mirror of
https://github.com/Virtual-World-RE/NeoGF.git
synced 2024-11-15 10:45:33 +01:00
added pzztool
This commit is contained in:
commit
6bcf47ef54
233
pzztool.py
Normal file
233
pzztool.py
Normal file
|
@ -0,0 +1,233 @@
|
|||
#!/usr/bin/env python3
|
||||
__version__ = "1.0"
|
||||
__author__ = "rigodron, algoflash"
|
||||
__OriginalAutor__ = "infval"
|
||||
|
||||
from pathlib import Path
|
||||
from struct import unpack
|
||||
|
||||
|
||||
def pzz_decompress(b):
|
||||
bout = bytearray()
|
||||
size_b = len(b)
|
||||
|
||||
cb = 0 # Control bytes
|
||||
cb_bit = -1
|
||||
i = 0
|
||||
while i < size_b:
|
||||
if cb_bit < 0:
|
||||
cb = b[i + 1]
|
||||
cb |= b[i + 0] << 8
|
||||
cb_bit = 15
|
||||
i += 2
|
||||
continue
|
||||
|
||||
compress_flag = cb & (1 << cb_bit)
|
||||
cb_bit -= 1
|
||||
|
||||
print(compress_flag)
|
||||
if compress_flag:
|
||||
c = b[i + 1]
|
||||
c |= b[i + 0] << 8
|
||||
offset = (c & 0x7FF) * 2
|
||||
if offset == 0:
|
||||
break # End of the compressed data
|
||||
count = (c >> 11) * 2
|
||||
if count == 0:
|
||||
i += 2
|
||||
c = b[i + 1]
|
||||
c |= b[i + 0] << 8
|
||||
count = c * 2
|
||||
|
||||
index = len(bout) - offset
|
||||
for j in range(count):
|
||||
bout.append(bout[index + j])
|
||||
else:
|
||||
bout.extend(b[i: i + 2])
|
||||
i += 2
|
||||
|
||||
return bout
|
||||
|
||||
|
||||
def pzz_compress(b):
|
||||
bout = bytearray()
|
||||
size_b = len(b) // 2 * 2
|
||||
|
||||
cb = 0 # Control bytes
|
||||
cb_bit = 15
|
||||
cb_pos = 0
|
||||
bout.extend(b"\x00\x00")
|
||||
|
||||
i = 0
|
||||
while i < size_b:
|
||||
start = max(i - 0x3FF * 2, 0)
|
||||
count_r = 0
|
||||
max_i = -1
|
||||
tmp = b[i: i + 2]
|
||||
init_count = len(tmp)
|
||||
while True:
|
||||
start = b.find(tmp, start, i + 1)
|
||||
if start != -1 and start % 2 != 0:
|
||||
start += 1
|
||||
continue
|
||||
if start != -1:
|
||||
count = init_count
|
||||
while i < size_b - count \
|
||||
and count < 0xFFFF * 2 \
|
||||
and b[start + count ] == b[i + count ] \
|
||||
and b[start + count + 1] == b[i + count + 1]:
|
||||
count += 2
|
||||
if count_r < count:
|
||||
count_r = count
|
||||
max_i = start
|
||||
start += 2
|
||||
else:
|
||||
break
|
||||
start = max_i
|
||||
|
||||
compress_flag = 0
|
||||
if count_r >= 4:
|
||||
compress_flag = 1
|
||||
offset = i - start
|
||||
offset //= 2
|
||||
count_r //= 2
|
||||
c = offset
|
||||
if count_r <= 0x1F:
|
||||
c |= count_r << 11
|
||||
bout.append((c >> 8))
|
||||
bout.append(c & 0xFF)
|
||||
else:
|
||||
bout.append((c >> 8))
|
||||
bout.append(c & 0xFF)
|
||||
bout.append((count_r >> 8))
|
||||
bout.append(count_r & 0xFF)
|
||||
i += count_r * 2
|
||||
else:
|
||||
bout.extend(b[i: i + 2])
|
||||
i += 2
|
||||
cb |= (compress_flag << cb_bit)
|
||||
cb_bit -= 1
|
||||
if cb_bit < 0:
|
||||
bout[cb_pos + 1] = cb & 0xFF
|
||||
bout[cb_pos + 0] = cb >> 8
|
||||
cb = 0x0000
|
||||
cb_bit = 15
|
||||
cb_pos = len(bout)
|
||||
bout.extend(b"\x00\x00")
|
||||
|
||||
cb |= (1 << cb_bit)
|
||||
bout[cb_pos + 1] = cb & 0xFF
|
||||
bout[cb_pos + 0] = cb >> 8
|
||||
bout.extend(b"\x00\x00")
|
||||
|
||||
return bout
|
||||
|
||||
|
||||
def pzz_unpack(path, dir_path):
|
||||
""" BMS script: https://zenhax.com/viewtopic.php?f=9&t=8724&p=39437#p39437
|
||||
"""
|
||||
with open(path, "rb") as f:
|
||||
file_count = f.read(4)
|
||||
file_count, = unpack(">I", file_count) # < big-endian uint32
|
||||
|
||||
#file_count contient la taille en groupes de 4 octets
|
||||
|
||||
size = f.read(file_count * 4)
|
||||
# size contient l'ensemble des octets du fichier
|
||||
size = unpack(">{}I".format(file_count), size)
|
||||
# size contient le fichier par groupes d'uint32
|
||||
|
||||
print("File count:", file_count)
|
||||
|
||||
offset = 0x800
|
||||
for i, s in enumerate(size): # on a un ensemble d'uint32 et leur index qu'on parcours
|
||||
is_compressed = (s & 0x40000000) != 0 # "and" avec le bit du poids le plus fort
|
||||
#print(s)
|
||||
s &= 0x3FFFFFFF # s contient maintenant tous les bits sans le bit du poids le plus fort
|
||||
#print(s & 0x3FFFFFFF)
|
||||
s *= 0x800 # 1000 0000 0000
|
||||
# s doit contenir la taille du fichier
|
||||
#print( s )
|
||||
if s == 0:
|
||||
continue
|
||||
comp_str = ""
|
||||
if is_compressed:
|
||||
comp_str = "_compressed"
|
||||
filename = "{}_{:03}{}".format(Path(path).stem, i, comp_str)
|
||||
p = (Path(dir_path) / filename).with_suffix(".dat")
|
||||
|
||||
print("Offset: {:010} - {}".format(offset, p))
|
||||
|
||||
f.seek(offset)
|
||||
p.write_bytes(f.read(s))
|
||||
offset += s
|
||||
|
||||
|
||||
def get_argparser():
|
||||
import argparse
|
||||
parser = argparse.ArgumentParser(description='PZZ (de)compressor & unpacker - [GameCube] Gotcha Force v' + __version__)
|
||||
parser.add_argument('--version', action='version', version='%(prog)s ' + __version__)
|
||||
parser.add_argument('input_path', metavar='INPUT', help='only relative if -bu, -bc, -bd')
|
||||
parser.add_argument('output_path', metavar='OUTPUT', help='directory if -u, -bu, -bc, -bd')
|
||||
group = parser.add_mutually_exclusive_group(required=True)
|
||||
#group.add_argument('-p', '--pack', action='store_true')
|
||||
group.add_argument('-u', '--unpack', action='store_true', help='PZZ files from AFS')
|
||||
group.add_argument('-c', '--compress', action='store_true')
|
||||
group.add_argument('-d', '--decompress', action='store_true', help='Unpacked files from PZZ')
|
||||
group.add_argument('-bu', '--batch-unpack', action='store_true', help='INPUT relative pattern; e.g. AFS_DATA\\*.pzz')
|
||||
group.add_argument('-bc', '--batch-compress', action='store_true', help='INPUT relative pattern; e.g. AFS_DATA\\*.bin')
|
||||
group.add_argument('-bd', '--batch-decompress', action='store_true', help='INPUT relative pattern; e.g. AFS_DATA\\*_compressed.dat')
|
||||
return parser
|
||||
|
||||
|
||||
if __name__ == '__main__':
|
||||
import sys
|
||||
parser = get_argparser()
|
||||
args = parser.parse_args()
|
||||
|
||||
p_input = Path(args.input_path)
|
||||
p_output = Path(args.output_path)
|
||||
if args.compress:
|
||||
print("### Compress")
|
||||
p_output.write_bytes(pzz_compress(p_input.read_bytes()))
|
||||
elif args.decompress:
|
||||
print("### Decompress")
|
||||
p_output.write_bytes(pzz_decompress(p_input.read_bytes()))
|
||||
elif args.batch_compress:
|
||||
print("### Batch Compress")
|
||||
p_output.mkdir(exist_ok=True)
|
||||
|
||||
p = Path('.')
|
||||
for filename in p.glob(args.input_path):
|
||||
print(filename)
|
||||
b = filename.read_bytes()
|
||||
(p_output / filename.name).with_suffix(".dat").write_bytes(pzz_compress(b))
|
||||
elif args.batch_decompress:
|
||||
print("### Batch Decompress")
|
||||
p_output.mkdir(exist_ok=True)
|
||||
|
||||
p = Path('.')
|
||||
for filename in p.glob(args.input_path):
|
||||
print(filename)
|
||||
try:
|
||||
b = filename.read_bytes()
|
||||
(p_output / filename.name).with_suffix(".bin").write_bytes(pzz_decompress(b))
|
||||
except IndexError:
|
||||
print("! Wrong PZZ file")
|
||||
#elif args.pack:
|
||||
# pass
|
||||
elif args.unpack:
|
||||
print("### Unpack")
|
||||
p_output.mkdir(exist_ok=True)
|
||||
pzz_unpack(p_input, p_output)
|
||||
#elif args.batch_pack:
|
||||
# pass
|
||||
elif args.batch_unpack:
|
||||
print("### Batch Unpack")
|
||||
p_output.mkdir(exist_ok=True)
|
||||
|
||||
p = Path('.')
|
||||
for filename in p.glob(args.input_path):
|
||||
print(filename)
|
||||
pzz_unpack(filename, p_output)
|
||||
|
Loading…
Reference in New Issue
Block a user