Compare commits
3 Commits
master
...
serializer
Author | SHA1 | Date | |
---|---|---|---|
1a0c63872c | |||
ba738e395d | |||
12ea8c4d1b |
|
@ -1,20 +1,21 @@
|
||||||
import struct
|
import struct
|
||||||
|
|
||||||
TYPE_CHAR = b"b"
|
TYPE_CHAR = b"b", 1
|
||||||
TYPE_BYTE = b"B"
|
TYPE_BYTE = b"B", 1
|
||||||
TYPE_BOOL = b"?"
|
TYPE_BOOL = b"?", 1
|
||||||
TYPE_SHORT = b"h"
|
TYPE_SHORT = b"h", 2
|
||||||
TYPE_USHORT = b"H"
|
TYPE_USHORT = b"H", 2
|
||||||
TYPE_INT = b"i"
|
TYPE_INT = b"i", 4
|
||||||
TYPE_UINT = b"I"
|
TYPE_UINT = b"I", 4
|
||||||
TYPE_LONG = b"l"
|
TYPE_LONG = b"l", 4
|
||||||
TYPE_ULONG = b"L"
|
TYPE_ULONG = b"L", 4
|
||||||
TYPE_LLONG = b"q"
|
TYPE_LLONG = b"q", 8
|
||||||
TYPE_ULLONG = b"Q"
|
TYPE_ULLONG = b"Q", 8
|
||||||
TYPE_FLOAT = b"f"
|
TYPE_FLOAT = b"f", 4
|
||||||
TYPE_DOUBLE = b"d"
|
TYPE_DOUBLE = b"d", 8
|
||||||
TYPE_STRING = b"B%ds"
|
TYPE_STRING = b"s", 1
|
||||||
TYPE_BYTESTREAM = b"%ds"
|
|
||||||
|
TYPE_ULEB = b"s", 1, True # True just so they are different tuples
|
||||||
|
|
||||||
BYTEORDER_NATIVE = b"@"
|
BYTEORDER_NATIVE = b"@"
|
||||||
BYTEORDER_LITTLE = b"<"
|
BYTEORDER_LITTLE = b"<"
|
||||||
|
@ -22,7 +23,6 @@ BYTEORDER_BIG = b">"
|
||||||
|
|
||||||
def encode_uleb(val):
|
def encode_uleb(val):
|
||||||
data = b""
|
data = b""
|
||||||
pos = 0
|
|
||||||
while val != 0:
|
while val != 0:
|
||||||
b = val & 0x7f
|
b = val & 0x7f
|
||||||
val >>= 7
|
val >>= 7
|
||||||
|
@ -73,10 +73,18 @@ def guess_type(value):
|
||||||
|
|
||||||
raise Exception("Unable to guess type")
|
raise Exception("Unable to guess type")
|
||||||
|
|
||||||
class Serializable:
|
def read(value_type, stream):
|
||||||
post_serialized = None
|
#if value_type == TYPE_STRING:
|
||||||
|
# read()
|
||||||
|
|
||||||
def __init__(self, value, type = None, post = None):
|
struct.unpack(value_type[0], stream[:value_type[1]])
|
||||||
|
|
||||||
|
class Serializable:
|
||||||
|
length = None
|
||||||
|
prefix = b""
|
||||||
|
filter = None
|
||||||
|
|
||||||
|
def __init__(self, value, type = None, length = None, prefix = b"", filter = None):
|
||||||
self.value = value
|
self.value = value
|
||||||
|
|
||||||
if type is not None:
|
if type is not None:
|
||||||
|
@ -84,29 +92,76 @@ class Serializable:
|
||||||
else:
|
else:
|
||||||
self.type = guess_type(type)
|
self.type = guess_type(type)
|
||||||
|
|
||||||
self.post_serialized = post
|
self.length = length
|
||||||
|
self.prefix = prefix
|
||||||
|
self.filter = filter
|
||||||
|
|
||||||
def get_struct_fmt(self):
|
def get_struct_fmt(self, length = 1, uleb_length_override = -1):
|
||||||
if self.type in [TYPE_STRING, TYPE_BYTESTREAM]:
|
struct_fmt = self.type[0]
|
||||||
return self.type % len(self.get_value()[-1])
|
if length > 1:
|
||||||
return self.type
|
struct_fmt = b"%d%b" % (length, self.type[0])
|
||||||
|
if self.length is not None:
|
||||||
|
if self.length == TYPE_ULEB:
|
||||||
|
struct_fmt = b"%d%b%b" % (len(self.get_value()[0]) if uleb_length_override == -1 else uleb_length_override, self.length[0], struct_fmt)
|
||||||
|
else:
|
||||||
|
struct_fmt = b"%b%b" % (self.length[0], struct_fmt)
|
||||||
|
return struct_fmt
|
||||||
|
|
||||||
def get_value(self):
|
def get_value(self):
|
||||||
def _get_value():
|
val = [self.value]
|
||||||
if self.type == TYPE_STRING:
|
if self.filter is not None:
|
||||||
return [0x0b, encode_uleb(len(self.value)) + self.value]
|
val[-1] = self.filter(val[-1], 0)
|
||||||
return [self.value]
|
if self.length is not None:
|
||||||
|
if self.length == TYPE_ULEB:
|
||||||
val = _get_value()
|
val.insert(0, encode_uleb(len(val[-1])))
|
||||||
if self.post_serialized is not None:
|
else:
|
||||||
val[-1] = self.post_serialized(val[-1])
|
val.insert(0, len(val[-1]))
|
||||||
|
|
||||||
return val
|
return val
|
||||||
|
|
||||||
def pack(self, byte_order = BYTEORDER_LITTLE):
|
def pack(self, byte_order = BYTEORDER_LITTLE):
|
||||||
struct_fmt = byte_order + self.get_struct_fmt()
|
length = 1
|
||||||
#print(struct_fmt, self.get_value())
|
if self.type == TYPE_STRING:
|
||||||
return struct.pack(struct_fmt, *self.get_value())
|
length = len(self.get_value()[-1])
|
||||||
|
struct_fmt = byte_order + self.get_struct_fmt(length)
|
||||||
|
packed = struct.pack(struct_fmt, *self.get_value())
|
||||||
|
if self.prefix is not None:
|
||||||
|
packed = self.prefix + packed
|
||||||
|
return packed
|
||||||
|
|
||||||
|
def unpack(self, stream, byte_order = BYTEORDER_LITTLE):
|
||||||
|
length = 1
|
||||||
|
read_pos = 0
|
||||||
|
extra_read = 0
|
||||||
|
remove_struct_fmt = []
|
||||||
|
if self.type == TYPE_STRING:
|
||||||
|
if self.length == TYPE_ULEB:
|
||||||
|
extra_read = 1
|
||||||
|
read_pos += 1 # Skip 0x0b
|
||||||
|
length, _read_pos = decode_uleb(stream[read_pos:])
|
||||||
|
read_pos += _read_pos - 1
|
||||||
|
else:
|
||||||
|
l = Serializable(0, self.length)
|
||||||
|
read_pos = l.unpack(stream)
|
||||||
|
length = l.get_value()[-1]
|
||||||
|
remove_struct_fmt.append(0)
|
||||||
|
|
||||||
|
|
||||||
|
struct_fmt = byte_order + self.get_struct_fmt(length, read_pos)
|
||||||
|
length += extra_read
|
||||||
|
|
||||||
|
if len(remove_struct_fmt):
|
||||||
|
for rem in remove_struct_fmt:
|
||||||
|
struct_fmt = struct_fmt[:rem + 1] + struct_fmt[rem + 2:]
|
||||||
|
|
||||||
|
data = struct.unpack(struct_fmt, stream[read_pos:read_pos + self.type[1] * length])
|
||||||
|
|
||||||
|
if self.filter is not None:
|
||||||
|
data = list(data)
|
||||||
|
data[-1] = self.filter(data[-1], 1)
|
||||||
|
|
||||||
|
self.value = data[-1]
|
||||||
|
|
||||||
|
return read_pos + self.type[1] * length
|
||||||
|
|
||||||
class Serializer:
|
class Serializer:
|
||||||
def __init__(self, byte_order = BYTEORDER_LITTLE):
|
def __init__(self, byte_order = BYTEORDER_LITTLE):
|
||||||
|
@ -119,6 +174,13 @@ class Serializer:
|
||||||
else:
|
else:
|
||||||
self.stack.append(serializable)
|
self.stack.append(serializable)
|
||||||
|
|
||||||
|
def serialize(self):
|
||||||
|
return b"".join(s.pack() for s in self.stack)
|
||||||
|
|
||||||
|
def deserialize(self, stream):
|
||||||
|
for s in self.stack:
|
||||||
|
stream = stream[s.unpack(stream):]
|
||||||
|
|
||||||
def flush(self):
|
def flush(self):
|
||||||
return b"".join(s.pack() for s in self.stack)
|
return b"".join(s.pack() for s in self.stack)
|
||||||
|
|
||||||
|
|
|
@ -1,6 +1,6 @@
|
||||||
import struct
|
import struct
|
||||||
import string
|
import string
|
||||||
#import lzma
|
import lzma
|
||||||
import pylzma # Required cause pythons standard library lzma has missing support
|
import pylzma # Required cause pythons standard library lzma has missing support
|
||||||
|
|
||||||
from .frame import ReplayFrame
|
from .frame import ReplayFrame
|
||||||
|
@ -18,10 +18,8 @@ from hashlib import md5 as _md5
|
||||||
import traceback
|
import traceback
|
||||||
|
|
||||||
def md5_str(data):
|
def md5_str(data):
|
||||||
print(data)
|
|
||||||
if type(data) is str:
|
if type(data) is str:
|
||||||
data = data.encode("ascii")
|
data = data.encode("ascii")
|
||||||
print(_md5(data).hexdigest().encode())
|
|
||||||
return _md5(data).hexdigest().encode()
|
return _md5(data).hexdigest().encode()
|
||||||
|
|
||||||
def md5_file(file):
|
def md5_file(file):
|
||||||
|
@ -36,20 +34,30 @@ def md5_file(file):
|
||||||
hash.update(chunk)
|
hash.update(chunk)
|
||||||
return hash.hexdigest()
|
return hash.hexdigest()
|
||||||
|
|
||||||
def append_and_compress(data):
|
def append_and_compress(data, mode):
|
||||||
|
if mode == 0:
|
||||||
return lzma_compress(data + b"-12345|0|0|1337,")
|
return lzma_compress(data + b"-12345|0|0|1337,")
|
||||||
|
else:
|
||||||
|
return lzma_decompress(data)[:-len(b"-12345|0|0|1337,")]
|
||||||
|
|
||||||
def lzma_compress(data):
|
def lzma_compress(data):
|
||||||
comp = pylzma.compress(data, dictionary = 21, fastBytes = 255, eos = False)
|
comp = pylzma.compress(data, dictionary = 21, fastBytes = 255, eos = False)
|
||||||
comp = comp[:5] + struct.pack(b"<Q", len(data)) + comp[5:] # Append uncompressed length
|
comp = comp[:5] + struct.pack(b"<Q", len(data)) + comp[5:] # Append uncompressed length
|
||||||
return struct.pack("<I", len(comp)) + comp # For some odd ass reason this doesnt use ULEB
|
return comp
|
||||||
|
|
||||||
|
def lzma_decompress(data):
|
||||||
|
comp = lzma.decompress(data)
|
||||||
|
return comp
|
||||||
|
|
||||||
class Replay:
|
class Replay:
|
||||||
_mode = Serializable(osuModes.STANDARD, TYPE_BYTE)
|
_mode = Serializable(osuModes.STANDARD, TYPE_BYTE)
|
||||||
_osu_version = Serializable(20181216, TYPE_INT)
|
_osu_version = Serializable(20181216, TYPE_INT)
|
||||||
_beatmap_hash = Serializable(b"d41d8cd98f00b204e9800998ecf8427e", TYPE_STRING)
|
#_beatmap_hash_length = Serializable(32, TYPE_ULEB)
|
||||||
_player_name = Serializable(b"osu!", TYPE_STRING)
|
_beatmap_hash = Serializable(b"d41d8cd98f00b204e9800998ecf8427e", TYPE_STRING, length = TYPE_ULEB, prefix = b"\x0b")
|
||||||
_score_hash = Serializable(b"d41d8cd98f00b204e9800998ecf8427e", TYPE_STRING)
|
#_player_name_length = Serializable(4, TYPE_ULEB)
|
||||||
|
_player_name = Serializable(b"osu!", TYPE_STRING, length = TYPE_ULEB, prefix = b"\x0b")
|
||||||
|
#_score_hash_length = Serializable(32, TYPE_ULEB)
|
||||||
|
_score_hash = Serializable(b"d41d8cd98f00b204e9800998ecf8427e", TYPE_STRING, length = TYPE_ULEB, prefix = b"\x0b")
|
||||||
|
|
||||||
_score_300s = Serializable(0, TYPE_USHORT)
|
_score_300s = Serializable(0, TYPE_USHORT)
|
||||||
_score_100s = Serializable(0, TYPE_USHORT)
|
_score_100s = Serializable(0, TYPE_USHORT)
|
||||||
|
@ -63,10 +71,12 @@ class Replay:
|
||||||
|
|
||||||
_mods = Serializable(osuMods.NOMOD, TYPE_INT)
|
_mods = Serializable(osuMods.NOMOD, TYPE_INT)
|
||||||
|
|
||||||
_lifebar_graph = Serializable(b"0|1,", TYPE_STRING)
|
#_lifebar_graph_length = Serializable(4, TYPE_ULEB)
|
||||||
|
_lifebar_graph = Serializable(b"0|1,", TYPE_STRING, length = TYPE_ULEB, prefix = b"\x0b")
|
||||||
_timestamp = Serializable(0, TYPE_ULLONG)
|
_timestamp = Serializable(0, TYPE_ULLONG)
|
||||||
|
|
||||||
_replay_data = Serializable(b"", TYPE_BYTESTREAM, post = append_and_compress)
|
#_replay_data_length = Serializable(0, TYPE_UINT)
|
||||||
|
_replay_data = Serializable(b"", TYPE_STRING, length = TYPE_UINT, filter = append_and_compress)
|
||||||
|
|
||||||
_online_score_id = Serializable(0, TYPE_ULLONG)
|
_online_score_id = Serializable(0, TYPE_ULLONG)
|
||||||
|
|
||||||
|
@ -245,4 +255,17 @@ class Replay:
|
||||||
serializer.add( self.__getattribute__(attrib) ) # Add value to serializer
|
serializer.add( self.__getattribute__(attrib) ) # Add value to serializer
|
||||||
|
|
||||||
with open(filename, "wb") as f:
|
with open(filename, "wb") as f:
|
||||||
f.write( serializer.flush() )
|
f.write( serializer.serialize() )
|
||||||
|
|
||||||
|
def load(self, filename):
|
||||||
|
serializer = Serializer()
|
||||||
|
|
||||||
|
attribs = [ a for a in self.__dir__() if not a.startswith("__") ]
|
||||||
|
for attrib in attribs:
|
||||||
|
if attrib == "_end_of_attributes":
|
||||||
|
break
|
||||||
|
|
||||||
|
serializer.add( self.__getattribute__(attrib) ) # Add value to serializer
|
||||||
|
|
||||||
|
with open(filename, "rb") as f:
|
||||||
|
serializer.deserialize( f.read() )
|
||||||
|
|
15
test.py
15
test.py
|
@ -3,17 +3,8 @@ from osuRepy.replay import Replay # Replay instance
|
||||||
|
|
||||||
from osuRepy.helpers import osuButtons, osuMods # Enum helpers
|
from osuRepy.helpers import osuButtons, osuMods # Enum helpers
|
||||||
|
|
||||||
replay = Replay(beatmap_hash = "9e66c0a0eadced7d07f06b3968a74cc0") # Create replay instance
|
replay = Replay() # Create replay instance
|
||||||
|
|
||||||
replay.write( ReplayFrame(2, 256, 192, osuButtons.M1) ) # Add replay frame (time, x, y, buttons)
|
replay.load("replay.osr")
|
||||||
|
|
||||||
replay.set_score(score = 29284624,
|
replay.save("test.osr")
|
||||||
s300 = 416,
|
|
||||||
s50 = 2,
|
|
||||||
miss = 1,
|
|
||||||
combo = 358) # Set score info
|
|
||||||
|
|
||||||
replay.set_mods(osuMods.HIDDEN | osuMods.DOUBLETIME) # Enable HDDT mods
|
|
||||||
replay.set_timestamp(636880715111611126, False)
|
|
||||||
|
|
||||||
replay.save("myReplay.osr") # Export replay to file
|
|
Loading…
Reference in New Issue
Block a user