mirror of
https://github.com/SickGear/SickGear.git
synced 2024-11-26 22:53:38 +00:00
251 lines
9.3 KiB
Python
251 lines
9.3 KiB
Python
|
"""
|
||
|
BZIP2 archive file
|
||
|
|
||
|
Author: Victor Stinner, Robert Xiao
|
||
|
"""
|
||
|
|
||
|
from hachoir.parser import Parser
|
||
|
from hachoir.core.tools import paddingSize
|
||
|
from hachoir.field import (Field, FieldSet, GenericVector,
|
||
|
ParserError, String,
|
||
|
PaddingBits, Bit, Bits, Character,
|
||
|
UInt32, Enum, CompressedField)
|
||
|
from hachoir.core.endian import BIG_ENDIAN
|
||
|
from hachoir.core.text_handler import textHandler, hexadecimal
|
||
|
from hachoir.parser.archive.zlib import build_tree, HuffmanCode
|
||
|
|
||
|
try:
|
||
|
from bz2 import BZ2Decompressor
|
||
|
|
||
|
class Bunzip2:
|
||
|
|
||
|
def __init__(self, stream):
|
||
|
self.bzip2 = BZ2Decompressor()
|
||
|
|
||
|
def __call__(self, size, data=''):
|
||
|
try:
|
||
|
return self.bzip2.decompress(data)
|
||
|
except EOFError:
|
||
|
return ''
|
||
|
|
||
|
has_deflate = True
|
||
|
except ImportError:
|
||
|
has_deflate = False
|
||
|
|
||
|
|
||
|
class ZeroTerminatedNumber(Field):
|
||
|
"""Zero (bit) terminated number: e.g. 11110 is 4."""
|
||
|
|
||
|
def __init__(self, parent, name, description=None):
|
||
|
Field.__init__(self, parent, name, 0, description)
|
||
|
|
||
|
endian = self.parent.endian
|
||
|
stream = self.parent.stream
|
||
|
addr = self.absolute_address
|
||
|
|
||
|
value = 0
|
||
|
while True:
|
||
|
bit = stream.readBits(addr, 1, endian)
|
||
|
addr += 1
|
||
|
self._size += 1
|
||
|
if not bit:
|
||
|
break
|
||
|
value += 1
|
||
|
self._value = value
|
||
|
|
||
|
def createValue(self):
|
||
|
return self._value
|
||
|
|
||
|
|
||
|
def move_to_front(seq, index):
|
||
|
seq[:] = seq[index:index + 1] + seq[0:index] + seq[index + 1:]
|
||
|
|
||
|
|
||
|
class Bzip2Bitmap(FieldSet):
|
||
|
|
||
|
def __init__(self, parent, name, nb_items, start_index, *args, **kwargs):
|
||
|
FieldSet.__init__(self, parent, name, *args, **kwargs)
|
||
|
self.nb_items = nb_items
|
||
|
self.start_index = start_index
|
||
|
|
||
|
def createFields(self):
|
||
|
for i in range(self.start_index, self.start_index + self.nb_items):
|
||
|
yield Bit(self, "symbol_used[%i]" % i, "Is the symbol %i (%r) used?" % (i, chr(i)))
|
||
|
|
||
|
|
||
|
class Bzip2Lengths(FieldSet):
|
||
|
|
||
|
def __init__(self, parent, name, symbols, *args, **kwargs):
|
||
|
FieldSet.__init__(self, parent, name, *args, **kwargs)
|
||
|
self.symbols = symbols
|
||
|
|
||
|
def createFields(self):
|
||
|
yield Bits(self, "start_length", 5)
|
||
|
length = self["start_length"].value
|
||
|
lengths = []
|
||
|
for i in range(self.symbols):
|
||
|
while True:
|
||
|
bit = Bit(
|
||
|
self, "change_length[%i][]" % i, "Should the length be changed for symbol %i?" % i)
|
||
|
yield bit
|
||
|
if not bit.value:
|
||
|
break
|
||
|
else:
|
||
|
bit = Enum(Bit(self, "length_decrement[%i][]" % i, "Decrement the value?"), {
|
||
|
True: "Decrement", False: "Increment"})
|
||
|
yield bit
|
||
|
if bit.value:
|
||
|
length -= 1
|
||
|
else:
|
||
|
length += 1
|
||
|
lengths.append(length)
|
||
|
self.final_length = length
|
||
|
self.tree = build_tree(lengths)
|
||
|
|
||
|
|
||
|
class Bzip2Selectors(FieldSet):
|
||
|
|
||
|
def __init__(self, parent, name, ngroups, *args, **kwargs):
|
||
|
FieldSet.__init__(self, parent, name, *args, **kwargs)
|
||
|
self.groups = list(range(ngroups))
|
||
|
|
||
|
def createFields(self):
|
||
|
for i in range(self["../selectors_used"].value):
|
||
|
field = ZeroTerminatedNumber(self, "selector_list[]")
|
||
|
move_to_front(self.groups, field.value)
|
||
|
field.realvalue = self.groups[0]
|
||
|
field._description = "MTF'ed selector index: raw value %i, real value %i" % (
|
||
|
field.value, field.realvalue)
|
||
|
yield field
|
||
|
|
||
|
|
||
|
class Bzip2Block(FieldSet):
|
||
|
|
||
|
def createFields(self):
|
||
|
yield textHandler(Bits(self, "blockheader", 48, "Block header"), hexadecimal)
|
||
|
if self["blockheader"].value != 0x314159265359: # pi
|
||
|
raise ParserError("Invalid block header!")
|
||
|
yield textHandler(UInt32(self, "crc32", "CRC32 for this block"), hexadecimal)
|
||
|
yield Bit(self, "randomized", "Is this block randomized?")
|
||
|
yield Bits(self, "orig_bwt_pointer", 24, "Starting pointer into BWT after untransform")
|
||
|
yield GenericVector(self, "huffman_used_map", 16, Bit, 'block_used', "Bitmap showing which blocks (representing 16 literals each) are in use")
|
||
|
symbols_used = []
|
||
|
for index, block_used in enumerate(self["huffman_used_map"].array('block_used')):
|
||
|
if block_used.value:
|
||
|
start_index = index * 16
|
||
|
field = Bzip2Bitmap(self, "huffman_used_bitmap[%i]" % index, 16, start_index, "Bitmap for block %i (literals %i to %i) showing which symbols are in use" % (
|
||
|
index, start_index, start_index + 15))
|
||
|
yield field
|
||
|
for i, used in enumerate(field):
|
||
|
if used.value:
|
||
|
symbols_used.append(start_index + i)
|
||
|
yield Bits(self, "huffman_groups", 3, "Number of different Huffman tables in use")
|
||
|
yield Bits(self, "selectors_used", 15, "Number of times the Huffman tables are switched")
|
||
|
yield Bzip2Selectors(self, "selectors_list", self["huffman_groups"].value)
|
||
|
trees = []
|
||
|
for group in range(self["huffman_groups"].value):
|
||
|
field = Bzip2Lengths(self, "huffman_lengths[]",
|
||
|
len(symbols_used) + 2)
|
||
|
yield field
|
||
|
trees.append(field.tree)
|
||
|
counter = 0
|
||
|
rle_run = 0
|
||
|
selector_tree = None
|
||
|
while True:
|
||
|
if counter % 50 == 0:
|
||
|
select_id = self["selectors_list"].array(
|
||
|
"selector_list")[counter // 50].realvalue
|
||
|
selector_tree = trees[select_id]
|
||
|
field = HuffmanCode(self, "huffman_code[]", selector_tree)
|
||
|
if field.realvalue in [0, 1]:
|
||
|
# RLE codes
|
||
|
if rle_run == 0:
|
||
|
rle_power = 1
|
||
|
rle_run += (field.realvalue + 1) * rle_power
|
||
|
rle_power <<= 1
|
||
|
field._description = "RLE Run Code %i (for %r); Total accumulated run %i (Huffman Code %i)" % (
|
||
|
field.realvalue, chr(symbols_used[0]), rle_run, field.value)
|
||
|
elif field.realvalue == len(symbols_used) + 1:
|
||
|
field._description = "Block Terminator (%i) (Huffman Code %i)" % (
|
||
|
field.realvalue, field.value)
|
||
|
yield field
|
||
|
break
|
||
|
else:
|
||
|
rle_run = 0
|
||
|
move_to_front(symbols_used, field.realvalue - 1)
|
||
|
field._description = "Literal %r (value %i) (Huffman Code %i)" % (
|
||
|
chr(symbols_used[0]), field.realvalue, field.value)
|
||
|
yield field
|
||
|
if field.realvalue == len(symbols_used) + 1:
|
||
|
break
|
||
|
counter += 1
|
||
|
|
||
|
|
||
|
class Bzip2Stream(FieldSet):
|
||
|
START_BLOCK = 0x314159265359 # pi
|
||
|
END_STREAM = 0x177245385090 # sqrt(pi)
|
||
|
|
||
|
def createFields(self):
|
||
|
end = False
|
||
|
while not end:
|
||
|
marker = self.stream.readBits(
|
||
|
self.absolute_address + self.current_size, 48, self.endian)
|
||
|
if marker == self.START_BLOCK:
|
||
|
yield Bzip2Block(self, "block[]")
|
||
|
elif marker == self.END_STREAM:
|
||
|
yield textHandler(Bits(self, "stream_end", 48, "End-of-stream marker"), hexadecimal)
|
||
|
yield textHandler(UInt32(self, "crc32", "CRC32 for entire stream"), hexadecimal)
|
||
|
padding = paddingSize(self.current_size, 8)
|
||
|
if padding:
|
||
|
yield PaddingBits(self, "padding[]", padding)
|
||
|
end = True
|
||
|
else:
|
||
|
raise ParserError("Invalid marker 0x%02X!" % marker)
|
||
|
|
||
|
|
||
|
class Bzip2Parser(Parser):
|
||
|
PARSER_TAGS = {
|
||
|
"id": "bzip2",
|
||
|
"category": "archive",
|
||
|
"file_ext": ("bz2",),
|
||
|
"mime": ("application/x-bzip2",),
|
||
|
"min_size": 10 * 8,
|
||
|
"magic": ((b'BZh', 0),),
|
||
|
"description": "bzip2 archive"
|
||
|
}
|
||
|
endian = BIG_ENDIAN
|
||
|
|
||
|
def validate(self):
|
||
|
if self.stream.readBytes(0, 3) != b'BZh':
|
||
|
return "Wrong file signature"
|
||
|
if not("1" <= self["blocksize"].value <= "9"):
|
||
|
return "Wrong blocksize"
|
||
|
return True
|
||
|
|
||
|
def createFields(self):
|
||
|
yield String(self, "id", 3, "Identifier (BZh)", charset="ASCII")
|
||
|
yield Character(self, "blocksize", "Block size (KB of memory needed to uncompress)")
|
||
|
|
||
|
if self._size is None: # TODO: is it possible to handle piped input?
|
||
|
raise NotImplementedError
|
||
|
|
||
|
size = (self._size - self.current_size) // 8
|
||
|
if size:
|
||
|
for tag, filename in self.stream.tags:
|
||
|
if tag == "filename" and filename.endswith(".bz2"):
|
||
|
filename = filename[:-4]
|
||
|
break
|
||
|
else:
|
||
|
filename = None
|
||
|
data = Bzip2Stream(self, "file", size=size * 8)
|
||
|
if has_deflate:
|
||
|
CompressedField(self, Bunzip2)
|
||
|
|
||
|
def createInputStream(**args):
|
||
|
if filename:
|
||
|
args.setdefault("tags", []).append(
|
||
|
("filename", filename))
|
||
|
return self._createInputStream(**args)
|
||
|
data._createInputStream = createInputStream
|
||
|
yield data
|