CA, 0xA02369B9, 0x655ABB50,
        0x40685A32, 0x3C2AB4B3, 0x319EE9D5, 0xC021B8F7,
        0x9B540B19, 0x875FA099, 0x95F7997E, 0x623D7DA8,
        0xF837889A, 0x97E32D77, 0x11ED935F, 0x16681281,
        0x0E358829, 0xC7E61FD6, 0x96DEDFA1, 0x7858BA99,
        0x57F584A5, 0x1B227263, 0x9B83C3FF, 0x1AC24696,
        0xCDB30AEB, 0x532E3054, 0x8FD948E4, 0x6DBC3128,
        0x58EBF2EF, 0x34C6FFEA, 0xFE28ED61, 0xEE7C3C73,
        0x5D4A14D9, 0xE864B7E3, 0x42105D14, 0x203E13E0,
        0x45EEE2B6, 0xA3AAABEA, 0xDB6C4F15, 0xFACB4FD0,
        0xC742F442, 0xEF6ABBB5, 0x654F3B1D, 0x41CD2105,
        0xD81E799E, 0x86854DC7, 0xE44B476A, 0x3D816250,
        0xCF62A1F2, 0x5B8D2646, 0xFC8883A0, 0xC1C7B6A3,
        0x7F1524C3, 0x69CB7492, 0x47848A0B, 0x5692B285,
        0x095BBF00, 0xAD19489D, 0x1462B174, 0x23820E00,
        0x58428D2A, 0x0C55F5EA, 0x1DADF43E, 0x233F7061,
        0x3372F092, 0x8D937E41, 0xD65FECF1, 0x6C223BDB,
        0x7CDE3759, 0xCBEE7460, 0x4085F2A7, 0xCE77326E,
        0xA6078084, 0x19F8509E, 0xE8EFD855, 0x61D99735,
        0xA969A7AA, 0xC50C06C2, 0x5A04ABFC, 0x800BCADC,
        0x9E447A2E, 0xC3453484, 0xFDD56705, 0x0E1E9EC9,
        0xDB73DBD3, 0x105588CD, 0x675FDA79, 0xE3674340,
        0xC5C43465, 0x713E38D8, 0x3D28F89E, 0xF16DFF20,
        0x153E21E7, 0x8FB03D4A, 0xE6E39F2B, 0xDB83ADF7)

    BF_S2 = (
        0xE93D5A68, 0x948140F7, 0xF64C261C, 0x94692934,
        0x411520F7, 0x7602D4F7, 0xBCF46B2E, 0xD4A20068,
        0xD4082471, 0x3320F46A, 0x43B7D4B7, 0x500061AF,
        0x1E39F62E, 0x97244546, 0x14214F74, 0xBF8B8840,
        0x4D95FC1D, 0x96B591AF, 0x70F4DDD3, 0x66A02F45,
        0xBFBC09EC, 0x03BD9785, 0x7FAC6DD0, 0x31CB8504,
        0x96EB27B3, 0x55FD3941, 0xDA2547E6, 0xABCA0A9A,
        0x28507825, 0x530429F4, 0x0A2C86DA, 0xE9B66DFB,
        0x68DC1462, 0xD7486900, 0x680EC0A4, 0x27A18DEE,
        0x4F3FFEA2, 0xE887AD8C, 0xB58CE006, 0x7AF4D6B6,
        0xAACE1E7C, 0xD3375FEC, 0xCE78A399, 0x406B2A42,
        0x20FE9E35, 0xD9F385B9, 0xEE39D7AB, 0x3B124E8B,
        0x1DC9FAF7, 0x4B6D1856, 0x26A36631, 0xEAE397B2,
        0x3A6EFA74, 0xDD5B4332, 0x6841E7F7, 0xCA7820FB,
        0xFB0AF54E, 0xD8FEB397, 0x454056AC, 0xBA489527,
        0x55533A3A, 0x20838D87, 0xFE6BA9B7, 0xD096954B,
        0x55A867BC, 0xA1159A58, 0xCCA92963, 0x99E1DB33,
        0xA62A4A56, 0x3F3125F9, 0x5EF47E1C, 0x9029317C,
        0xFDF8E802, 0x04272F70, 0x80BB155C, 0x05282CE3,
        0x95C11548, 0xE4C66D22, 0x48C1133F, 0xC70F86DC,
        0x07F9C9EE, 0x41041F0F, 0x404779A4, 0x5D886E17,
        0x325F51EB, 0xD59BC0D1, 0xF2BCC18F, 0x41113564,
        0x257B7834, 0x602A9C60, 0xDFF8E8A3, 0x1F636C1B,
        0x0E12B4C2, 0x02E1329E, 0xAF664FD1, 0xCAD18115,
        0x6B2395E0, 0x333E92E1, 0x3B240B62, 0xEEBEB922,
        0x85B2A20E, 0xE6BA0D99, 0xDE720C8C, 0x2DA2F728,
        0xD0127845, 0x95B794FD, 0x647D0862, 0xE7CCF5F0,
        0x5449A36F, 0x877D48FA, 0xC39DFD27, 0xF33E8D1E,
        0x0A476341, 0x992EFF74, 0x3A6F6EAB, 0xF4F8FD37,
        0xA812DC60, 0xA1EBDDF8, 0x991BE14C, 0xDB6E6B0D,
        0xC67B5510, 0x6D672C37, 0x2765D43B, 0xDCD0E804,
        0xF1290DC7, 0xCC00FFA3, 0xB5390F92, 0x690FED0B,
        0x667B9FFB, 0xCEDB7D9C, 0xA091CF0B, 0xD9155EA3,
        0xBB132F88, 0x515BAD24, 0x7B9479BF, 0x763BD6EB,
        0x37392EB3, 0xCC115979, 0x8026E297, 0xF42E312D,
        0x6842ADA7, 0xC66A2B3B, 0x12754CCC, 0x782EF11C,
        0x6A124237, 0xB79251E7, 0x06A1BBE6, 0x4BFB6350,
        0x1A6B1018, 0x11CAEDFA, 0x3D25BDD8, 0xE2E1C3C9,
        0x44421659, 0x0A121386, 0xD90CEC6E, 0xD5ABEA2A,
        0x64AF674E, 0xDA86A85F, 0xBEBFE988, 0x64E4C3FE,
        0x9DBC8057, 0xF0F7C086, 0x60787BF8, 0x6003604D,
        0xD1FD8346, 0xF6381FB0, 0x7745AE04, 0xD736FCCC,
        0x83426B33, 0xF01EAB71, 0xB0804187, 0x3C005E5F,
        0x77A057BE, 0xBDE8AE24, 0x55464299, 0xBF582E61,
        0x4E58F48F, 0xF2DDFDA2, 0xF474EF38, 0x8789BDC2,
        0x5366F9C3, 0xC8B38E74, 0xB475F255, 0x46FCD9B9,
        0x7AEB2661, 0x8B1DDF84, 0x846A0E79, 0x915F95E2,
        0x466E598E, 0x20B45770, 0x8CD55591, 0xC902DE4C,
        0xB90BACE1, 0xBB8205D0, 0x11A86248, 0x7574A99E,
        0xB77F19B6, 0xE0A9DC09, 0x662D09A1, 0xC4324633,
        0xE85A1F02, 0x09F0BE8C, 0x4A99A025, 0x1D6EFE10,
        0x1AB93D1D, 0x0BA5A4DF, 0xA186F20F, 0x2868F169,
        0xDCB7DA83, 0x573906FE, 0xA1E2CE9B, 0x4FCD7F52,
        0x50115E01, 0xA70683FA, 0xA002B5C4, 0x0DE6D027,
        0x9AF88C27, 0x773F8641, 0xC3604C06, 0x61A806B5,
        0xF0177A28, 0xC0F586E0, 0x006058AA, 0x30DC7D62,
        0x11E69ED7, 0x2338EA63, 0x53C2DD94, 0xC2C21634,
        0xBBCBEE56, 0x90BCB6DE, 0xEBFC7DA1, 0xCE591D76,
        0x6F05E409, 0x4B7C0188, 0x39720A3D, 0x7C927C24,
        0x86E3725F, 0x724D9DB9, 0x1AC15BB4, 0xD39EB8FC,
        0xED545578, 0x08FCA5B5, 0xD83D7CD3, 0x4DAD0FC4,
        0x1E50EF5E, 0xB161E6F8, 0xA28514D9, 0x6C51133C,
        0x6FD5C7E7, 0x56E14EC4, 0x362ABFCE, 0xDDC6C837,
        0xD79A3234, 0x92638212, 0x670EFA8E, 0x406000E0)

    BF_S3 = (
        0x3A39CE37, 0xD3FAF5CF, 0xABC27737, 0x5AC52D1B,
        0x5CB0679E, 0x4FA33742, 0xD3822740, 0x99BC9BBE,
        0xD5118E9D, 0xBF0F7315, 0xD62D1C7E, 0xC700C47B,
        0xB78C1B6B, 0x21A19045, 0xB26EB1BE, 0x6A366EB4,
        0x5748AB2F, 0xBC946E79, 0xC6A376D2, 0x6549C2C8,
        0x530FF8EE, 0x468DDE7D, 0xD5730A1D, 0x4CD04DC6,
        0x2939BBDB, 0xA9BA4650, 0xAC9526E8, 0xBE5EE304,
        0xA1FAD5F0, 0x6A2D519A, 0x63EF8CE2, 0x9A86EE22,
        0xC089C2B8, 0x43242EF6, 0xA51E03AA, 0x9CF2D0A4,
        0x83C061BA, 0x9BE96A4D, 0x8FE51550, 0xBA645BD6,
        0x2826A2F9, 0xA73A3AE1, 0x4BA99586, 0xEF5562E9,
        0xC72FEFD3, 0xF752F7DA, 0x3F046F69, 0x77FA0A59,
        0x80E4A915, 0x87B08601, 0x9B09E6AD, 0x3B3EE593,
        0xE990FD5A, 0x9E34D797, 0x2CF0B7D9, 0x022B8B51,
        0x96D5AC3A, 0x017DA67D, 0xD1CF3ED6, 0x7C7D2D28,
        0x1F9F25CF, 0xADF2B89B, 0x5AD6B472, 0x5A88F54C,
        0xE029AC71, 0xE019A5E6, 0x47B0ACFD, 0xED93FA9B,
        0xE8D3C48D, 0x283B57CC, 0xF8D56629, 0x79132E28,
        0x785F0191, 0xED756055, 0xF7960E44, 0xE3D35E8C,
        0x15056DD4, 0x88F46DBA, 0x03A16125, 0x0564F0BD,
        0xC3EB9E15, 0x3C9057A2, 0x97271AEC, 0xA93A072A,
        0x1B3F6D9B, 0x1E6321F5, 0xF59C66FB, 0x26DCF319,
        0x7533D928, 0xB155FDF5, 0x03563482, 0x8ABA3CBB,
        0x28517711, 0xC20AD9F8, 0xABCC5167, 0xCCAD925F,
        0x4DE81751, 0x3830DC8E, 0x379D5862, 0x9320F991,
        0xEA7A90C2, 0xFB3E7BCE, 0x5121CE64, 0x774FBE32,
        0xA8B6E37E, 0xC3293D46, 0x48DE5369, 0x6413E680,
        0xA2AE0810, 0xDD6DB224, 0x69852DFD, 0x09072166,
        0xB39A460A, 0x6445C0DD, 0x586CDECF, 0x1C20C8AE,
        0x5BBEF7DD, 0x1B588D40, 0xCCD2017F, 0x6BB4E3BB,
        0xDDA26A7E, 0x3A59FF45, 0x3E350A44, 0xBCB4CDD5,
        0x72EACEA8, 0xFA6484BB, 0x8D6612AE, 0xBF3C6F47,
        0xD29BE463, 0x542F5D9E, 0xAEC2771B, 0xF64E6370,
        0x740E0D8D, 0xE75B1357, 0xF8721671, 0xAF537D5D,
        0x4040CB08, 0x4EB4E2CC, 0x34D2466A, 0x0115AF84,
        0xE1B00428, 0x95983A1D, 0x06B89FB4, 0xCE6EA048,
        0x6F3F3B82, 0x3520AB82, 0x011A1D4B, 0x277227F8,
        0x611560B1, 0xE7933FDC, 0xBB3A792B, 0x344525BD,
        0xA08839E1, 0x51CE794B, 0x2F32C9B7, 0xA01FBAC9,
        0xE01CC87E, 0xBCC7D1F6, 0xCF0111C3, 0xA1E8AAC7,
        0x1A908749, 0xD44FBD9A, 0xD0DADECB, 0xD50ADA38,
        0x0339C32A, 0xC6913667, 0x8DF9317C, 0xE0B12B4F,
        0xF79E59B7, 0x43F5BB3A, 0xF2D519FF, 0x27D9459C,
        0xBF97222C, 0x15E6FC2A, 0x0F91FC71, 0x9B941525,
        0xFAE59361, 0xCEB69CEB, 0xC2A86459, 0x12BAA8D1,
        0xB6C1075E, 0xE3056A0C, 0x10D25065, 0xCB03A442,
        0xE0EC6E0E, 0x1698DB3B, 0x4C98A0BE, 0x3278E964,
        0x9F1F9532, 0xE0D392DF, 0xD3A0342B, 0x8971F21E,
        0x1B0A7441, 0x4BA3348C, 0xC5BE7120, 0xC37632D8,
        0xDF359F8D, 0x9B992F2E, 0xE60B6F47, 0x0FE3F11D,
        0xE54CDA54, 0x1EDAD891, 0xCE6279CF, 0xCD3E7E6F,
        0x1618B166, 0xFD2C1D05, 0x848FD2C5, 0xF6FB2299,
        0xF523F357, 0xA6327623, 0x93A83531, 0x56CCCD02,
        0xACF08162, 0x5A75EBB5, 0x6E163697, 0x88D273CC,
        0xDE966292, 0x81B949D0, 0x4C50901B, 0x71C65614,
        0xE6C6C7BD, 0x327A140A, 0x45E1D006, 0xC3F27B9A,
        0xC9AA53FD, 0x62A80F00, 0xBB25BFE2, 0x35BDD2F6,
        0x71126905, 0xB2040222, 0xB6CBCF7C, 0xCD769C2B,
        0x53113EC0, 0x1640E3D3, 0x38ABBD60, 0x2547ADF0,
        0xBA38209C, 0xF746CE76, 0x77AFA1C5, 0x20756060,
        0x85CBFE4E, 0x8AE88DD8, 0x7AAAF9B0, 0x4CF9AA7E,
        0x1948C25C, 0x02FB8A8C, 0x01C36AE4, 0xD6EBE1F9,
        0x90D4F869, 0xA65CDEA0, 0x3F09252D, 0xC208E69F,
        0xB74E6132, 0xCE77E25B, 0x578FDFE3, 0x3AC372E6)

    def __init__(self, pwd):
        """
        Initialize decryption engine with a key derived from password *pwd*,
        which can be str or bytes.
        """
        if isinstance(pwd, str):
            pwd = pwd.encode('utf-8')
        self._bf_init(self._derive_key(pwd))

    def _derive_key(self, pwd):
        """
        Derive the decryption key from password bytes *pwd* using a single
        application of SHA-1 using non-standard padding.  The password is
        truncated to a maximum of 50 bytes before key derivation.

        >>> AceBlowfish._derive_key(None, b'123456789')
        (3071200156, 3325860325, 4058316933, 1308772094, 896611998)
        """
        if len(pwd) > 50:
            pwd = pwd[:50]
        buf = pwd + bytes([0x80] + [0] * (64 - len(pwd) - 5))
        state = []
        state.extend(struct.unpack('<15L', buf))
        state.append(len(pwd) << 3)
        for i in range(len(state), 80):
            state.append(state[i-16] ^ state[i-14] ^ state[i-8] ^ state[i-3])
        a = AceBlowfish.SHA1_A
        b = AceBlowfish.SHA1_B
        c = AceBlowfish.SHA1_C
        d = AceBlowfish.SHA1_D
        e = AceBlowfish.SHA1_E
        for i in range(20):
            a, b, c, d, e = \
                c_sum32(c_rot32(a, 5), ((b&c)|(~b&d)), e, state[i],
                        0x5a827999), a, c_rot32(b, 30), c, d
        for i in range(20, 40):
            a, b, c, d, e = \
                c_sum32(c_rot32(a, 5), (b^c^d), e, state[i],
                        0x6ed9eba1), a, c_rot32(b, 30), c, d
        for i in range(40, 60):
            a, b, c, d, e = \
                c_sum32(c_rot32(a, 5), ((b&c)|(b&d)|(c&d)), e, state[i],
                        0x8f1bbcdc), a, c_rot32(b, 30), c, d
        for i in range(60, 80):
            a, b, c, d, e = \
                c_sum32(c_rot32(a, 5), (b^c^d), e, state[i],
                        0xca62c1d6), a, c_rot32(b, 30), c, d
        a = c_add32(a, AceBlowfish.SHA1_A)
        b = c_add32(b, AceBlowfish.SHA1_B)
        c = c_add32(c, AceBlowfish.SHA1_C)
        d = c_add32(d, AceBlowfish.SHA1_D)
        e = c_add32(e, AceBlowfish.SHA1_E)
        return (a, b, c, d, e)

    def _bf_init(self, key):
        """
        Initialize blowfish state using 160-bit key *key* as list or tuple of
        integers.
        """
        self.__p = [self.BF_P[i] ^ key[i % len(key)] \
                for i in list(range(len(self.BF_P)))]
        self.__s = (list(self.BF_S0), list(self.BF_S1),
                    list(self.BF_S2), list(self.BF_S3))
        self.__lastcl = 0
        self.__lastcr = 0
        l = r = 0
        for i in range(0, 18, 2):
            l, r = self._bf_encrypt_block(l, r)
            self.__p[i] = l
            self.__p[i + 1] = r
        for i in range(0, 4):
            for j in range(0, 256, 2):
                l, r = self._bf_encrypt_block(l, r)
                self.__s[i][j] = l
                self.__s[i][j + 1] = r

    def _bf_func(self, x):
        """
        The blowfish round function operating on an integer.
        """
        h = c_add32(self.__s[0][x >> 24], self.__s[1][x >> 16 & 0xff])
        return c_add32((h ^ self.__s[2][x >> 8 & 0xff]), self.__s[3][x & 0xff])

    def _bf_encrypt_block(self, l, r):
        """
        Encrypt a single block consisting of integers *l* and *r*.
        """
        for i in range(0, 16, 2):
            l ^= self.__p[i]
            r ^= self._bf_func(l)
            r ^= self.__p[i+1]
            l ^= self._bf_func(r)
        l ^= self.__p[16]
        r ^= self.__p[17]
        return (r, l)

    def _bf_decrypt_block(self, l, r):
        """
        Decrypt a single block consisting of integers *l* and *r*.
        """
        for i in range(16, 0, -2):
            l ^= self.__p[i+1]
            r ^= self._bf_func(l)
            r ^= self.__p[i]
            l ^= self._bf_func(r)
        l ^= self.__p[1]
        r ^= self.__p[0]
        return (r, l)

    def decrypt(self, buf):
        """
        Decrypt a buffer of complete blocks, i.e. of length that is a multiple
        of the block size returned by the blocksize property.
        AceBlowfish uses Blowfish in CBC mode with an IV of all zeroes on the
        first call, and an IV of the last ciphertext block on subsequent calls.
        Does not remove any padding.
        """
        assert len(buf) % self.blocksize == 0
        out = []
        for i in range(0, len(buf), 8):
            cl, cr = struct.unpack('<LL', buf[i:i+8])
            pl, pr = self._bf_decrypt_block(cl, cr)
            pl ^= self.__lastcl
            pr ^= self.__lastcr
            self.__lastcl = cl
            self.__lastcr = cr
            out.append(struct.pack('<LL', pl, pr))
        return b''.join(out)

    @property
    def blocksize(self):
        """
        Return the block size of the decryption engine in bytes.
        The decrypt() method will only accept buffers containing a multiple of
        the block size of bytes.
        """
        return 8



class AceCRC32:
    """
    Calculate an ACE CRC-32 checksum.

    ACE CRC-32 uses the standard CRC-32 polynomial, bit ordering and
    initialization vector, but does not invert the resulting checksum.
    This implementation uses :meth:`zlib.crc32` with inverted state,
    inverted initialization vector and inverted output in order to
    construct ACE CRC-32 from standard CRC-32.

    >>> crc = AceCRC32()
    >>> crc += b"12345"
    >>> crc += b"6789"
    >>> crc.sum
    873187033
    >>> crc == 873187033
    True
    >>> crc2 = AceCRC32(b"123456789")
    >>> crc2.sum
    873187033
    >>> crc == crc2
    True
    """

    def __init__(self, buf=b''):
        """
        Initialize and add bytes in *buf* into checksum.
        """
        self.__state = 0
        if len(buf) > 0:
            self += buf

    def __iadd__(self, buf):
        """
        Adding a buffer of bytes into the checksum, updating the rolling
        checksum from all previously added buffers.
        """
        self.__state = zlib.crc32(buf, self.__state)
        return self

    def __eq__(self, other):
        """
        Compare the checksum to a fixed value or another ACE CRC32 object.
        """
        return self.sum == other

    def __format__(self, format_spec):
        """
        Format the checksum for printing.
        """
        return self.sum.__format__(format_spec)

    def __str__(self):
        """
        String representation of object is hex value of checksum.
        """
        return "0x%08x" % self.sum

    @property
    def sum(self):
        """
        The final checksum.
        """
        return self.__state ^ 0xFFFFFFFF

class AceCRC16(AceCRC32):
    """
    Calculate an ACE CRC-16 checksum, which is actually just the lower 16 bits
    of an ACE CRC-32.

    >>> crc = AceCRC16()
    >>> crc += b"12345"
    >>> crc += b"6789"
    >>> crc.sum
    50905
    >>> crc == 50905
    True
    """
    def __str__(self):
        """
        String representation of object is hex value of checksum.
        """
        return "0x%04x" % self.sum

    @property
    def sum(self):
        """
        The checksum.
        """
        return super().sum & 0xFFFF

def ace_crc32(buf):
    """
    Return the ACE CRC-32 checksum of the bytes in *buf*.

    >>> ace_crc32(b"123456789")
    873187033
    """
    return AceCRC32(buf).sum

def ace_crc16(buf):
    """
    Return the ACE CRC-16 checksum of the bytes in *buf*.

    >>> ace_crc16(b"123456789")
    50905
    """
    return AceCRC16(buf).sum



class BitStream:
    """
    Intel-endian 32bit-byte-swapped, MSB first bitstream, reading from an
    underlying file-like object that does not need to be seekable, but is
    expected to be a multiple of 4 in length.

    >>> bs = BitStream(io.BytesIO(b'01234567'))
    >>> bs.peek_bits(31)
    429463704
    >>> bs.read_bits(31)
    429463704
    >>> bs.skip_bits(3)
    >>> bs.read_bits(5)
    27
    >>> bs.read_golomb_rice(3)
    20
    >>> bs.read_golomb_rice(2, True)
    -2
    >>> bs.read_knownwidth_uint(10)
    618
    >>> bs.read_bits(7)
    52
    >>> bs.peek_bits(31)
    0
    >>> bs.read_bits(1)
    Traceback (most recent call last):
        ...
    EOFError
    >>> BitStream(io.BytesIO(b'012')).read_bits(31)
    Traceback (most recent call last):
        ...
    ValueError
    """

    @staticmethod
    def _getbits(value, start, length):
        """
        Return *length* bits from byte *value*, starting at position *start*.
        Behaviour is undefined for start < 0, length < 0 or start + length > 32.
        """
        #assert start >= 0 and length >= 0 and start + length <= 32
        mask = ((0xFFFFFFFF << (32 - length)) & 0xFFFFFFFF) >> start
        return (value & mask) >> (32 - length - start)

    def __init__(self, f):
        """
        Initialize BitStream reading from file-like object *f* until EOF.
        """
        self.__file = f
        self.__buf = array.array('I')
        self.__len = 0                  # in bits
        self.__pos = 0                  # in bits
        self._refill()

    def _refill(self):
        """
        Refill the internal buffer with data read from file.
        """
        tmpbuf = self.__file.read(FILE_BLOCKSIZE)
        if len(tmpbuf) == 0:
            raise EOFError("Cannot refill beyond EOF")
        if len(tmpbuf) % 4 != 0:
            raise ValueError("Truncated 32-bit word from file-like object")

        newbuf = self.__buf[-1:]
        for i in range(0, len(tmpbuf), 4):
            newbuf.append(struct.unpack('<L', tmpbuf[i:i+4])[0])
        if self.__pos > 0:
            self.__pos -= (self.__len - 32)
        self.__buf = newbuf
        self.__len = 32 * len(newbuf)

    def skip_bits(self, bits):
        """
        Skip *bits* bits in the stream.
        Raise EOFError when skipping beyond the end of the input file data.
        The pure-python implementation supports skipping arbitrarily many
        *bits* while the c implementation is limited to a maximum of 31.
        """
        if self.__pos + bits > self.__len:
            self._refill()
        self.__pos += bits

    def peek_bits(self, bits):
        """
        Peek at next *bits* bits in the stream without incrementing position.
        A maximum of 31 bits beyond the end of the input file data are
        guaranteed to be peekable; these bits are always unset.
        The pure-python implementation supports peeking arbitrarily many
        *bits* while the c implementation is limited to a maximum of 31.
        """
        if self.__pos + bits > self.__len:
            try:
                self._refill()
            except EOFError:
                if len(self.__buf) * 32 == self.__len:
                    self.__buf.append(0)
                if self.__pos + bits > self.__len + 31:
                    raise

        peeked = min(bits, 32 - (self.__pos % 32))
        res = self._getbits(self.__buf[self.__pos // 32],
                            self.__pos % 32, peeked)
        while bits - peeked >= 32:
            res <<= 32
            res += self.__buf[(self.__pos + peeked) // 32]
            peeked += 32
        if bits - peeked > 0:
            res <<= bits - peeked
            res += self._getbits(self.__buf[(self.__pos + peeked) // 32],
                                 0, bits - peeked)
        return res

    def read_bits(self, bits):
        """
        Read *bits* bits from bitstream and increment position accordingly.
        The pure-python implementation supports reading arbitrarily many
        *bits* while the c implementation is limited to a maximum of 31.
        """
        value = self.peek_bits(bits)
        self.skip_bits(bits)
        return value

    def read_golomb_rice(self, r_bits, signed=False):
        """
        Read a Golomb-Rice code with *r_bits* remainder bits and an arbitrary
        number of quotient bits from bitstream and return the represented
        value.  Iff *signed* is True, interpret the lowest order bit as sign
        bit and return a signed integer.
        """
        if r_bits == 0:
            value = 0
        else:
            assert r_bits > 0
            value = self.read_bits(r_bits)
        while self.read_bits(1) == 1:
            value += 1 << r_bits
        if signed == False:
            return value
        if value & 1:
            return - (value >> 1) - 1
        else:
            return value >> 1

    def read_knownwidth_uint(self, bits):
        """
        Read an unsigned integer with previously known bit width *bits* from
        stream.  The most significant bit is not encoded in the bit stream,
        because it is always 1.
        """
        if bits < 2:
            return bits
        bits -= 1
        return self.read_bits(bits) + (1 << bits)



if acebitstream != None:
    class BitStream_c(acebitstream.BitStream):
        read_golomb_rice = BitStream.read_golomb_rice
        read_knownwidth_uint = BitStream.read_knownwidth_uint

    BitStream_c.__doc__ = BitStream.__doc__
    BitStream = BitStream_c



class AceMode:
    """
    Represent and parse compression submode information from a bitstream.
    """
    @classmethod
    def read_from(cls, bs):
        mode = cls(bs.read_bits(8))
        if mode.mode == ACE.MODE_LZ77_DELTA:
            mode.delta_dist = bs.read_bits(8)
            mode.delta_len = bs.read_bits(17)
        elif mode.mode == ACE.MODE_LZ77_EXE:
            mode.exe_mode = bs.read_bits(8)
        if DEBUG:
            eprint(mode)
        return mode

    def __init__(self, mode):
        self.mode = mode

    def __str__(self):
        args = ''
        if self.mode == ACE.MODE_LZ77_DELTA:
            args = " delta_dist=%i delta_len=%i" % (self.delta_dist,
                                                    self.delta_len)
        elif self.mode == ACE.MODE_LZ77_EXE:
            args = " exe_mode=%i" % self.exe_mode
        return "mode %s(%i)%s" % (ACE.mode_str(self.mode), self.mode, args)



class Huffman:
    """
    Huffman decoder engine.
    """

    class Tree:
        """
        Huffman tree reconstructed from bitstream, internally represented by
        a table mapping (length-extended) codes to symbols and a table mapping
        symbols to bit widths.
        """
        def __init__(self, codes, widths, max_width):
            self.codes = codes
            self.widths = widths
            self.max_width = max_width

        def read_symbol(self, bs):
            """
            Read a single Huffman symbol from BitStream *bs* by peeking the
            maximum code length in bits from the bit stream, looking up the
            symbol and its width, and finally skipping the actual width of
            the code for the symbol in the bit stream.
            """
            maxwidth_code = bs.peek_bits(self.max_width)
            if maxwidth_code >= len(self.codes):
                # Not sure if we could prevent this from happening on malformed
                # input by improving how we choose max_width and codes; so far
                # only happens on malformed input.
                raise CorruptedArchiveError("maxwidth_code >= len(codes)")
            symbol = self.codes[maxwidth_code]
            bs.skip_bits(self.widths[symbol])
            return symbol


    WIDTHWIDTHBITS      = 3
    MAXWIDTHWIDTH       = (1 << WIDTHWIDTHBITS) - 1

    @staticmethod
    def _quicksort(keys, values):
        """
        In-place quicksort of lists *keys* and *values* in descending order of
        *keys*.  Python uses a stable sorting algorithm, while the
        reconstruction of the correct Huffman trees depends on the sorting
        being unstable in exactly the way of this quicksort implementation.

        >>> k, v = [1, 0, 0, 1, 2, 0, 0], list(range(7))
        >>> Huffman._quicksort(k, v)
        >>> (k, v)
        ([2, 1, 1, 0, 0, 0, 0], [4, 0, 3, 5, 6, 2, 1])
        """
        def _quicksort_subrange(left, right):
            def _list_swap(_list, a, b):
                """
                >>> a = list(range(9))
                >>> _list_swap(a, 3, 6)
                [0, 1, 2, 6, 4, 5, 3, 7, 8, 9]
                """
                _list[a], _list[b] = _list[b], _list[a]

            new_left = left
            new_right = right
            m = keys[right]
            while True:
                while keys[new_left] > m:
                    new_left += 1
                while keys[new_right] < m:
                    new_right -= 1
                if new_left <= new_right:
                    _list_swap(keys,   new_left, new_right)
                    _list_swap(values, new_left, new_right)
                    new_left += 1
                    new_right -= 1
                if new_left >= new_right:
                    break
            if left < new_right:
                if left < new_right - 1:
                    _quicksort_subrange(left, new_right)
                else:
                    if keys[left] < keys[new_right]:
                        _list_swap(keys,   left, new_right)
                        _list_swap(values, left, new_right)
            if right > new_left:
                if new_left < right - 1:
                    _quicksort_subrange(new_left, right)
                else:
                    if keys[new_left] < keys[right]:
                        _list_swap(keys,   new_left, right)
                        _list_swap(values, new_left, right)

        assert len(keys) == len(values)
        _quicksort_subrange(0, len(keys) - 1)

    @staticmethod
    def _make_tree(widths, max_width):
        """
        Calculate the list of Huffman codes corresponding to the symbols
        implicitly described by the list of *widths* and maximal width
        *max_width*, and return a Huffman.Tree object representing the
        resulting Huffman tree.
        """
        sorted_symbols  = list(range(len(widths)))
        sorted_widths   = list(widths)
        Huffman._quicksort(sorted_widths, sorted_symbols)

        used = 0
        while used < len(sorted_widths) and sorted_widths[used] != 0:
            used += 1

        if used < 2:
            widths[sorted_symbols[0]] = 1
            if used == 0:
                used += 1
        del sorted_symbols[used:]
        del sorted_widths[used:]

        codes = []
        max_codes = 1 << max_width
        for sym, wdt in zip(reversed(sorted_symbols), reversed(sorted_widths)):
            if wdt > max_width:
                raise CorruptedArchiveError("wdt > max_width")
            repeat = 1 << (max_width - wdt)
            codes.extend([sym] * repeat)
            if len(codes) > max_codes:
                raise CorruptedArchiveError("len(codes) > max_codes")

        return Huffman.Tree(codes, widths, max_width)

    @staticmethod
    def read_tree(bs, max_width, num_codes):
        """
        Read a Huffman tree consisting of codes and their widths from
        BitStream *bs*.  The caller specifies the maximum width of a single
        code *max_width* and the number of codes *num_codes*; these are
        required to reconstruct the Huffman tree from the widths stored in
        the bit stream.
        """
        num_widths = bs.read_bits(9) + 1
        if num_widths > num_codes + 1:
            num_widths = num_codes + 1
        lower_width = bs.read_bits(4)
        upper_width = bs.read_bits(4)

        width_widths = []
        width_num_widths = upper_width + 1
        for i in range(width_num_widths):
            width_widths.append(bs.read_bits(Huffman.WIDTHWIDTHBITS))
        width_tree = Huffman._make_tree(width_widths, Huffman.MAXWIDTHWIDTH)

        widths = []
        while len(widths) < num_widths:
            symbol = width_tree.read_symbol(bs)
            if symbol < upper_width:
                widths.append(symbol)
            else:
                length = bs.read_bits(4) + 4
                length = min(length, num_widths - len(widths))
                widths.extend([0] * length)

        if upper_width > 0:
            for i in range(1, len(widths)):
                widths[i] = (widths[i] + widths[i - 1]) % upper_width

        for i in range(len(widths)):
            if widths[i] > 0:
                widths[i] += lower_width

        return Huffman._make_tree(widths, max_width)



class LZ77:
    """
    ACE 1.0 and ACE 2.0 LZ77 mode decompression engine.

    Plain LZ77 compression over a Huffman-encoded symbol stream.
    """

    class SymbolReader:
        """
        Read blocks of Huffman-encoded LZ77 symbols.
        Two Huffman trees are used, one for the LZ77 symbols (main codes) and
        one for the length parameters (len codes).
        """

        def __init__(self):
            self.__syms_to_read = 0

        def _read_trees(self, bs):
            """
            Read the Huffman trees as well as the blocksize from BitStream
            *bs*; essentially this starts reading into a next block of symbols.
            """
            self.__main_tree = Huffman.read_tree(bs, LZ77.MAXCODEWIDTH,
                                                     LZ77.NUMMAINCODES)
            self.__len_tree  = Huffman.read_tree(bs, LZ77.MAXCODEWIDTH,
                                                     LZ77.NUMLENCODES)
            self.__syms_to_read = bs.read_bits(15)

        def read_main_symbol(self, bs):
            """
            Read a main symbol from BitStream *bs*.
            """
            if self.__syms_to_read == 0:
                self._read_trees(bs)
            self.__syms_to_read -= 1
            return self.__main_tree.read_symbol(bs)

        def read_len_symbol(self, bs):
            """
            Read a length symbol from BitStream *bs*.
            """
            return self.__len_tree.read_symbol(bs)


    class DistHist:
        """
        Distance value cache for storing the last SIZE used LZ77 distances.

        >>> dh = LZ77.DistHist()
        >>> dh.append(1);dh.append(2);dh.append(3);dh.append(4);dh.append(5)
        >>> dh.retrieve(2)
        3
        >>> dh.retrieve(0)
        3
        >>> dh.retrieve(1)
        5
        >>> dh.retrieve(1)
        3
        """
        SIZE = 4

        def __init__(self):
            self.__hist = [0] * self.SIZE

        def append(self, dist):
            self.__hist.pop(0)
            self.__hist.append(dist)

        def retrieve(self, offset):
            assert offset >= 0 and offset < self.SIZE
            dist = self.__hist.pop(self.SIZE - offset - 1)
            self.__hist.append(dist)
            return dist


    class Dictionary:
        """
        LZ77 dictionary.  Stores at least the last dictionary-size number of
        decompressed bytes and supports the LZ77 copy operation.  Also doubles
        as decompressed bytes buffer.  Consequently, the dictionary will grow
        as bytes are appended to it until copyout or copyin are called.

        >>> dic = LZ77.Dictionary(4, 8)
        >>> dic.append(1); dic.append(2); dic.extend((3,4))
        >>> dic.copy(4, 4)
        >>> dic.copyout(8)
        [1, 2, 3, 4, 1, 2, 3, 4]
        >>> dic.copy(9, 1)
        Traceback (most recent call last):
            ...
        CorruptedArchiveError
        """
        def __init__(self, minsize, maxsize):
            self.__dicdata = []
            self.__dicsize = minsize
            self.__maxsize = maxsize

        def set_size(self, dicsize):
            """
            Set expected dictionary size for next decompression run.
            """
            self.__dicsize = min(max(dicsize, self.__dicsize), self.__maxsize)

        def append(self, char):
            """
            Append output byte *char* to dictionary.
            """
            self.__dicdata.append(char)

        def extend(self, buf):
            """
            Append output bytes *buf* to dictionary.
            """
            self.__dicdata.extend(buf)

        def copy(self, dist, n):
            """
            Copy *n* previously produced output bytes to end of dictionary,
            starting from position *dist* away from the end.
            """
            source_pos = len(self.__dicdata) - dist
            if source_pos < 0:
                raise CorruptedArchiveError("LZ77 copy src out of bounds")
            # copy needs to be byte-wise for overlapping src and dst
            for i in range(source_pos, source_pos + n):
                self.__dicdata.append(self.__dicdata[i])

        def copyin(self, buf):
            """
            Copy output bytes produced by other decompression methods
            from *buf* into dictionary.  This operation may cause the
            dictionary to shrink to the indended dictionary size; make sure
            to copyout all needed output bytes before calling copyin.
            """
            self.extend(buf)
            self._truncate()

        def copyout(self, n):
            """
            Return copy of last *n* appended output bytes for handing them
            to the caller.  This operation may cause the dictionary to shrink
            to the intended dictionary size.
            """
            if n > 0:
                assert n <= len(self.__dicdata)
                chunk = self.__dicdata[-n:]
            else:
                chunk = []
            self._truncate()
            return chunk

        def _truncate(self):
            # only perform actual truncation when dictionary exceeds 4 times
            # it's supposed size in order to prevent excessive data copying
            if len(self.__dicdata) > 4 * self.__dicsize:
                self.__dicdata = self.__dicdata[-self.__dicsize:]


    #   0..255  character literals
    # 256..259  copy from dictionary, dist from dist history -1..-4
    # 260..282  copy from dictionary, dist 0..22 bits from bitstream
    # 283       type code
    MAXCODEWIDTH        = 11
    MAXLEN              = 259
    MAXDISTATLEN2       = 255
    MAXDISTATLEN3       = 8191
    MINDICBITS          = 10
    MAXDICBITS          = 22
    MINDICSIZE          = 1 << MINDICBITS
    MAXDICSIZE          = 1 << MAXDICBITS
    TYPECODE            = 260 + MAXDICBITS + 1
    NUMMAINCODES        = 260 + MAXDICBITS + 2
    NUMLENCODES         = 256 - 1

    def __init__(self):
        self.__dictionary = LZ77.Dictionary(LZ77.MINDICSIZE, LZ77.MAXDICSIZE)

    def reinit(self):
        """
        Reinitialize the LZ77 decompression engine.
        Reset all data dependent state to initial values.
        """
        self.__symreader = LZ77.SymbolReader()
        self.__disthist = LZ77.DistHist()
        self.__leftover = []

    def dic_setsize(self, dicsize):
        """
        Set the required dictionary size for the next LZ77 decompression run.
        """
        self.__dictionary.set_size(dicsize)

    def dic_register(self, buf):
        """
        Register bytes in *buf* produced by other decompression modes into
        the LZ77 dictionary.
        """
        self.__dictionary.copyin(buf)

    def read(self, bs, want_size):
        """
        Read a block of LZ77 compressed data from BitStream *bs*.
        Reading will stop when *want_size* output bytes can be provided,
        or when a block ends, i.e. when a mode instruction is found.
        Returns a tuple of the output byte-like and the mode instruction.
        """
        assert want_size > 0
        have_size = 0

        if len(self.__leftover) > 0:
            self.__dictionary.extend(self.__leftover)
            have_size += len(self.__leftover)
            self.__leftover = []

        next_mode = None
        while have_size < want_size:
            symbol = self.__symreader.read_main_symbol(bs)
            if symbol <= 255:
                self.__dictionary.append(symbol)
                have_size += 1
            elif symbol < LZ77.TYPECODE:
                if symbol <= 259:
                    copy_len = self.__symreader.read_len_symbol(bs)
                    offset = symbol & 0x03
                    copy_dist = self.__disthist.retrieve(offset)
                    if offset > 1:
                        copy_len += 3
                    else:
                        copy_len += 2
                else:
                    copy_dist = bs.read_knownwidth_uint(symbol - 260)
                    copy_len = self.__symreader.read_len_symbol(bs)
                    self.__disthist.append(copy_dist)
                    if copy_dist <= LZ77.MAXDISTATLEN2:
                        copy_len += 2
                    elif copy_dist <= LZ77.MAXDISTATLEN3:
                        copy_len += 3
                    else:
                        copy_len += 4
                copy_dist += 1
                if have_size + copy_len > want_size:
                    raise CorruptedArchiveError("LZ77 copy exceeds want_size")
                self.__dictionary.copy(copy_dist, copy_len)
                have_size += copy_len
            elif symbol == LZ77.TYPECODE:
                next_mode = AceMode.read_from(bs)
                break
            else:
                raise CorruptedArchiveError("LZ77 symbol > LZ77.TYPECODE")

        chunk = self.__dictionary.copyout(have_size)
        return (chunk, next_mode)



class Sound:
    """
    ACE 2.0 SOUND mode decompression engine.

    Multi-channel audio predictor over Huffman-encoding, resulting in a higher
    compression ratio for uncompressed mono/stereo 8/16 bit sound data.
    """

    class SymbolReader:
        """
        Read blocks of Huffman-encoded SOUND symbols.
        For each channel, three Huffman trees are used.
        """

        def __init__(self, num_models):
            self.__trees = [None] * num_models
            self.__syms_to_read = 0

        def _read_trees(self, bs):
            """
            Read the Huffman trees as well as the blocksize from BitStream
            *bs*; essentially this starts reading into a next block of symbols.
            """
            for i in range(len(self.__trees)):
                self.__trees[i] = Huffman.read_tree(bs, Sound.MAXCODEWIDTH,
                                                        Sound.NUMCODES)
            self.__syms_to_read = bs.read_bits(15)

        def read_symbol(self, bs, model):
            """
            Read a symbol from BitStream *bs* using the Huffman tree for model
            *model*.
            """
            if self.__syms_to_read == 0:
                self._read_trees(bs)
            self.__syms_to_read -= 1
            return self.__trees[model].read_symbol(bs)


    def classinit_sound_quantizer(cls):
        """
        Decorator that adds the static quantizer table to class *cls*.
        """
        cls._quantizer = [0] * 256
        for i in range(1, 129):
            # [-i] is equivalent to [256 - i]
            cls._quantizer[-i] = cls._quantizer[i] = i.bit_length()
        return cls

    @classinit_sound_quantizer
    class Channel:
        """
        Decompression parameters and methods for a single audio channel.
        """
        def __init__(self, symreader, channel_idx):
            """
            Initialize a channel with index *channel_idx*, using symbol
            reader *symreader* to fetch new symbols.
            """
            self.__symreader            = symreader
            self.__model_base_idx       = 3 * channel_idx
            self.__pred_dif_cnt         = [0] * 2
            self.__last_pred_dif_cnt    = [0] * 2
            self.__rar_dif_cnt          = [0] * 4
            self.__rar_coeff            = [0] * 4
            self.__rar_dif              = [0] * 9
            self.__byte_count           = 0
            self.__last_sample          = 0
            self.__last_delta           = 0
            self.__adapt_model_cnt      = 0
            self.__adapt_model_use      = 0
            self.__get_state            = 0
            self.__get_code             = 0

        def _get_symbol(self, bs):
            """
            Get next symbol from BitStream *bs*.
            """
            model = self.__get_state << 1
            if model == 0:
                model += self.__adapt_model_use
            model += self.__model_base_idx
            return self.__symreader.read_symbol(bs, model)

        def get(self, bs):
            """
            Get next sample, reading from BitStream *bs* if necessary.
            """
            if self.__get_state != 2:
                self.__get_code = self._get_symbol(bs)
                if self.__get_code == Sound.TYPECODE:
                    return AceMode.read_from(bs)

            if self.__get_state == 0:
                if self.__get_code >= Sound.RUNLENCODES:
                    value = self.__get_code - Sound.RUNLENCODES
                    self.__adapt_model_cnt = \
                        (self.__adapt_model_cnt * 7 >> 3) + value
                    if self.__adapt_model_cnt > 40:
                        self.__adapt_model_use = 1
                    else:
                        self.__adapt_model_use = 0
                else:
                    self.__get_state = 2
            elif self.__get_state == 1:
                value = self.__get_code
                self.__get_state = 0

            if self.__get_state == 2:
                if self.__get_code == 0:
                    self.__get_state = 1
                else:
                    self.__get_code -= 1
                value = 0

            if value & 1:
                return 255 - (value >> 1)
            else:
                return value >> 1

        def rar_predict(self):
            if self.__pred_dif_cnt[0] > self.__pred_dif_cnt[1]:
                return self.__last_sample
            else:
                return self._get_predicted_sample()

        def rar_adjust(self, sample):
            self.__byte_count += 1
            pred_sample = self._get_predicted_sample()
            pred_dif = c_schar(pred_sample - sample) << 3
            self.__rar_dif[0] += abs(pred_dif - self.__rar_dif_cnt[0])
            self.__rar_dif[1] += abs(pred_dif + self.__rar_dif_cnt[0])
            self.__rar_dif[2] += abs(pred_dif - self.__rar_dif_cnt[1])
            self.__rar_dif[3] += abs(pred_dif + self.__rar_dif_cnt[1])
            self.__rar_dif[4] += abs(pred_dif - self.__rar_dif_cnt[2])
            self.__rar_dif[5] += abs(pred_dif + self.__rar_dif_cnt[2])
            self.__rar_dif[6] += abs(pred_dif - self.__rar_dif_cnt[3])
            self.__rar_dif[7] += abs(pred_dif + self.__rar_dif_cnt[3])
            self.__rar_dif[8] += abs(pred_dif)

            self.__last_delta = c_schar(sample - self.__last_sample)
            self.__pred_dif_cnt[0] += self._quantizer[pred_dif >> 3]
            self.__pred_dif_cnt[1] += self._quantizer[self.__last_sample-sample]
            self.__last_sample = sample

            if self.__byte_count & 0x1F == 0:
                min_dif = 0xFFFF
                for i in reversed(range(9)):
                    if self.__rar_dif[i] <= min_dif:
                        min_dif = self.__rar_dif[i]
                        min_dif_pos = i
                    self.__rar_dif[i] = 0
                if min_dif_pos != 8:
                    i = min_dif_pos >> 1
                    if min_dif_pos & 1 == 0:
                        if self.__rar_coeff[i] >= -16:
                            self.__rar_coeff[i] -= 1
                    else:
                        if self.__rar_coeff[i] <= 16:
                            self.__rar_coeff[i] += 1
                if self.__byte_count & 0xFF == 0:
                    for i in range(2):
                        self.__pred_dif_cnt[i] -= self.__last_pred_dif_cnt[i]
                        self.__last_pred_dif_cnt[i] = self.__pred_dif_cnt[i]

            self.__rar_dif_cnt[3] = self.__rar_dif_cnt[2]
            self.__rar_dif_cnt[2] = self.__rar_dif_cnt[1]
            self.__rar_dif_cnt[1] = self.__last_delta - self.__rar_dif_cnt[0]
            self.__rar_dif_cnt[0] = self.__last_delta

        def _get_predicted_sample(self):
            return c_uchar((8 * self.__last_sample + \
                            self.__rar_coeff[0] * self.__rar_dif_cnt[0] + \
                            self.__rar_coeff[1] * self.__rar_dif_cnt[1] + \
                            self.__rar_coeff[2] * self.__rar_dif_cnt[2] + \
                            self.__rar_coeff[3] * self.__rar_dif_cnt[3]) >> 3)


    RUNLENCODES         = 32
    TYPECODE            = 256 + RUNLENCODES
    NUMCODES            = 256 + RUNLENCODES + 1
    MAXCODEWIDTH        = 10
    NUMCHANNELS         = (1, 2, 3, 3)
    USECHANNELS         = ((0, 0, 0, 0),
                           (0, 1, 0, 1),
                           (0, 1, 0, 2),
                           (1, 0, 2, 0))

    def __init__(self):
        self.__mode     = None
        self.__channels = None

    def reinit(self, mode):
        """
        Reinitialize the SOUND decompression engine.
        Reset all data dependent state to initial values.
        """
        self.__mode     = mode - ACE.MODE_SOUND_8
        num_channels    = Sound.NUMCHANNELS[self.__mode]
        num_models      = num_channels * 3
        sr              = Sound.SymbolReader(num_models)
        self.__channels = [self.Channel(sr, i) for i in range(num_channels)]

    def read(self, bs, want_size):
        """
        Read a block of SOUND compressed data from BitStream *bs*.
        Reading will stop when *want_size* output bytes can be provided,
        or when a block ends, i.e. when a mode instruction is found.
        Returns a tuple of the output byte-like and the mode instruction.
        """
        assert want_size > 0
        chunk = []
        next_mode = None
        for i in range(want_size & 0xFFFFFFFC):
            channel = Sound.USECHANNELS[self.__mode][i % 4]
            value = self.__channels[channel].get(bs)
            if isinstance(value, AceMode):
                next_mode = value
                break
            sample = c_uchar(value + self.__channels[channel].rar_predict())
            chunk.append(sample)
            self.__channels[channel].rar_adjust(c_schar(sample))
        return (chunk, next_mode)



class Pic:
    """
    ACE 2.0 PIC mode decompression engine.

    Two-dimensional pixel value predictor over Huffman encoding, resulting in a
    higher compression ratio for uncompressed picture data.
    """

    class ErrContext:
        """
        A prediction error context.
        """
        def __init__(self):
            self.used_counter = 0
            self.predictor_number = 0
            self.average_counter = 4
            self.error_counters = [0] * 4


    class ErrModel:
        """
        A prediction error model comprising of N error contexts.
        """
        N = 365

        def __init__(self):
            self.contexts = [Pic.ErrContext() for _ in range(Pic.ErrModel.N)]


    def classinit_pic_dif_bit_width(cls):
        """
        Decorator that adds the PIC dif_bit_width static table to *cls*.
        """
        cls._dif_bit_width = []
        for i in range(0, 128):
            cls._dif_bit_width.append((2 * i).bit_length())
        for i in range(-128, 0):
            cls._dif_bit_width.append((- 2 * i - 1).bit_length())
        return cls

    def classinit_pic_quantizers(cls):
        """
        Decorator that adds the PIC quantizer static tables to *cls*.
        """
        cls._quantizer   = []
        cls._quantizer9  = []
        cls._quantizer81 = []
        for i in range(-255, -20):
            cls._quantizer.append(-4)
        for i in range(-20, -6):
            cls._quantizer.append(-3)
        for i in range(-6, -2):
            cls._quantizer.append(-2)
        for i in range(-2, 0):
            cls._quantizer.append(-1)
        cls._quantizer.append(0)
        for i in range(1, 3):
            cls._quantizer.append(1)
        for i in range(3, 7):
            cls._quantizer.append(2)
        for i in range(7, 21):
            cls._quantizer.append(3)
        for i in range(21, 256):
            cls._quantizer.append(4)
        for q in cls._quantizer:
            cls._quantizer9.append(9 * q)
            cls._quantizer81.append(81 * q)
        return cls

    @classinit_pic_dif_bit_width
    @classinit_pic_quantizers
    class PixelDecoder:
        _producers = []

        @classmethod
        def register(cls, othercls):
            cls._producers.append(othercls)
            return othercls

        @classmethod
        def read_from(cls, bs):
            """
            Read a pixel decoder identifier from BitStream *bs* and return
            the appropriate PixelDecoder instance.
            """
            try:
                return cls._producers[bs.read_bits(2)]()
            except IndexError:
                raise CorruptedArchiveError("Unknown producer requested")

        def shift_pixels(self):
            """
            Shift pixels to the left to prepare for the next column.

                C A D
                B X
            """
            self._pixel_c = self._pixel_a
            self._pixel_a = self._pixel_d
            self._pixel_b = self._pixel_x

        def get_context(self):
            """
            Calculate the error context to use based on the differences
            between the neighbouring pixels D-A, A-C and C-B:

                C A D
                B X
            """
            ctx = self._quantizer81[255 + self._pixel_d - self._pixel_a] + \
                  self._quantizer9 [255 + self._pixel_a - self._pixel_c] + \
                  self._quantizer  [255 + self._pixel_c - self._pixel_b]
            return abs(ctx)

        def _predict(self, use_predictor):
            """
            With X being the current position, the predictors use the pixels
            above (A), to the left (B) and in the corner (C):

                C A D
                B X
            """
            if use_predictor == 0:
                return self._pixel_a
            elif use_predictor == 1:
                return self._pixel_b
            elif use_predictor == 2:
                return (self._pixel_a + self._pixel_b) >> 1
            elif use_predictor == 3:
                return c_uchar(self._pixel_a + self._pixel_b - self._pixel_c)

        def update_pixel_x(self, bs, context):
            """
            Read the next data point from BitStream *bs* and store the
            resulting pixel X based on ErrContext *context*.
            """
            context.used_counter += 1
            r = c_div(context.average_counter, context.used_counter)
            epsilon = bs.read_golomb_rice(r.bit_length(), signed=True)
            predicted = self._predict(context.predictor_number)
            self._pixel_x = c_uchar(predicted + epsilon)

            context.average_counter += abs(epsilon)
            if context.used_counter == 128:
                context.used_counter >>= 1
                context.average_counter >>= 1

            for i in range(len(context.error_counters)):
                context.error_counters[i] += \
                        self._dif_bit_width[self._pixel_x - self._predict(i)]
                if i == 0 or context.error_counters[i] < \
                             context.error_counters[best_predictor]:
                    best_predictor = i
            context.predictor_number = best_predictor

            if any([ec > 0x7F for ec in context.error_counters]):
                for i in range(len(context.error_counters)):
                    context.error_counters[i] >>= 1

    @PixelDecoder.register
    class PixelDecoder0(PixelDecoder):
        def __init__(self):
            self._pixel_a = 0
            self._pixel_b = 0
            self._pixel_c = 0
            self._pixel_x = 0

        def update_pixel_d(self, thisplane_d, refplane_d):
            self._pixel_d = thisplane_d

        def produce(self, refplane_x):
            return self._pixel_x

    class DifferentialPixelDecoder(PixelDecoder):
        def __init__(self):
            self._pixel_a = 128
            self._pixel_b = 128
            self._pixel_c = 128
            self._pixel_x = 128

    @PixelDecoder.register
    class PixelDecoder1(DifferentialPixelDecoder):
        def update_pixel_d(self, thisplane_d, refplane_d):
            self._pixel_d = c_uchar(128 + thisplane_d - refplane_d)

        def produce(self, refplane_x):
            return c_uchar(self._pixel_x + refplane_x - 128)

    @PixelDecoder.register
    class PixelDecoder2(DifferentialPixelDecoder):
        def update_pixel_d(self, thisplane_d, refplane_d):
            self._pixel_d = c_uchar(128 + thisplane_d - (refplane_d * 11 >> 4))

        def produce(self, refplane_x):
            return c_uchar(self._pixel_x + (refplane_x * 11 >> 4) - 128)


    def __init__(self):
        pass

    def reinit(self, bs):
        """
        Reinitialize the PIC decompression engine.
        Read width and planes from BitStream *bs* and reset all data dependent
        state to initial values.
        Note that width does not need to be a multiple of planes.
        """
        self.__width = bs.read_golomb_rice(12)
        self.__planes = bs.read_golomb_rice(2)
        self.__errmodel_plane0    = self.ErrModel()
        self.__errmodel_plane1toN = self.ErrModel()
        self.__prevrow = [0] * (self.__width + self.__planes)
        self.__leftover = []

    def _row(self, bs):
        """
        Decompress a row of pixels.
        """
        # NOTE
        # Some indices into row and self.__prevrow are outside of the
        # expected range 0..width, as indicated below.  Additionally, when
        # processing the first row, self.__prevrow is all zeroes.
        # Furthermore, self.__width is not necessarily a multiple of
        # self.__planes.
        row = [0] * (self.__width + self.__planes)
        for plane in range(self.__planes):
            if plane == 0:
                errmodel = self.__errmodel_plane0
                decoder = Pic.PixelDecoder0()
            else:
                errmodel = self.__errmodel_plane1toN
                decoder = Pic.PixelDecoder.read_from(bs)
            # plane-1 is -1 for first plane
            decoder.update_pixel_d(self.__prevrow[plane],
                                   self.__prevrow[plane - 1])

            for col in range(plane, self.__width, self.__planes):
                decoder.shift_pixels()
                # col+self.__planes is > width for last col in plane
                decoder.update_pixel_d(self.__prevrow[col + self.__planes],
                                       self.__prevrow[col + self.__planes - 1])
                context = errmodel.contexts[decoder.get_context()]
                decoder.update_pixel_x(bs, context)
                # col-1 is -1 for first col in first plane
                row[col] = decoder.produce(row[col - 1])

        self.__prevrow = row
        return row[:self.__width]

    def read(self, bs, want_size):
        """
        Read a block of PIC compressed data from BitStream *bs*.
        Reading will stop when *want_size* output bytes can be provided,
        or when a block ends, i.e. when a mode instruction is found.
        Returns a tuple of the output byte-like and the mode instruction.
        """
        assert want_size > 0
        chunk = []
        next_mode = None
        if len(self.__leftover) > 0:
            chunk.extend(self.__leftover)
            self.__leftover = []
        while len(chunk) < want_size:
            if bs.read_bits(1) == 0:
                next_mode = AceMode.read_from(bs)
                break
            data = self._row(bs)
            n = min(want_size - len(chunk), len(data))
            if n == len(data):
                chunk.extend(data)
            else:
                chunk.extend(data[0:n])
                self.__leftover = data[n:]
        return (chunk, next_mode)



class ACE:
    """
    Core decompression engine for ACE compression up to version 2.0.
    """
    MODE_LZ77               = 0     # LZ77
    MODE_LZ77_DELTA         = 1     # LZ77 after byte reordering
    MODE_LZ77_EXE           = 2     # LZ77 after patching JMP/CALL targets
    MODE_SOUND_8            = 3     # 8 bit sound compression
    MODE_SOUND_16           = 4     # 16 bit sound compression
    MODE_SOUND_32A          = 5     # 32 bit sound compression, variant 1
    MODE_SOUND_32B          = 6     # 32 bit sound compression, variant 2
    MODE_PIC                = 7     # picture compression
    MODE_STRINGS            = ('LZ77', 'LZ77_DELTA', 'LZ77_EXE',
                               'SOUND_8', 'SOUND_16', 'SOUND_32A', 'SOUND_32B',
                               'PIC')

    @staticmethod
    def mode_str(mode):
        try:
            return ACE.MODE_STRINGS[mode]
        except IndexError:
            return '?'

    @staticmethod
    def decompress_comment(buf):
        """
        Decompress an ACE MAIN or FILE comment from bytes *buf* and return the
        decompressed bytes.
        """
        bs = BitStream(io.BytesIO(buf))
        want_size = bs.read_bits(15)
        huff_tree = Huffman.read_tree(bs, LZ77.MAXCODEWIDTH, LZ77.NUMMAINCODES)
        comment = []
        htab = [0] * 511
        while len(comment) < want_size:
            if len(comment) > 1:
                hval = comment[-1] + comment[-2]
                source_pos = htab[hval]
                htab[hval] = len(comment)
            else:
                source_pos = 0
            code = huff_tree.read_symbol(bs)
            if code < 256:
                comment.append(code)
            else:
                for i in range(code - 256 + 2):
                    comment.append(comment[source_pos + i])
        return bytes(comment)

    def __init__(self):
        self.__lz77  = LZ77()
        self.__sound = Sound()
        self.__pic   = Pic()

    def decompress_stored(self, f, filesize, dicsize):
        """
        Decompress data compressed using the store method from file-like-object
        *f* containing compressed bytes that will be decompressed to *filesize*
        bytes.  Decompressed data will be yielded in blocks of undefined size
        upon availability.  Empty files will return without yielding anything.
        """
        self.__lz77.dic_setsize(dicsize)
        producedsize = 0
        while producedsize < filesize:
            wantsize = min(filesize - producedsize, FILE_BLOCKSIZE)
            outchunk = f.read(wantsize)
            if len(outchunk) == 0:
                raise CorruptedArchiveError("Truncated stored file")
            self.__lz77.dic_register(outchunk)
            yield outchunk
            producedsize += len(outchunk)

    def decompress_lz77(self, f, filesize, dicsize):
        """
        Decompress data compressed using the ACE 1.0 legacy LZ77 method from
        file-like-object *f* containing compressed bytes that will be
        decompressed to *filesize* bytes.  Decompressed data will be yielded
        in blocks of undefined size upon availability.
        """
        self.__lz77.dic_setsize(dicsize)
        self.__lz77.reinit()
        bs = BitStream(f)
        producedsize = 0
        while producedsize < filesize:
            outchunk, next_mode = self.__lz77.read(bs, filesize)
            if next_mode:
                raise CorruptedArchiveError("LZ77.TYPECODE in ACE 1.0 LZ77")
            yield bytes(outchunk)
            producedsize += len(outchunk)

    def decompress_blocked(self, f, filesize, dicsize):
        """
        Decompress data compressed using the ACE 2.0 blocked method from
        file-like-object *f* containing compressed bytes that will be
        decompressed to *filesize* bytes.  Decompressed data will be yielded
        in blocks of undefined size upon availability.
        """
        bs = BitStream(f)
        self.__lz77.dic_setsize(dicsize)
        self.__lz77.reinit()

        # LZ77_EXE
        exe_leftover = []

        # LZ77_DELTA
        last_delta = 0

        next_mode = None
        mode = AceMode(ACE.MODE_LZ77)

        producedsize = 0
        while producedsize < filesize:
            if next_mode != None:
                if mode.mode != next_mode.mode:
                    if next_mode.mode in (ACE.MODE_SOUND_8,
                                          ACE.MODE_SOUND_16,
                                          ACE.MODE_SOUND_32A,
                                          ACE.MODE_SOUND_32B):
                        self.__sound.reinit(next_mode.mode)
                    elif next_mode.mode == ACE.MODE_PIC:
                        self.__pic.reinit(bs)

                mode = next_mode
                next_mode = None

            outchunk = []
            if mode.mode == ACE.MODE_LZ77_DELTA:
                # Preprocessor that rearranges chunks of data and calculates
                # differences between byte values, resulting in a higher
                # LZ77 compression ratio for some inputs.
                delta = []
                while len(delta) < mode.delta_len:
                    chunk, nm = self.__lz77.read(bs,
                                                 mode.delta_len - len(delta))
                    if len(delta) == 0:
                        # avoid costly copy
                        delta = chunk
                    else:
                        delta.extend(chunk)
                    if nm != None:
                        if next_mode:
                            raise CorruptedArchiveError("DELTA clobbers mode")
                        next_mode = nm
                        if len(delta) == 0:
                            break
                if len(delta) == 0 and next_mode != None:
                    continue

                for i in range(len(delta)):
                    delta[i] = c_uchar(delta[i] + last_delta)
                    last_delta = delta[i]

                delta_plane = 0
                delta_plane_pos = 0
                delta_plane_size = mode.delta_len // mode.delta_dist
                while delta_plane_pos < delta_plane_size:
                    while delta_plane < mode.delta_len:
                        outchunk.append(delta[delta_plane + delta_plane_pos])
                        delta_plane += delta_plane_size
                    delta_plane = 0
                    delta_plane_pos += 1
                # end of ACE.MODE_LZ77_DELTA

            elif mode.mode in (ACE.MODE_LZ77, ACE.MODE_LZ77_EXE):
                if len(exe_leftover) > 0:
                    outchunk.extend(exe_leftover)
                    exe_leftover = []
                chunk, next_mode = self.__lz77.read(bs,
                        filesize - producedsize - len(outchunk))
                outchunk.extend(chunk)

                if mode.mode == ACE.MODE_LZ77_EXE:
                    # Preprocessor that adjusts target addresses of
                    # x86 JMP and CALL instructions in order to achieve a
                    # higher LZ77 compression ratio for executables.
                    it = iter(range(len(outchunk)))
                    for i in it:
                        if i + 4 >= len(outchunk):
                            break
                        if outchunk[i] == 0xE8:   # CALL rel16/rel32
                            pos = producedsize + i
                            if mode.exe_mode == 0:
                                # rel16
                                #assert i + 2 < len(outchunk)
                                rel16 = outchunk[i+1] + (outchunk[i+2] << 8)
                                rel16 = (rel16 - pos) & 0xFFFF
                                outchunk[i+1] =  rel16       & 0xFF
                                outchunk[i+2] = (rel16 >> 8) & 0xFF
                                next(it); next(it)
                            else:
                                # rel32
                                #assert i + 4 < len(outchunk)
                                rel32 =  outchunk[i+1]        + \
                                        (outchunk[i+2] <<  8) + \
                                        (outchunk[i+3] << 16) + \
                                        (outchunk[i+4] << 24)
                                rel32 = (rel32 - pos) & 0xFFFFFFFF
                                outchunk[i+1] =  rel32        & 0xFF
                                outchunk[i+2] = (rel32 >>  8) & 0xFF
                                outchunk[i+3] = (rel32 >> 16) & 0xFF
                                outchunk[i+4] = (rel32 >> 24) & 0xFF
                                next(it); next(it); next(it); next(it)
                        elif outchunk[i] == 0xE9: # JMP  rel16/rel32
                            pos = producedsize + i
                            # rel16
                            #assert i + 2 < len(outchunk)
                            rel16 = outchunk[i+1] + (outchunk[i+2] << 8)
                            rel16 = (rel16 - pos) & 0xFFFF
                            outchunk[i+1] =  rel16       & 0xFF
                            outchunk[i+2] = (rel16 >> 8) & 0xFF
                            next(it); next(it)
                    # store max 4 bytes for next loop; this can happen when
                    # changing between different exe modes after the opcode
                    # but before completing the machine instruction
                    for i in it:
                        #assert i + 4 >= len(outchunk)
                        if outchunk[i] == 0xE8 or outchunk[i] == 0xE9:
                            exe_leftover = outchunk[i:]
                            outchunk = outchunk[:i]
                    # end of ACE.MODE_LZ77_EXE
                # end of ACE.MODE_LZ77 or ACE.MODE_LZ77_EXE

            elif mode.mode in (ACE.MODE_SOUND_8,   ACE.MODE_SOUND_16,
                               ACE.MODE_SOUND_32A, ACE.MODE_SOUND_32B):
                outchunk, next_mode = self.__sound.read(bs,
                                                        filesize - producedsize)
                self.__lz77.dic_register(outchunk)
                # end of ACE.MODE_SOUND_*

            elif mode.mode == ACE.MODE_PIC:
                outchunk, next_mode = self.__pic.read(bs,
                                                      filesize - producedsize)
                self.__lz77.dic_register(outchunk)
                # end of ACE.MODE_PIC

            else:
                raise CorruptedArchiveError("unknown mode: %s" % mode)

            yield bytes(outchunk)
            producedsize += len(outchunk)
            # end of block loop
        return producedsize



class Header:
    """
    Base class for all ACE file format headers.
    Header classes are dumb by design and only serve as fancy structs.
    """
    MAGIC               = b'**ACE**'

    TYPE_MAIN           = 0
    TYPE_FILE32         = 1
    TYPE_RECOVERY32     = 2
    TYPE_FILE64         = 3
    TYPE_RECOVERY64A    = 4
    TYPE_RECOVERY64B    = 5
    TYPE_STRINGS        = ('MAIN', 'FILE32', 'RECOVERY32',
                           'FILE64', 'RECOVERY64A', 'RECOVERY64B')

    FLAG_ADDSIZE        = 1 <<  0   # 1 iff addsize field present           MFR
    FLAG_COMMENT        = 1 <<  1   # 1 iff comment present                 MF-
    FLAG_64BIT          = 1 <<  2   # 1 iff 64bit addsize field             -FR
    FLAG_V20FORMAT      = 1 <<  8   # 1 iff ACE 2.0 format                  M--
    FLAG_SFX            = 1 <<  9   # 1 iff self extracting archive         M--
    FLAG_LIMITSFXJR     = 1 << 10   # 1 iff dict size limited to 256K       M--
    FLAG_NTSECURITY     = 1 << 10   # 1 iff NTFS security data present      -F-
    FLAG_MULTIVOLUME    = 1 << 11   # 1 iff archive has multiple volumes    M--
    FLAG_ADVERT         = 1 << 12   # 1 iff advert string present           M--
    FLAG_CONTPREV       = 1 << 12   # 1 iff continued from previous volume  -F-
    FLAG_RECOVERY       = 1 << 13   # 1 iff recovery record present         M--
    FLAG_CONTNEXT       = 1 << 13   # 1 iff continued in next volume        -F-
    FLAG_LOCKED         = 1 << 14   # 1 iff archive is locked               M--
    FLAG_PASSWORD       = 1 << 14   # 1 iff password encrypted              -F-
    FLAG_SOLID          = 1 << 15   # 1 iff archive is solid                MF-
    FLAG_STRINGS_M      = ('ADDSIZE',   'COMMENT',  '4',          '8',
                           '16',        '32',       '64',         '128',
                           'V20FORMAT', 'SFX',      'LIMITSFXJR', 'MULTIVOLUME',
                           'ADVERT',    'RECOVERY', 'LOCKED',     'SOLID')
    FLAG_STRINGS_F      = ('ADDSIZE',   'COMMENT',  '64BIT',      '8',
                           '16',        '32',       '64',         '128',
                           '256',       '512',      'NTSECURITY', '2048',
                           'CONTPREV',  'CONTNEXT', 'PASSWORD',   'SOLID')
    FLAG_STRINGS_R      = ('ADDSIZE',   '2',        '64BIT',      '8',
                           '16',        '32',       '64',         '128',
                           '256',       '512',      '1024',       '2048',
                           '4096',      '8192',     '16384',      '32768')
    FLAG_STRINGS_BYTYPE = (FLAG_STRINGS_M, FLAG_STRINGS_F, FLAG_STRINGS_R,
                           FLAG_STRINGS_F, FLAG_STRINGS_R, FLAG_STRINGS_R)

    HOST_MSDOS          =  0
    HOST_OS2            =  1
    HOST_WIN32          =  2
    HOST_UNIX           =  3
    HOST_MAC_OS         =  4
    HOST_WIN_NT         =  5
    HOST_PRIMOS         =  6
    HOST_APPLE_GS       =  7
    HOST_ATARI          =  8
    HOST_VAX_VMS        =  9
    HOST_AMIGA          = 10
    HOST_NEXT           = 11
    HOST_LINUX          = 12
    HOST_STRINGS        = ('MS-DOS', 'OS/2', 'Win32', 'Unix', 'Mac OS',
                           'Win NT', 'Primos', 'Apple GS', 'ATARI', 'VAX VMS',
                           'AMIGA', 'NeXT', 'Linux')

    COMP_STORED         = 0
    COMP_LZ77           = 1
    COMP_BLOCKED        = 2
    COMP_STRINGS        = ('stored', 'lz77', 'blocked')

    QUAL_NONE           = 0
    QUAL_FASTEST        = 1
    QUAL_FAST           = 2
    QUAL_NORMAL         = 3
    QUAL_GOOD           = 4
    QUAL_BEST           = 5
    QUAL_STRINGS        = ('store', 'fastest', 'fast', 'normal', 'good', 'best')

    # winnt.h
    ATTR_READONLY               = 0x00000001
    ATTR_HIDDEN                 = 0x00000002
    ATTR_SYSTEM                 = 0x00000004
    ATTR_VOLUME_ID              = 0x00000008
    ATTR_DIRECTORY              = 0x00000010
    ATTR_ARCHIVE                = 0x00000020
    ATTR_DEVICE                 = 0x00000040
    ATTR_NORMAL                 = 0x00000080
    ATTR_TEMPORARY              = 0x00000100
    ATTR_SPARSE_FILE            = 0x00000200
    ATTR_REPARSE_POINT          = 0x00000400
    ATTR_COMPRESSED             = 0x00000800
    ATTR_OFFLINE                = 0x00001000
    ATTR_NOT_CONTENT_INDEXED    = 0x00002000
    ATTR_ENCRYPTED              = 0x00004000
    ATTR_INTEGRITY_STREAM       = 0x00008000
    ATTR_VIRTUAL                = 0x00010000
    ATTR_NO_SCRUB_DATA          = 0x00020000
    ATTR_EA                     = 0x00040000
    ATTR_STRINGS                = ('READONLY', 'HIDDEN', 'SYSTEM', 'VOLUME_ID',
                                   'DIRECTORY', 'ARCHIVE', 'DEVICE', 'NORMAL',
                                   'TEMPORARY', 'SPARSE_FILE',
                                   'REPARSE_POINT', 'COMPRESSED',
                                   'OFFLINE', 'NOT_CONTENT_INDEXED',
                                   'ENCRYPTED', 'INTEGRITY_STREAM',
                                   'VIRTUAL', 'NO_SCRUB_DATA', 'EA')

    @staticmethod
    def _format_bitfield(strings, field):
        labels = []
        for i in range(field.bit_length()):
            bit = 1 << i
            if field & bit == bit:
                try:
                    labels.append(strings[i])
                except IndexError:
                    labels.append(str(bit))
        return '|'.join(labels)

    def __init__(self, crc, size, type, flags):
        self.hdr_crc    = crc       # uint16    header crc without crc,sz
        self.hdr_size   = size      # uint16    header size without crc,sz
        self.hdr_type   = type      # uint8
        self.hdr_flags  = flags     # uint16

    def __str__(self):
        return """header
    hdr_crc     0x%04x
    hdr_size    %i
    hdr_type    0x%02x        %s
    hdr_flags   0x%04x      %s""" % (
                self.hdr_crc,
                self.hdr_size,
                self.hdr_type, self.hdr_type_str,
                self.hdr_flags, self.hdr_flags_str)

    def flag(self, flag):
        return self.hdr_flags & flag == flag

    @property
    def hdr_type_str(self):
        try:
            return Header.TYPE_STRINGS[self.hdr_type]
        except IndexError:
            return '?'

    @property
    def hdr_flags_str(self):
        try:
            strings = self.FLAG_STRINGS_BYTYPE[self.hdr_type]
            return self._format_bitfield(strings, self.hdr_flags)
        except IndexError:
            return '?'



class UnknownHeader(Header):
    pass



class MainHeader(Header):
    def __init__(self, *args):
        super().__init__(*args)
        self.magic      = None      # uint8[7]  **ACE**
        self.eversion   = None      # uint8     extract version
        self.cversion   = None      # uint8     creator version
        self.host       = None      # uint8     platform
        self.volume     = None      # uint8     volume number
        self.datetime   = None      # uint32    date/time in MS-DOS format
        self.reserved1  = None      # uint8[8]
        self.advert     = b''       # [uint8]   optional
        self.comment    = b''       # [uint16]  optional, compressed
        self.reserved2  = None      # [?]       optional

    def __str__(self):
        return super().__str__() + """
    magic       %r
    eversion    %i          %s
    cversion    %i          %s
    host        0x%02x        %s
    volume      %i
    datetime    0x%08x  %s
    reserved1   %02x %02x %02x %02x %02x %02x %02x %02x
    advert      %r
    comment     %r
    reserved2   %r""" % (
                self.magic,
                self.eversion, self.eversion/10,
                self.cversion, self.cversion/10,
                self.host, self.host_str,
                self.volume,
                self.datetime,
                _dt_fromdos(self.datetime).strftime('%Y-%m-%d %H:%M:%S'),
                self.reserved1[0], self.reserved1[1],
                self.reserved1[2], self.reserved1[3],
                self.reserved1[4], self.reserved1[5],
                self.reserved1[6], self.reserved1[7],
                self.advert,
                self.comment,
                self.reserved2)

    @property
    def host_str(self):
        try:
            return Header.HOST_STRINGS[self.host]
        except IndexError:
            return '?'



class FileHeader(Header):
    def __init__(self, *args):
        super().__init__(*args)
        self.packsize   = None      # uint32|64 packed size
        self.origsize   = None      # uint32|64 original size
        self.datetime   = None      # uint32    ctime
        self.attribs    = None      # uint32    file attributes
        self.crc32      = None      # uint32    checksum over compressed file
        self.comptype   = None      # uint8     compression type
        self.compqual   = None      # uint8     compression quality
        self.params     = None      # uint16    decompression parameters
        self.reserved1  = None      # uint16
        self.filename   = None      # [uint16]
        self.comment    = b''       # [uint16]  optional, compressed
        self.ntsecurity = b''       # [uint16]  optional
        self.reserved2  = None      # ?
        self.dataoffset = None      #           position of data after hdr

    def __str__(self):
        return super().__str__() + """
    packsize    %i
    origsize    %i
    datetime    0x%08x  %s
    attribs     0x%08x  %s
    crc32       0x%08x
    comptype    0x%02x        %s
    compqual    0x%02x        %s
    params      0x%04x
    reserved1   0x%04x
    filename    %r
    comment     %r
    ntsecurity  %r
    reserved2   %r""" % (
                self.packsize,
                self.origsize,
                self.datetime,
                _dt_fromdos(self.datetime).strftime('%Y-%m-%d %H:%M:%S'),
                self.attribs, self.attribs_str,
                self.crc32,
                self.comptype, self.comptype_str,
                self.compqual, self.compqual_str,
                self.params,
                self.reserved1,
                self.filename,
                self.comment,
                self.ntsecurity,
                self.reserved2)

    def attrib(self, attrib):
        return self.attribs & attrib == attrib

    @property
    def attribs_str(self):
        return self._format_bitfield(Header.ATTR_STRINGS, self.attribs)

    @property
    def comptype_str(self):
        try:
            return Header.COMP_STRINGS[self.comptype]
        except IndexError:
            return '?'

    @property
    def compqual_str(self):
        try:
            return Header.QUAL_STRINGS[self.compqual]
        except IndexError:
            return '?'



class RecoveryHeader(Header):
    def __init__(self, *args):
        super().__init__(*args)
                                    # R32    R64A   R64B
        self.rcvrsize       = None  # uint32 uint64 uint64  recovery data size
        self.magic          = None  # uint8[7]              **ACE**
        self.relstart       = None  # uint32 uint64 uint64  ?
        self.cluster        = None  # uint32 uint32 -       ?
        self.sectors        = None  # -      -      uint16  ?
        self.spc            = None  # -      -      uint16  ?
        self.clustersize    = None  # uint32 uint32 uint32  ?
        self.rcvrcrc        = None  # uint16 uint16 -       recovery data crc

    def __str__(self):
        segments = [super().__str__()]
        segments.append("""
    rcvrsize    %i
    magic       %r
    relstart    %i""" % (
                self.rcvrsize,
                self.magic,
                self.relstart))
        if self.hdr_type in (Header.TYPE_RECOVERY32,
                             Header.TYPE_RECOVERY64A):
            segments.append("""
    cluster     %i
    clustersize %i
    rcvrcrc     0x%02x""" % (
                self.cluster,
                self.clustersize,
                self.rcvrcrc))
        elif self.hdr_type == Header.TYPE_RECOVERY64B:
            segments.append("""
    sectors     %i
    spc         %i
    clustersize %i""" % (
                self.sectors,
                self.spc,
                self.clustersize))
        return ''.join(segments)



class AceError(Exception):
    """
    Base class for all :mod:`acefile` exceptions.
    """
    pass

class MainHeaderNotFoundError(AceError):
    """
    The main ACE header marked by the magic bytes ``**ACE**`` could not be
    found.
    Either the *search* argument was to small or the archive is not an ACE
    format archive.
    """
    pass

class MultiVolumeArchiveError(AceError):
    """
    A multi-volume archive was expected but a normal archive was found, or
    mismatching volumes were provided, or while reading a member from a
    multi-volume archive, the member headers indicate that the member
    continues in the next volume, but no next volume was found or provided.
    """
    pass

class CorruptedArchiveError(AceError):
    """
    Archive is corrupted.  Either a header or data CRC check failed, an invalid
    value was read from the archive or the archive is truncated.
    """
    pass

class EncryptedArchiveError(AceError):
    """
    Archive member is encrypted but either no password was provided, or
    decompression failed with the given password.
    Also raised when processing an encrypted solid archive member out of order,
    when any previous archive member uses a different password than the archive
    member currently being accessed.

    .. note::

        Due to the lack of a password verifier in the ACE file format, there is
        no straightforward way to distinguish a wrong password from a corrupted
        archive.  If the CRC check of an encrypted archive member fails or an
        :class:`CorruptedArchiveError` is encountered during decompression, it
        is assumed that the password was wrong and as a consequence,
        :class:`EncryptedArchiveError` is raised.
    """
    pass

class UnknownCompressionMethodError(AceError):
    """
    Data was compressed using an unknown compression method and therefore
    cannot be decompressed using this implementation.  This should not happen
    for ACE 1.0 or ACE 2.0 archives since this implementation implements all
    existing compression methods.
    """
    pass



class AceMember:
    """
    Represents a single archive member, potentially spanning multiple
    archive volumes.
    :class:`AceMember` is not directly instantiated; instead, instances are
    returned by :meth:`AceArchive.getmember` and :meth:`AceArchive.getmembers`.
    """

    # https://msdn.microsoft.com/en-us/library/windows/desktop/aa365247
    if platform.system() == 'Windows':
        RESERVED_NAMES = ('CON', 'PRN', 'AUX', 'NUL', 'COM1', 'COM2', 'COM3',
                           'COM4', 'COM5', 'COM6', 'COM7', 'COM8', 'COM9',
                           'LPT1', 'LPT2', 'LPT3', 'LPT4', 'LPT5', 'LPT6',
                           'LPT7', 'LPT8', 'LPT9')
    else:
        RESERVED_NAMES = ()

    RESERVED_CHARS = ':<>"?*|' + ''.join([chr(x) for x in range(1,32)])
    TRANSLATION_TAB = str.maketrans(dict.fromkeys(RESERVED_CHARS))

    @staticmethod
    def _sanitize_filename(filename):
        """
        Decode and sanitize filename for security and platform independence.
        Returns either a sanitized relative path, or an empty string.
        This python implementation was not vulnerable to CVE-2018-20250, but we
        include a test vector to make sure anyway.
        While the author believes this sanitization function to be safe, it is
        entirely possible that it is not.  For maximum safety against path
        traversal attacks, for instance when unpacking malicious code, you may
        want to ignore the filenames in the archive headers and instead
        generate your own filename for each archive member.

        >>> AceMember._sanitize_filename(b'a.exe\\0b.txt')
        'a.exe'
        >>> AceMember._sanitize_filename(b'\\\\etc\\\\foo/bar\\\\baz.txt')
        'etc/foo/bar/baz.txt'
        >>> AceMember._sanitize_filename(b'a/b/../b/.//.././c/.//../d/file.txt')
        'a/d/file.txt'
        >>> AceMember._sanitize_filename(b'/etc/passwd')
        'etc/passwd'
        >>> AceMember._sanitize_filename(b'.././.././.././.././../etc/passwd')
        'etc/passwd'
        >>> AceMember._sanitize_filename(b'C:\\\\Windows\\\\foo.exe')
        'C/Windows/foo.exe'
        >>> AceMember._sanitize_filename(b'\\\\\\\\server\\\\share\\\\file')
        'server/share/file'
        >>> AceMember._sanitize_filename(b'\\\\\\\\.\\\\CdRom0')
        'CdRom0'
        >>> AceMember._sanitize_filename(b'\\\\\\\\?\\\\raw\\\\path')
        'raw/path'
        >>> AceMember._sanitize_filename(b'hello\x05world')
        'helloworld'
        >>> AceMember._sanitize_filename(b'.././.././.././.././../etc/../')
        ''
        >>> AceMember._sanitize_filename(b'c:\\\\c:\\\\CVE-2018-20250\\\\p.lnk')
        'c/c/CVE-2018-20250/p.lnk'
        """
        filename = filename.decode('utf-8', errors='replace')
        # treat null byte as filename terminator
        nullbyte = filename.find(chr(0))
        if nullbyte >= 0:
            filename = filename[0:nullbyte]
        # ensure path separators are consistent with current platform
        if os.sep != '/':
            filename = filename.replace('/', os.sep)
        if os.sep != '\\':
            filename = filename.replace('\\', os.sep)
        # eliminate characters illegal on some platforms, including some
        # characters that are relevant for path traversal attacks (i.e. colon)
        filename = filename.translate(AceMember.TRANSLATION_TAB)
        # first eliminate all /./, foo/../ and similar sequences, then remove
        # all remaining .. labels in order to avoid path traversal attacks but
        # still allow a safe subset of dot syntax in the filename
        filename = os.path.normpath(filename)
        escsep = re.escape(os.sep)
        pattern = r'(^|%s)(?:\.\.(?:%s|$))+' % (escsep, escsep)
        filename = re.sub(pattern, r'\1', filename)
        # remove leading path separators to ensure a relative path
        filename = filename.lstrip(os.sep)
        # avoid reserved file names
        if filename in AceMember.RESERVED_NAMES:
            return '_' + filename
        return filename

    def __init__(self, idx, filehdrs, f):
        """
        Initialize an :class:`AceMember` object with index within archive *idx*,
        initial file header *filehdr* and underlying file-like object *f*.
        """
        self._idx           = idx
        self._file          = f
        self._headers       = filehdrs
        self.__attribs      = filehdrs[0].attribs
        self.__comment      = filehdrs[0].comment.decode('utf-8',
                                                         errors='replace')
        self.__crc32        = filehdrs[-1].crc32
        self.__comptype     = filehdrs[0].comptype
        self.__compqual     = filehdrs[0].compqual
        self.__datetime     = _dt_fromdos(filehdrs[0].datetime)
        self.__dicsizebits  = (filehdrs[0].params & 15) + 10
        self.__dicsize      = 1 << self.__dicsizebits
        self.__raw_filename = filehdrs[0].filename
        self.__filename     = self._sanitize_filename(filehdrs[0].filename)
        if self.__filename == '':
            self.__filename = 'file%04i' % self._idx
        self.__ntsecurity   = filehdrs[0].ntsecurity
        self.__size         = filehdrs[0].origsize
        self.__packsize     = 0
        for hdr in filehdrs:
            self.__packsize += hdr.packsize

    def is_dir(self):
        """
        True iff :class:`AceMember` instance describes a directory.
        """
        return self.attribs & Header.ATTR_DIRECTORY != 0

    def is_enc(self):
        """
        True iff :class:`AceMember` instance describes an encrypted archive
        member.
        """
        return self._headers[0].flag(Header.FLAG_PASSWORD)

    def is_reg(self):
        """
        True iff :class:`AceMember` instance describes a regular file.
        """
        return not self.is_dir()

    @property
    def attribs(self):
        """
        DOS/Windows file attribute bit field, as :class:`int`,
        as produced by the Windows :func:`GetFileAttributes` API.
        """
        return self.__attribs

    @property
    def comment(self):
        """
        File-level comment, as :class:`str`.
        If absent, empty :class:`str`.
        """
        return self.__comment

    @property
    def comptype(self):
        """
        Compression type used; one of
        :data:`COMP_STORED`,
        :data:`COMP_LZ77` or
        :data:`COMP_BLOCKED`.
        """
        return self.__comptype

    @property
    def compqual(self):
        """
        Compression quality used; one of
        :data:`QUAL_NONE`,
        :data:`QUAL_FASTEST`,
        :data:`QUAL_FAST`,
        :data:`QUAL_NORMAL`,
        :data:`QUAL_GOOD` or
        :data:`QUAL_BEST`.
        """
        return self.__compqual

    @property
    def crc32(self):
        """
        ACE CRC-32 checksum of decompressed data as recorded in the archive,
        as :class:`int`.
        ACE CRC-32 is the bitwise inverse of standard CRC-32.
        """
        return self.__crc32

    @property
    def datetime(self):
        """
        Timestamp as recorded in the archive, as :class:`datetime.datetime`
        instance.
        """
        return self.__datetime

    @property
    def dicsize(self):
        """
        LZ77 dictionary size required for extraction of this archive member
        in literal symbols, ranging from 1K to 4M.
        """
        return 1 << self.__dicsizebits

    @property
    def dicsizebits(self):
        """
        LZ77 dictionary size bit length, i.e. the base-two logarithm of the
        dictionary size required for extraction of this archive member.
        """
        return self.__dicsizebits

    @property
    def filename(self):
        """
        Sanitized filename, as :class:`str`, safe for use with file
        operations on the current platform.
        """
        return self.__filename

    @property
    def ntsecurity(self):
        """
        NT security descriptor as :class:`bytes`, describing the owner, primary
        group and discretionary access control list (DACL) of the archive
        member, as produced by the Windows :func:`GetFileSecurity` API with the
        :data:`OWNER_SECURITY_INFORMATION`,
        :data:`GROUP_SECURITY_INFORMATION` and
        :data:`DACL_SECURITY_INFORMATION` flags set.
        If absent, empty :class:`bytes`.
        """
        return self.__ntsecurity

    @property
    def packsize(self):
        """
        Size before decompression (packed size).
        """
        return self.__packsize

    @property
    def raw_filename(self):
        """
        Raw, unsanitized filename, as :class:`bytes`, not safe for use with
        file operations and possibly using path syntax from other platforms.
        """
        return self.__raw_filename

    @property
    def size(self):
        """
        Size after decompression (original size).
        """
        return self.__size

    def __str__(self):
        return "member idx %i sz %i/%i type %i qual %i: %s" % (
                self._idx, self.packsize, self.size,
                self.comptype, self.compqual,
                self.filename)



class AceVolume:
    """
    Parse and represent a single archive volume.
    """

    def __init__(self, file, mode='r', *, search=524288, _idx=0, _am=None):
        if mode != 'r':
            raise NotImplementedError("mode != 'r' not implemented")
        if isinstance(file, str):
            self.__file = builtins.open(file, 'rb')
            self.__filename = file
        else:
            if not file.seekable():
                raise TypeError("file must be filename or "
                                "seekable file-like object")
            self.__file = file
            self.__filename = '-'
        self.__file.seek(0, os.SEEK_END)
        self.__filesize = self.__file.tell()
        self.__main_header = None
        self.__file_headers = []
        self.__recovery_headers = []
        self.__all_headers = []
        try:
            self._parse_headers(search)
            if self.__main_header == None:
                raise CorruptedArchiveError("no main header")
        except:
            self.close()
            raise

    def close(self):
        """
        Close the underlying file object for this volume.
        Can safely be called multiple times.
        """
        if self.__file != None:
            self.__file.close()
            self.__file = None

    def dumpheaders(self, file=sys.stdout):
        """
        Dump ACE headers in this archive volume to *file*.
        """
        print("""volume
    filename    %s
    filesize    %i
    headers     MAIN:1 FILE:%i RECOVERY:%i others:%i""" % (
                self.__filename,
                self.__filesize,
                len(self.__file_headers),
                len(self.__recovery_headers),
                len(self.__all_headers) - len(self.__file_headers) - \
                        len(self.__recovery_headers) - 1),
            file=file)
        for h in self.__all_headers:
            print(h, file=file)

    def file_segment_for(self, fhdr):
        """
        Returns a :class:`FileSegmentIO` object for the file header *fhdr*
        belonging to this volume.
        """
        assert fhdr in self.__file_headers
        return FileSegmentIO(self.__file, fhdr.dataoffset, fhdr.packsize)

    def _next_filename(self):
        """
        Derive the filename of the next volume after this one.
        If the filename ends in ``.[cC]XX``, XX is incremented by 1.
        Otherwise self is assumed to be the first in the series and
        ``.[cC]00`` is used as extension.
        Returns the derived filename in two variants, upper and lower case,
        to allow for finding the file on fully case-sensitive filesystems.
        """
        base, ext = os.path.splitext(self.__filename)
        ext = ext.lower()
        if ext[:2] == '.c':
            try:
                n = int(ext[2:])
                return (base + ('.c%02i' % (n + 1)),
                        base + ('.C%02i' % (n + 1)))
            except ValueError:
                pass
        return (base + '.c00',
                base + '.C00')

    def try_load_next_volume(self, mode):
        """
        Open the next volume following this one in a multi-volume archive
        and return the instantiated :class:`AceVolume` object.
        """
        for nextname in self._next_filename():
            try:
                return AceVolume(nextname, mode=mode, search=0)
            except FileNotFoundError:
                continue
        return None

    def _parse_headers(self, search):
        """
        Parse ACE headers from self.__file.  If *search* is > 0, search for
        the magic bytes in the first *search* bytes of the file.
        Raises MainHeaderNotFoundError if the main header could not be located.
        Raises other exceptions if parsing fails for other reasons.
        On success, loads all the parsed headers into self.__main_header,
        self.__file_headers, self.__recovery_headers and self.__all_headers.
        """
        self.__file.seek(0)
        buf = self.__file.read(512)
        found_at_start = False
        if buf[7:14] == MainHeader.MAGIC:
            self.__file.seek(0)
            try:
                self._parse_header()
                found_at_start = True
            except CorruptedArchiveError:
                pass
        if not found_at_start:
            if search == 0:
                raise MainHeaderNotFoundError("no ACE header at offset 0")
            self.__file.seek(0)
            buf = self.__file.read(search)
            magicpos = 7
            while magicpos < search:
                magicpos = buf.find(MainHeader.MAGIC, magicpos + 1, search)
                if magicpos == -1:
                    raise MainHeaderNotFoundError(
                            "no ACE header within first %i bytes" % search)
                self.__file.seek(magicpos - 7)
                try:
                    self._parse_header()
                    break
                except CorruptedArchiveError:
                    continue
        while self.__file.tell() < self.__filesize:
            self._parse_header()

    def _parse_header(self):
        """
        Parse a single header from self.__file at the current file position.
        Raises CorruptedArchiveError if the header cannot be parsed.
        Guarantees that no data is written to object state
        if an exception is thrown, otherwise the header is added to
        self.__main_header, self.__file_headers, self.__recovery_headers
        and self.__all_headers.
        """
        buf = self.__file.read(4)
        if len(buf) < 4:
            raise CorruptedArchiveError("truncated header")
        hcrc, hsize = struct.unpack('<HH', buf)
        buf = self.__file.read(hsize)
        if len(buf) < hsize:
            raise CorruptedArchiveError("truncated header")
        if ace_crc16(buf) != hcrc:
            raise CorruptedArchiveError("header CRC failed")
        htype, hflags = struct.unpack('<BH', buf[0:3])
        i = 3

        if htype == Header.TYPE_MAIN:
            header = MainHeader(hcrc, hsize, htype, hflags)
            if header.flag(Header.FLAG_ADDSIZE):
                raise CorruptedArchiveError("main header has addsize > 0")
            header.magic = buf[3:10]
            if header.magic != MainHeader.MAGIC:
                raise CorruptedArchiveError("main header without magic")
            header.eversion, \
            header.cversion, \
            header.host, \
            header.volume, \
            header.datetime = struct.unpack('<BBBBL', buf[10:18])
            header.reserved1 = buf[18:26]
            i = 26
            if header.flag(Header.FLAG_ADVERT):
                if i + 1 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                avsz, = struct.unpack('<B', buf[i:i+1])
                i += 1
                if i + avsz > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.advert = buf[i:i+avsz]
                i += avsz
            if header.flag(Header.FLAG_COMMENT):
                if i + 2 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                cmsz, = struct.unpack('<H', buf[i:i+2])
                i += 2
                if i + cmsz > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.comment = ACE.decompress_comment(buf[i:i+cmsz])
                i += cmsz
            header.reserved2 = buf[i:]
            if self.__main_header != None:
                raise CorruptedArchiveError("multiple main headers")
            self.__main_header = header

        elif htype in (Header.TYPE_FILE32, Header.TYPE_FILE64):
            header = FileHeader(hcrc, hsize, htype, hflags)
            if not header.flag(Header.FLAG_ADDSIZE):
                raise CorruptedArchiveError("file header with addsize == 0")
            if header.flag(Header.FLAG_64BIT):
                if htype != Header.TYPE_FILE64:
                    raise CorruptedArchiveError("64 bit flag in 32 bit header")
                if i + 16 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.packsize, \
                header.origsize, = struct.unpack('<QQ', buf[i:i+16])
                i += 16
            else:
                if htype != Header.TYPE_FILE32:
                    raise CorruptedArchiveError("32 bit flag in 64 bit header")
                if i + 8 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.packsize, \
                header.origsize, = struct.unpack('<LL', buf[i:i+8])
                i += 8
            if i + 20 > len(buf):
                raise CorruptedArchiveError("truncated header")
            header.datetime, \
            header.attribs, \
            header.crc32, \
            header.comptype, \
            header.compqual, \
            header.params, \
            header.reserved1, \
            fnsz = struct.unpack('<LLLBBHHH', buf[i:i+20])
            i += 20
            if i + fnsz > len(buf):
                raise CorruptedArchiveError("truncated header")
            header.filename = buf[i:i+fnsz]
            i += fnsz
            if header.flag(Header.FLAG_COMMENT):
                if i + 2 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                cmsz, = struct.unpack('<H', buf[i:i+2])
                i += 2
                if i + cmsz > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.comment = ACE.decompress_comment(buf[i:i+cmsz])
                i += cmsz
            if header.flag(Header.FLAG_NTSECURITY):
                if i + 2 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                nssz, = struct.unpack('<H', buf[i:i+2])
                i += 2
                if i + nssz > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.ntsecurity = buf[i:i+nssz]
                i += nssz
            header.reserved2 = buf[i:]
            header.dataoffset = self.__file.tell()
            self.__file_headers.append(header)
            self.__file.seek(header.packsize, os.SEEK_CUR)

        elif htype in (Header.TYPE_RECOVERY32,
                       Header.TYPE_RECOVERY64A,
                       Header.TYPE_RECOVERY64B):
            header = RecoveryHeader(hcrc, hsize, htype, hflags)
            if not header.flag(Header.FLAG_ADDSIZE):
                raise CorruptedArchiveError("recovery header with addsize == 0")
            if header.flag(Header.FLAG_64BIT) and \
               htype == Header.TYPE_RECOVERY32:
                raise CorruptedArchiveError("64 bit flag in 32 bit header")
            if not header.flag(Header.FLAG_64BIT) and \
               htype != Header.TYPE_RECOVERY32:
                raise CorruptedArchiveError("32 bit flag in 64 bit header")
            if header.flag(Header.FLAG_64BIT):
                if i + 23 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.rcvrsize, \
                header.magic, \
                header.relstart = struct.unpack('<Q7sQ', buf[i:i+23])
                i += 23
            else:
                if i + 15 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.rcvrsize, \
                header.magic, \
                header.relstart = struct.unpack('<L7sL', buf[i:i+15])
                i += 15
            if htype == Header.TYPE_RECOVERY64B:
                if i + 8 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.sectors, \
                header.spc, \
                header.clustersize = struct.unpack('<HHL', buf[i:i+8])
                i += 8
            else:
                if i + 10 > len(buf):
                    raise CorruptedArchiveError("truncated header")
                header.cluster, \
                header.clustersize, \
                header.rcvrcrc = struct.unpack('<LLH', buf[i:i+10])
                i += 10
            header.dataoffset = self.__file.tell()
            self.__recovery_headers.append(header)
            self.__file.seek(header.rcvrsize, os.SEEK_CUR)

        else:
            header = UnknownHeader(hcrc, hsize, htype, hflags)
            addsz = 0
            if header.flag(Header.FLAG_ADDSIZE):
                if header.flag(Header.FLAG_64BIT):
                    if i + 8 > len(buf):
                        raise CorruptedArchiveError("truncated header")
                    addsz, = struct.unpack('<Q', buf[i:i+8])
                else:
                    if i + 4 > len(buf):
                        raise CorruptedArchiveError("truncated header")
                    addsz, = struct.unpack('<L', buf[i:i+4])
            self.__file.seek(addsz, os.SEEK_CUR)

        self.__all_headers.append(header)

    def get_file_headers(self):
        return self.__file_headers

    def get_recovery_headers(self):
        return self.__recovery_headers

    def is_locked(self):
        return self.__main_header.flag(Header.FLAG_LOCKED)

    def is_multivolume(self):
        return self.__main_header.flag(Header.FLAG_MULTIVOLUME)

    def is_solid(self):
        return self.__main_header.flag(Header.FLAG_SOLID)

    @property
    def advert(self):
        return self.__main_header.advert.decode('utf-8', errors='replace')

    @property
    def comment(self):
        return self.__main_header.comment.decode('utf-8', errors='replace')

    @property
    def cversion(self):
        return self.__main_header.cversion

    @property
    def eversion(self):
        return self.__main_header.eversion

    @property
    def filename(self):
        return self.__filename

    @property
    def datetime(self):
        return _dt_fromdos(self.__main_header.datetime)

    @property
    def platform(self):
        return self.__main_header.host_str

    @property
    def volume(self):
        return self.__main_header.volume



class AceArchive:
    """
    Represents an ACE archive, possibly consisting of multiple volumes.
    :class:`AceArchive` is not directly instantiated; instead, instances are
    returned by :meth:`acefile.open`.

    When used as a context manager, :class:`AceArchive` ensures that
    :meth:`AceArchive.close` is called after the block.
    When used as an iterator, :class:`AceArchive` yields instances of
    :class:`AceMember` representing all archive members in order of
    appearance in the archive.
    """

    @classmethod
    def _open(cls, file, mode='r', *, search=524288):
        """
        Open archive from *file*, which is either a filename or seekable
        file-like object, and return an instance of :class:`AceArchive`
        representing the opened archive that can function as a context
        manager.
        Only *mode* 'r' is implemented.
        If *search* is 0, the archive must start at position 0 in *file*,
        otherwise the first *search* bytes are searched for the magic bytes
        ``**ACE**`` that mark the ACE main header.
        For 1:1 compatibility with the official unace, 1024 sectors are
        searched by default, even though none of the SFX stubs that come with
        ACE compressors are that large.

        Multi-volume archives are represented by a single :class:`AceArchive`
        object to the caller, all operations transparently read into subsequent
        volumes as required.
        To load a multi-volume archive, either open the first volume of the
        series by filename, or provide a list or tuple of all file-like
        objects or filenames in the correct order in *file*.
        """
        return cls(file, mode, search=search)

    def __init__(self, file, mode='r', *, search=524288):
        """
        See :meth:`AceArchive._open`.
        """
        if mode != 'r':
            raise NotImplementedError("mode != 'r' not implemented")
        if isinstance(file, (tuple, list)):
            if len(file) == 0:
                raise ValueError("file is empty tuple/list")
        else:
            file = (file,)

        self.__volumes = []
        try:
            # load volumes
            self.__tmp_file = file[0]
            self.__volumes.append(AceVolume(file[0], mode, search=search))
            self.__tmp_file = None
            if self.__volumes[0].is_multivolume():
                for f in file[1:]:
                    self.__tmp_file = f
                    self.__volumes.append(AceVolume(f, mode, search=0))
                    self.__tmp_file = None
                if len(self.__volumes) == 1 and isinstance(file[0], str):
                    vol = self.__volumes[0]
                    while True:
                        vol = vol.try_load_next_volume(mode)
                        if not vol:
                            break
                        self.__volumes.append(vol)

            # check volume linkage
            if len(self.__volumes) > 1:
                last_volume = None
                for vol in self.__volumes:
                    if not vol.is_multivolume():
                        raise MultiVolumeArchiveError("single-volume archive")
                    if last_volume != None and vol.volume != last_volume + 1:
                        raise MultiVolumeArchiveError("volumes do not match")
                    last_volume = vol.volume

            # build list of members and their file segments across volumes
            self.__members = []
            headers = []
            segments = []
            for volume in self.__volumes:
                for hdr in volume.get_file_headers():
                    if len(headers) == 0:
                        if hdr.flag(Header.FLAG_CONTPREV):
                            if len(self.__members) > 0:
                                raise MultiVolumeArchiveError("incomplete file")
                            # don't raise an error if this is the first file
                            # in the first volume, to allow opening subsequent
                            # volumes of multi-volume archives separately
                            continue
                    else:
                        if not hdr.flag(Header.FLAG_CONTPREV):
                            raise MultiVolumeArchiveError("unexpected new file")
                        if hdr.filename != headers[-1].filename:
                            raise MultiVolumeArchiveError("filename mismatch")
                    headers.append(hdr)
                    segments.append(volume.file_segment_for(hdr))
                    if not hdr.flag(Header.FLAG_CONTNEXT):
                        if len(segments) > 1:
                            f = MultipleFilesIO(segments)
                        else:
                            f = segments[0]
                        self.__members.append(AceMember(len(self.__members),
                                                        headers, f))
                        headers = []
                        segments = []

            self.__next_read_idx = 0
            self.__ace = ACE()
        except:
            self.close()
            raise

    def __enter__(self):
        """
        Using :class:`AceArchive` as a context manager ensures that
        :meth:`AceArchive.close` is called after leaving the block.
        """
        return self

    def __exit__(self, type, value, traceback):
        self.close()

    def __iter__(self):
        """
        Using :class:`AceArchive` as an iterater will iterate over
        :class:`AceMember` objects for all archive members.
        """
        return self.__members.__iter__()

    def __repr__(self):
        return "<%s %r at %#x>" % (self.__class__.__name__,
                                   self.filename,
                                   id(self))

    def close(self):
        """
        Close the archive and all open files.
        No other methods may be called after having called
        :meth:`AceArchive.close`, but calling :meth:`AceArchive.close`
        multiple times is permitted.
        """
        if self.__tmp_file != None:
            if not isinstance(self.__tmp_file, str):
                self.__tmp_file.close()
            self.__tmp_file = None
        for volume in self.__volumes:
            volume.close()

    def _getmember_byname(self, name):
        """
        Return an :class:`AceMember` object corresponding to archive member
        name *name*.
        Raise :class:`KeyError` if *name* is not present in the archive.
        If *name* occurs multiple times in the archive, then the last occurence
        is returned.
        """
        match = None
        for am in self.__members:
            if am.filename == name:
                match = am
        if match == None:
            raise KeyError("no member '%s' in archive" % name)
        return match

    def _getmember_byidx(self, idx):
        """
        Return an :class:`AceMember` object corresponding to archive member
        index *idx*.
        Raise :class:`IndexError` if *idx* is not present in the archive.
        """
        return self.__members[idx]

    def getmember(self, member):
        """
        Return an :class:`AceMember` object corresponding to archive
        member *member*.
        Raise :class:`KeyError` or :class:`IndexError` if *member* is not
        found in archive.
        *Member* can refer to an :class:`AceMember` object, a member name or
        an index into the archive member list.
        If *member* is a name and it occurs multiple times in the archive,
        then the last member with matching filename is returned.
        """
        if isinstance(member, int):
            return self._getmember_byidx(member)
        elif isinstance(member, AceMember):
            return member
        elif isinstance(member, str):
            return self._getmember_byname(member)
        else:
            raise TypeError("member argument has unsupported type")

    def getmembers(self):
        """
        Return a list of :class:`AceMember` objects for the members of the
        archive.
        The objects are in the same order as they are in the archive.
        For simply iterating over the members of an archive, it is more concise
        and functionally equivalent to directly iterate over the
        :class:`AceArchive` instance instead of over the list returned by
        :meth:`AceArchive.getmembers`.
        """
        return self.__members

    def getnames(self):
        """
        Return a list of the (file)names of all the members in the archive
        in the order they are in the archive.
        """
        return [am.filename for am in self.getmembers()]

    def extract(self, member, *, path=None, pwd=None, restore=False):
        """
        Extract an archive member to *path* or the current working directory.
        *Member* can refer to an :class:`AceMember` object, a member name or
        an index into the archive member list.
        Password *pwd* is used to decrypt the archive member if it is
        encrypted.
        Raises :class:`EncryptedArchiveError` if an archive member is
        encrypted but no password was provided.
        Iff *restore* is True, restore mtime and atime for non-dir members,
        file attributes and NT security information as far as supported by
        the platform.

        .. note::

            For **solid** archives, extracting members in a different order
            than they appear in the archive works, but is potentially very
            slow, because the decompressor needs to restart decompression at
            the beginning of the solid archive to restore internal decompressor
            state.
            For **encrypted solid** archives, out of order access may fail when
            archive members use different passwords.
        """
        am = self.getmember(member)

        if path != None:
            fn = os.path.join(path, am.filename)
        else:
            fn = am.filename
        if am.is_dir():
            try:
                os.mkdir(fn)
            except FileExistsError:
                pass
        else:
            basedir = os.path.dirname(fn)
            if basedir != '':
                os.makedirs(basedir, exist_ok=True)
            with builtins.open(fn, 'wb') as f:
                for buf in self.readblocks(am, pwd=pwd):
                    f.write(buf)
        if restore:
            if SetFileAttributes:
                SetFileAttributes(fn, am.attribs)
            elif am.attribs & Header.ATTR_READONLY != 0:
                mode = stat.S_IMODE(os.lstat(fn).st_mode)
                all_w = stat.S_IWUSR | stat.S_IWGRP | stat.S_IWOTH
                os.chmod(fn, mode & ~all_w)
            if SetFileSecurity and am.ntsecurity:
                SetFileSecurity(fn, 0x7, am.ntsecurity)
            if not am.is_dir():
                ts = am.datetime.timestamp()
                os.utime(fn, (ts, ts))

    def extractall(self, *, path=None, members=None, pwd=None, restore=False):
        """
        Extract *members* or all members from archive to *path* or the current
        working directory.
        *Members* can contain :class:`AceMember` objects, member names or
        indexes into the archive member list.
        Password *pwd* is used to decrypt encrypted archive members.
        To extract archives that use multiple different passwords for different
        archive members, you must use :meth:`AceArchive.extract` instead.
        Raises :class:`EncryptedArchiveError` if an archive member is
        encrypted but no password was provided.
        Iff *restore* is True, restore mtime and atime for non-dir members,
        file attributes and NT security information as far as supported by
        the platform.
        """
        if members == None or members == []:
            members = self.getmembers()
        else:
            if self.is_solid():
                # ensure members subset is in order of appearance
                sorted_members = []
                for member in self.getmembers():
                    if member in members or \
                       member.filename in members or \
                       member._idx in members:
                        sorted_members.append(member)
                members = sorted_members
        for am in members:
            self.extract(am, path=path, pwd=pwd, restore=restore)

    def read(self, member, *, pwd=None):
        """
        Read the decompressed bytes of an archive member.
        *Member* can refer to an :class:`AceMember` object, a member name or
        an index into the archive member list.
        Password *pwd* is used to decrypt the archive member if it is
        encrypted.
        Raises :class:`EncryptedArchiveError` if the archive member is
        encrypted but no password was provided.

        .. note::

            For **solid** archives, reading members in a different order than
            they appear in the archive works, but is potentially very slow,
            because the decompressor needs to restart decompression at the
            beginning of the solid archive to restore internal decompressor
            state.
            For **encrypted solid** archives, out of order access may fail when
            archive members use different passwords.

        .. note::

            Using :meth:`AceArchive.read` for large files is inefficient and
            may fail for very large files.
            Using :meth:`AceArchive.readblocks` to write the data to disk in
            blocks ensures that large files can be handled efficiently.
        """
        return b''.join(self.readblocks(member, pwd=pwd))

    def readblocks(self, member, *, pwd=None):
        """
        Read the archive member by yielding blocks of decompressed bytes.
        *Member* can refer to an :class:`AceMember` object, a member name or
        an index into the archive member list.
        Password *pwd* is used to decrypt the archive member if it is
        encrypted.
        Raises :class:`EncryptedArchiveError` if the archive member is
        encrypted but no password was provided.

        .. note::

            For **solid** archives, reading members in a different order than
            they appear in the archive works, but is potentially very slow,
            because the decompressor needs to restart decompression at the
            beginning of the solid archive to restore internal decompressor
            state.
            For **encrypted solid** archives, out of order access may fail when
            archive members use different passwords.
        """
        am = self.getmember(member)

        if DEBUG:
            eprint(am)

        # Need first volume available to read from solid multi-volume archives.
        if self.is_solid() and self.is_multivolume() and self.volume > 0:
            raise MultiVolumeArchiveError("need first volume")

        # For solid archives, ensure the LZ77 state corresponds to the state
        # after extracting the previous file by re-starting extraction from
        # the beginning or the last extracted file.  This is what makes out
        # of order access to solid archive members prohibitively slow.
        if self.is_solid() and self.__next_read_idx != am._idx:
            if self.__next_read_idx < am._idx:
                restart_idx = self.__next_read_idx
            else:
                restart_idx = self.__next_read_idx = 0
            for i in range(restart_idx, am._idx):
                if not self.test(i):
                    raise CorruptedArchiveError("failed to restore solid state")

        if (not am.is_dir()) and am.size > 0:
            f = am._file
            f.seek(0)

            # For password protected members, wrap the file-like object in
            # a decrypting wrapper object.
            if am.is_enc():
                if not pwd:
                    raise EncryptedArchiveError("need password")
                f = EncryptedFileIO(f, AceBlowfish(pwd))

            # Choose the matching decompressor based on the first header.
            if am.comptype == Header.COMP_STORED:
                decompressor = self.__ace.decompress_stored
            elif am.comptype == Header.COMP_LZ77:
                decompressor = self.__ace.decompress_lz77
            elif am.comptype == Header.COMP_BLOCKED:
                decompressor = self.__ace.decompress_blocked
            else:
                raise UnknownCompressionMethodError(
                        "method %i unknown" % am.comptype)

            # Decompress and calculate CRC over full decompressed data,
            # i.e. after decryption and across all segments that may have
            # been read from different volumes.
            crc = AceCRC32()
            try:
                for block in decompressor(f, am.size, am.dicsize):
                    crc += block
                    yield block
            except ValueError:
                if am.is_enc():
                    raise EncryptedArchiveError("wrong password or corrupted")
                else:
                    raise CorruptedArchiveError("ValueError during decomp")
            except CorruptedArchiveError:
                if am.is_enc():
                    raise EncryptedArchiveError("wrong password or corrupted")
                raise
            if crc != am.crc32:
                if am.is_enc():
                    raise EncryptedArchiveError("wrong password or corrupted")
                raise CorruptedArchiveError("CRC mismatch")

        self.__next_read_idx += 1

    def test(self, member, *, pwd=None):
        """
        Test an archive member.  Returns False if any corruption was
        found, True if the header and decompression was okay.
        *Member* can refer to an :class:`AceMember` object, a member name or
        an index into the archive member list.
        Password *pwd* is used to decrypt the archive member if it is
        encrypted.
        Raises :class:`EncryptedArchiveError` if the archive member is
        encrypted but no password was provided.

        .. note::

            For **solid** archives, testing members in a different order than
            they appear in the archive works, but is potentially very slow,
            because the decompressor needs to restart decompression at the
            beginning of the solid archive to restore internal decompressor
            state.
            For **encrypted solid** archives, out of order access may fail when
            archive members use different passwords.
        """
        try:
            for buf in self.readblocks(member, pwd=pwd):
                pass
            return True
        except EncryptedArchiveError:
            raise
        except AceError:
            if DEBUG:
                raise
            return False

    def testall(self, *, pwd=None):
        """
        Test all the members in the archive.  Returns the name of the first
        archive member with a failing header or content CRC, or None if all
        members were okay.
        Password *pwd* is used to decrypt encrypted archive members.
        To test archives that use multiple different passwords for different
        archive members, use :meth:`AceArchive.test` instead.
        Raises :class:`EncryptedArchiveError` if an archive member is
        encrypted but no password was provided.
        """
        for am in self.getmembers():
            if not self.test(am, pwd=pwd):
                return am.filename
        return None

    def dumpheaders(self, file=sys.stdout):
        """
        Dump all ACE file format headers in this archive and all its volumes
        to *file*.
        """
        for volume in self.__volumes:
            volume.dumpheaders()

    def is_locked(self):
        """
        Return True iff archive is locked for further modifications.
        Since this implementation does not support writing to archives,
        presence or absence of the flag in an archive does not change any
        behaviour of :mod:`acefile`.
        """
        return self.__volumes[0].is_locked()

    def is_multivolume(self):
        """
        Return True iff archive is a multi-volume archive as determined
        by the archive headers.  When opening the last volume of a
        multi-volume archive, this returns True even though only a single
        volume was loaded.
        """
        return self.__volumes[0].is_multivolume()

    def is_solid(self):
        """
        Return True iff archive is a solid archive, i.e. iff the archive
        members are linked to each other by sharing the same LZ77 dictionary.
        Members of solid archives should always be read/tested/extracted in
        the order they appear in the archive in order to avoid costly
        decompression restarts from the beginning of the archive.
        """
        return self.__volumes[0].is_solid()

    @property
    def advert(self):
        """
        ACE archive advert string as :class:`str`.
        Unregistered versions of ACE compressors communicate that they are
        unregistered by including an advert string of
        ``*UNREGISTERED VERSION*`` in archives they create.
        If absent, empty :class:`str`.
        """
        return self.__volumes[0].advert

    @property
    def comment(self):
        """
        ACE archive level comment as :class:`str`.
        If absent, empty :class:`str`.
        """
        return self.__volumes[0].comment

    @property
    def cversion(self):
        """
        ACE creator version.  This is equal to the major version of the ACE
        compressor used to create the archive, which equals the highest
        version of the ACE format supported by the ACE compressor which
        produced the archive.
        """
        return self.__volumes[0].cversion

    @property
    def eversion(self):
        """
        ACE extractor version.  This is the version of the ACE decompressor
        required to extract, which equals the version of the ACE format this
        archive is compliant with.
        """
        return self.__volumes[0].eversion

    @property
    def filename(self):
        """
        ACE archive filename.  This is not a property of the archive but rather
        just the filename passed to :func:`acefile.open`.
        """
        return self.__volumes[0].filename

    @property
    def datetime(self):
        """
        Archive timestamp as :class:`datetime.datetime` object.
        """
        return self.__volumes[0].datetime

    @property
    def platform(self):
        """
        String describing the platform on which the ACE archive was created.
        This is derived from the *host* field in the archive header.
        """
        return self.__volumes[0].platform

    @property
    def volume(self):
        """
        ACE archive volume number of the first volume of this ACE archive.
        """
        return self.__volumes[0].volume

    @property
    def volumes_loaded(self):
        """
        Number of loaded volumes in this archives.  When opening a subsequent
        volume of a multi-volume archive, this may be lower than the
        theoretical volume count.
        """
        return len(self.__volumes)



def is_acefile(file, *, search=524288):
    """
    Return True iff *file* refers to an ACE archive by filename or seekable
    file-like object.
    If *search* is 0, the archive must start at position 0 in *file*,
    otherwise the first *search* bytes are searched for the magic bytes
    ``**ACE**`` that mark the ACE main header.
    For 1:1 compatibility with the official unace, 1024 sectors are
    searched by default, even though none of the SFX stubs that come with
    ACE compressors are that large.
    """
    try:
        with open(file, search=search) as f:
            pass
        return True
    except AceError:
        return False



#: The compression type constant for no compression.
COMP_STORED  = Header.COMP_STORED
#: The compression type constant for ACE 1.0 LZ77 mode.
COMP_LZ77    = Header.COMP_LZ77
#: The compression type constant for ACE 2.0 blocked mode.
COMP_BLOCKED = Header.COMP_BLOCKED

#: The compression quality constant for no compression.
QUAL_NONE    = Header.QUAL_NONE
#: The compression quality constant for fastest compression.
QUAL_FASTEST = Header.QUAL_FASTEST
#: The compression quality constant for fast compression.
QUAL_FAST    = Header.QUAL_FAST
#: The compression quality constant for normal compression.
QUAL_NORMAL  = Header.QUAL_NORMAL
#: The compression quality constant for good compression.
QUAL_GOOD    = Header.QUAL_GOOD
#: The compression quality constant for best compression.
QUAL_BEST    = Header.QUAL_BEST

open = AceArchive._open

__all__ = ['is_acefile', 'open']
__all__.extend(filter(lambda name: name.startswith('COMP_'),
                      sorted(list(globals()))))
__all__.extend(filter(lambda name: name.startswith('QUAL_'),
                      sorted(list(globals()))))
__all__.extend(filter(lambda name: name.endswith('Error'),
                      sorted(list(globals()))))



def unace():
    import argparse
    import getpass
    import signal

    def title(docstr):
        return docstr.strip().split('\n', 1)[0]

    class Status:
        def __init__(self, argv0, action, archive):
            self.argv0 = os.path.basename(argv0)
            self.action = action + 'ing'
            self.archive = os.path.basename(archive)
            self.member = ''

        def __str__(self):
            return "%s: %s %s %s" % (self.argv0, self.action,
                                     self.archive, self.member)

    status = None

    def siginfo_handler(signum, frame):
        eprint(status)

    parser = argparse.ArgumentParser(description=title(__doc__))

    parser.add_argument('archive', type=str,
            help='archive to read from')
    parser.add_argument('file', nargs='*', type=str,
            help='file(s) in archive to operate on, default all')

    parser.add_argument('-V', '--version', action='version',
            version='acefile %s' % __version__,
            help='show version and exit')

    group = parser.add_mutually_exclusive_group()
    group.add_argument('--extract', '-x', default='extract',
            action='store_const', dest='mode', const='extract',
            help='extract files in archive (default)')
    group.add_argument('--test', '-t',
            action='store_const', dest='mode', const='test',
            help='test archive integrity')
    group.add_argument('--list', '-l',
            action='store_const', dest='mode', const='list',
            help='list files in archive')
    group.add_argument('--headers',
            action='store_const', dest='mode', const='headers',
            help='dump archive headers')
    group.add_argument('--selftest',
            action='store_const', dest='mode', const='selftest',
            help=argparse.SUPPRESS)

    parser.add_argument('-d', '--basedir', type=str, default='.', metavar='X',
            help='base directory for extraction')
    parser.add_argument('-p', '--password', type=str, metavar='X',
            help='password for decryption')
    parser.add_argument('-r', '--restore', action='store_true',
            help='restore mtime/atime, attribs and ntsecurity on extraction')
    parser.add_argument('-b', '--batch', action='store_true',
            help='suppress all interactive input')
    parser.add_argument('-v', '--verbose', action='store_true',
            help='be more verbose')
    parser.add_argument('--debug', action='store_true',
            help='show mode transitions and expose internal exceptions')

    # not implemented arguments that other unace implementations have:
    # --(no-)full-path              always full path extraction
    # --(no-)show-comments          show comments iff verbose
    # --(no-)overwrite-files        always overwrite files
    # --(no-)full-path-matching     always full path matching
    # --exclude(-list)              feature not implemented

    args = parser.parse_args()

    if args.mode != 'extract' and len(args.file) > 0:
        eprint("%s: error: not extracting, but files were specified" %
               os.path.basename(sys.argv[0]))
        sys.exit(1)

    if args.debug:
        global DEBUG
        DEBUG = True

    if hasattr(signal, 'SIGINFO'):
        signal.signal(signal.SIGINFO, siginfo_handler)
        status = Status(sys.argv[0], args.mode, args.archive)

    if args.archive == '-':
        if sys.stdin.buffer.seekable() and platform.system() != 'Windows':
            archive = sys.stdin.buffer
        else:
            archive = io.BytesIO(sys.stdin.buffer.read())
    else:
        archive = args.archive

    try:
        with open(archive) as f:
            if args.verbose:
                if acebitstream == None:
                    eprint(("warning: acebitstream c extension unavailable, "
                            "using pure-python bit stream"))
                eprint("processing archive %s" % f.filename)
                eprint("loaded %i volume(s) starting at volume %i" % (
                       f.volumes_loaded, f.volume))
                archinfo = []
                if not f.is_locked():
                    archinfo.append('not ')
                archinfo.append('locked, ')
                if not f.is_multivolume():
                    archinfo.append('not ')
                archinfo.append('multi-volume, ')
                if not f.is_solid():
                    archinfo.append('not ')
                archinfo.append('solid')
                eprint("archive is", ''.join(archinfo))
                eprint("last modified %s" % (
                       f.datetime.strftime('%Y-%m-%d %H:%M:%S')))
                eprint("created on %s with ACE %s for extraction with %s+" % (
                       f.platform, f.cversion/10, f.eversion/10))
                if f.advert:
                    eprint("advert [%s]" % f.advert)

                if f.is_multivolume() and f.volume > 0:
                    eprint("warning: this is not the initial volume of this "
                           "multi-volume archive")
                if f.comment:
                    eprint(asciibox(f.comment, title='archive comment'))

            if args.mode == 'extract':
                if f.is_multivolume() and f.volume > 0 and f.is_solid():
                    eprint(("error: need complete set of volumes to extract "
                            "from solid multivolume archive"))
                    sys.exit(1)
                failed = 0
                password = args.password
                if args.file:
                    members = [f.getmember(m) for m in args.file]
                else:
                    members = f.getmembers()
                for am in members:
                    if status:
                        status.member = am.filename
                    if am.is_enc() and password == None and not args.batch:
                        try:
                            password = getpass.getpass("%s password: " % \
                                                        am.filename)
                        except EOFError:
                            password = None
                    while True:
                        try:
                            f.extract(am, path=args.basedir,
                                          pwd=password,
                                          restore=args.restore)
                            if args.verbose:
                                eprint("%s" % am.filename)
                            break
                        except EncryptedArchiveError:
                            if args.verbose or args.batch or not password:
                                eprint("%s failed to decrypt" % am.filename)
                            if args.batch or not password:
                                failed += 1
                                break
                            try:
                                password = getpass.getpass("%s password: " % \
                                                            am.filename)
                            except EOFError:
                                password = ''
                            if password == '':
                                password = args.password
                                eprint("%s skipped" % am.filename)
                                failed += 1
                                break
                        except AceError:
                            eprint("%s failed to extract" % am.filename)
                            failed += 1
                            break
                    if f.is_solid() and failed > 0:
                        eprint("error extracting from solid archive, aborting")
                        sys.exit(1)
                    if args.verbose and am.comment:
                        eprint(asciibox(am.comment, title='file comment'))
                if failed > 0:
                    sys.exit(1)

            elif args.mode == 'test':
                if f.is_multivolume() and f.volume > 0 and f.is_solid():
                    eprint(("error: need complete set of volumes to test "
                            "solid multivolume archive"))
                    sys.exit(1)
                failed = 0
                ok = 0
                password = args.password
                for am in f:
                    if status:
                        status.member = am.filename
                    if f.is_solid() and failed > 0:
                        print("failure  %s" % am.filename)
                        failed += 1
                        continue
                    if am.is_enc() and password == None and not args.batch:
                        try:
                            password = getpass.getpass("%s password: " % \
                                                        am.filename)
                        except EOFError:
                            password = None
                    while True:
                        try:
                            if f.test(am, pwd=password):
                                print("success  %s" % am.filename)
                                ok += 1
                            else:
                                print("failure  %s" % am.filename)
                                failed += 1
                            break
                        except EncryptedArchiveError:
                            if args.batch or not password:
                                print("needpwd  %s" % am.filename)
                                failed += 1
                                break
                            eprint("last used password failed")
                            try:
                                password = getpass.getpass("%s password: " % \
                                                            am.filename)
                            except EOFError:
                                password = ''
                            if password == '':
                                password = args.password
                                print("needpwd  %s" % am.filename)
                                failed += 1
                                break
                    if args.verbose and am.comment:
                        eprint(asciibox(am.comment, title='file comment'))
                eprint("total %i tested, %i ok, %i failed" % (
                       ok + failed, ok, failed))
                if failed > 0:
                    sys.exit(1)

            elif args.mode == 'list':
                if args.verbose:
                    eprint(("CQD FES      size     packed   rel  "
                            "timestamp            filename"))
                    count = count_size = count_packsize = 0
                    for am in f:
                        if am.is_dir():
                            ft = 'd'
                        else:
                            ft = 'f'
                        if am.is_enc():
                            en = '+'
                        else:
                            en = ' '
                        if am.ntsecurity:
                            ns = 's'
                        else:
                            ns = ' '
                        if am.size > 0:
                            ratio = (100 * am.packsize) // am.size
                        else:
                            ratio = 100
                        print("%i%i%s %s%s%s %9i  %9i  %3i%%  %s  %s" % (
                            am.comptype, am.compqual,
                            hex(am.dicsizebits - 10)[2:],
                            ft, en, ns,
                            am.size,
                            am.packsize,
                            ratio,
                            am.datetime.strftime('%Y-%m-%d %H:%M:%S'),
                            am.filename))
                        if am.comment:
                            eprint(asciibox(am.comment, title='file comment'))
                        count_size += am.size
                        count_packsize += am.packsize
                        count += 1
                    eprint("total %i members, %i bytes, %i bytes compressed" % (
                           count, count_size, count_packsize))
                else:
                    for fn in f.getnames():
                        print("%s" % fn)

            elif args.mode == 'headers':
                f.dumpheaders()

            elif args.mode == 'selftest':
                eprint('dumpheaders():')
                f.dumpheaders()
                eprint('-' * 78)
                eprint('getnames():')
                for fn in f.getnames():
                    eprint("%s" % fn)
                eprint('-' * 78)
                eprint('testall():')
                rv = f.testall()
                if rv != None:
                    eprint("Test failed: member %s is corrupted" % rv)
                    sys.exit(1)
                eprint('-' * 78)
                eprint('test() in order:')
                for member in f:
                    if f.test(member):
                        eprint("%s: CRC OK" % member.filename)
                    else:
                        eprint("%s: CRC FAILED" % member.filename)
                        sys.exit(1)
                eprint('-' * 78)
                eprint('test() in reverse order:')
                for member in reversed(f.getmembers()):
                    if f.test(member):
                        eprint("%s: CRC OK" % member.filename)
                    else:
                        eprint("%s: CRC FAILED" % member.filename)
                        sys.exit(1)
            # end of with open

    except AceError as e:
        if DEBUG:
            raise
        eprint("%s: %s: %s" % (args.archive, type(e).__name__, e))
        sys.exit(1)

    sys.exit(0)



def testsuite():
    import doctest
    return doctest.DocTestSuite(optionflags=doctest.IGNORE_EXCEPTION_DETAIL)

def test():
    import doctest
    fails, tests = doctest.testmod(optionflags=doctest.IGNORE_EXCEPTION_DETAIL)
    sys.exit(min(1, fails))



if __name__ == '__main__':
    if '--doctest' in sys.argv:
        test()
    unace()