1
0
mirror of synced 2024-11-27 23:50:47 +01:00
bemaniutils/bemani/format/ifs.py

250 lines
10 KiB
Python

import hashlib
import io
import os
import struct
from PIL import Image # type: ignore
from typing import Dict, List, Tuple
from bemani.protocol.binary import BinaryEncoding
from bemani.protocol.xml import XmlEncoding
from bemani.protocol.lz77 import Lz77
from bemani.protocol.node import Node
class IFS:
"""
Best-effort utility for decoding the `.ifs` file format. There are better tools out
there, but this was developed before their existence. This should work with most of
the games out there including non-rhythm games that use this format.
"""
def __init__(self, data: bytes, decode_binxml: bool=False, decode_textures: bool=False) -> None:
self.__files: Dict[str, bytes] = {}
self.__formats: Dict[str, str] = {}
self.__compressed: Dict[str, bool] = {}
self.__imgsize: Dict[str, Tuple[int, int, int, int]] = {}
self.__uvsize: Dict[str, Tuple[int, int, int, int]] = {}
self.__decode_binxml = decode_binxml
self.__decode_textures = decode_textures
self.__parse_file(data)
def __fix_name(self, filename: str) -> str:
if filename[0] == '_' and filename[1].isdigit():
filename = filename[1:]
filename = filename.replace('_E', '.')
filename = filename.replace('__', '_')
return filename
def __parse_file(self, data: bytes) -> None:
# Grab the magic values and make sure this is an IFS
(signature, version, version_crc, pack_time, unpacked_header_size, data_index) = struct.unpack(
'>IHHIII',
data[0:20],
)
if signature != 0x6CAD8F89:
raise Exception('Invalid IFS file!')
if version ^ version_crc != 0xFFFF:
raise Exception('Corrupt version in IFS file!')
if version == 1:
# No header MD5
header_offset = 20
else:
# Make room for header MD5, at byte offset 20-36
header_offset = 36
# First, try as binary
benc = BinaryEncoding()
header = benc.decode(data[header_offset:data_index])
if header is None:
# Now, try as XML
xenc = XmlEncoding()
header = xenc.decode(
b'<?xml encoding="ascii"?>' +
data[header_offset:data_index].split(b'\0')[0]
)
if header is None:
raise Exception('Invalid IFS file!')
files: Dict[str, Tuple[int, int, int]] = {}
if header.name != 'imgfs':
raise Exception('Unknown IFS format!')
def get_children(parent: str, node: Node) -> None:
real_name = self.__fix_name(node.name)
if node.data_type == '3s32':
node_name = os.path.join(parent, real_name).replace(f'{os.sep}imgfs{os.sep}', '')
files[node_name] = (node.value[0] + data_index, node.value[1], node.value[2])
else:
for subchild in node.children:
get_children(os.path.join(parent, f"{real_name}{os.sep}"), subchild)
# Recursively walk the entire filesystem extracting files and their locations.
get_children(os.sep, header)
for fn in files:
(start, size, pack_time) = files[fn]
filedata = data[start:(start + size)]
self.__files[fn] = filedata
# Now, find all of the index files that are available.
for filename in list(self.__files.keys()):
abs_filename = (os.sep if filename.startswith(os.sep) else "") + filename
if abs_filename.endswith(f"{os.sep}texturelist.xml"):
# This is a texture index.
texdir = os.path.dirname(filename)
benc = BinaryEncoding()
texdata = benc.decode(self.__files[filename])
if texdata.name != 'texturelist':
raise Exception(f"Unexpected name {texdata.name} in texture list!")
if texdata.attribute('compress') == 'avslz':
compressed = True
else:
compressed = False
for child in texdata.children:
if child.name != 'texture':
continue
textfmt = child.attribute('format')
for subchild in child.children:
if subchild.name != 'image':
continue
md5sum = hashlib.md5(subchild.attribute('name').encode(benc.encoding)).hexdigest()
oldname = os.path.join(texdir, md5sum)
newname = os.path.join(texdir, subchild.attribute('name'))
if oldname in self.__files:
supported = False
if self.__decode_textures:
if textfmt in ["argb8888rev"]:
# This is a supported file to decode
newname += ".png"
supported = True
# Remove old index, update file to new index.
self.__files[newname] = self.__files[oldname]
del self.__files[oldname]
# Remember the attributes for this file so we can extract it later.
self.__compressed[newname] = compressed
if supported:
# Only pop down the format and sizes if we support extracting.
self.__formats[newname] = textfmt
rect = subchild.child_value('imgrect')
if rect is not None:
self.__imgsize[newname] = (
rect[0] // 2,
rect[1] // 2,
rect[2] // 2,
rect[3] // 2,
)
rect = subchild.child_value('uvrect')
if rect is not None:
self.__uvsize[newname] = (
rect[0] // 2,
rect[1] // 2,
rect[2] // 2,
rect[3] // 2,
)
elif abs_filename.endswith(f"{os.sep}afplist.xml"):
# This is a texture index.
afpdir = os.path.dirname(filename)
bsidir = os.path.join(afpdir, "bsi")
geodir = os.path.join(os.path.dirname(afpdir), "geo")
benc = BinaryEncoding()
afpdata = benc.decode(self.__files[filename])
if afpdata.name != 'afplist':
raise Exception(f"Unexpected name {afpdata.name} in afp list!")
for child in afpdata.children:
if child.name != 'afp':
continue
# First, fix up the afp files themselves.
name = child.attribute('name')
md5sum = hashlib.md5(name.encode(benc.encoding)).hexdigest()
for fixdir in [afpdir, bsidir]:
oldname = os.path.join(fixdir, md5sum)
newname = os.path.join(fixdir, name)
if oldname in self.__files:
# Remove old index, update file to new index.
self.__files[newname] = self.__files[oldname]
del self.__files[oldname]
# Now, fix up the shape files as well.
geodata = child.child_value("geo")
if geodata is not None:
for geoid in geodata:
geoname = f"{name}_shape{geoid}"
md5sum = hashlib.md5(geoname.encode(benc.encoding)).hexdigest()
oldname = os.path.join(geodir, md5sum)
newname = os.path.join(geodir, geoname)
if oldname in self.__files:
# Remove old index, update file to new index.
self.__files[newname] = self.__files[oldname]
del self.__files[oldname]
@property
def filenames(self) -> List[str]:
return [f for f in self.__files]
def read_file(self, filename: str) -> bytes:
# First, figure out if this file is stored compressed or not. If it is, decompress
# it so that we have the raw data available to us.
decompress = self.__compressed.get(filename, False)
filedata = self.__files[filename]
if decompress:
uncompressed_size, compressed_size = struct.unpack('>II', filedata[0:8])
if len(filedata) == compressed_size + 8:
lz77 = Lz77()
filedata = lz77.decompress(filedata[8:])
else:
raise Exception('Unrecognized compression!')
if self.__decode_binxml and os.path.splitext(filename)[1] == '.xml':
benc = BinaryEncoding()
filexml = benc.decode(filedata)
if filexml is not None:
filedata = str(filexml).encode('utf-8')
if self.__decode_textures and filename in self.__formats and filename in self.__imgsize and filename in self.__uvsize:
fmt = self.__formats[filename]
img = self.__imgsize[filename]
crop = self.__uvsize[filename]
# Decode the image data itself.
if fmt == "argb8888rev":
width = img[1] - img[0]
height = img[3] - img[2]
if len(filedata) < (width * height * 4):
left = (width * height * 4) - len(filedata)
filedata = filedata + b'\x00' * left
png = Image.frombytes('RGBA', (width, height), filedata, 'raw', 'BGRA')
png = png.crop((
crop[0] - img[0],
crop[2] - img[2],
crop[1] - img[0],
crop[3] - img[2],
))
b = io.BytesIO()
png.save(b, format='PNG')
filedata = b.getvalue()
return filedata