Telethon/telethon_generator/parsers/tlobject.py

291 lines
10 KiB
Python

import re
from zlib import crc32
from ..utils import snake_to_camel_case
CORE_TYPES = (
0xbc799737, # boolFalse#bc799737 = Bool;
0x997275b5, # boolTrue#997275b5 = Bool;
0x3fedd339, # true#3fedd339 = True;
0x1cb5c415, # vector#1cb5c415 {t:Type} # [ t ] = Vector t;
)
# https://github.com/telegramdesktop/tdesktop/blob/4bf66cb6e93f3965b40084771b595e93d0b11bcd/Telegram/SourceFiles/codegen/scheme/codegen_scheme.py#L57-L62
WHITELISTED_MISMATCHING_IDS = {
# 0 represents any layer
0: {'ipPortSecret', 'accessPointRule', 'help.configSimple'},
77: {'channel'},
78: {'channel'},
79: {'channel'},
80: {'channel'},
}
class TLObject:
def __init__(self, fullname, object_id, args, result, is_function, layer):
"""
Initializes a new TLObject, given its properties.
:param fullname: The fullname of the TL object (namespace.name)
The namespace can be omitted.
:param object_id: The hexadecimal string representing the object ID
:param args: The arguments, if any, of the TL object
:param result: The result type of the TL object
:param is_function: Is the object a function or a type?
:param layer: The layer this TLObject belongs to.
"""
# The name can or not have a namespace
self.fullname = fullname
if '.' in fullname:
self.namespace, self.name = fullname.split('.', maxsplit=1)
else:
self.namespace, self.name = None, fullname
self.args = args
self.result = result
self.is_function = is_function
self.id = None
if object_id is None:
self.id = self.infer_id()
else:
self.id = int(object_id, base=16)
whitelist = WHITELISTED_MISMATCHING_IDS[0] |\
WHITELISTED_MISMATCHING_IDS.get(layer, set())
if self.fullname not in whitelist:
assert self.id == self.infer_id(),\
'Invalid inferred ID for ' + repr(self)
self.class_name = snake_to_camel_case(
self.name, suffix='Request' if self.is_function else '')
self.real_args = list(a for a in self.sorted_args() if not
(a.flag_indicator or a.generic_definition))
def sorted_args(self):
"""Returns the arguments properly sorted and ready to plug-in
into a Python's method header (i.e., flags and those which
can be inferred will go last so they can default =None)
"""
return sorted(self.args,
key=lambda x: x.is_flag or x.can_be_inferred)
def __repr__(self, ignore_id=False):
if self.id is None or ignore_id:
hex_id = ''
else:
hex_id = '#{:08x}'.format(self.id)
if self.args:
args = ' ' + ' '.join([repr(arg) for arg in self.args])
else:
args = ''
return '{}{}{} = {}'.format(self.fullname, hex_id, args, self.result)
def infer_id(self):
representation = self.__repr__(ignore_id=True)
representation = representation\
.replace(':bytes ', ':string ')\
.replace('?bytes ', '?string ')\
.replace('<', ' ').replace('>', '')\
.replace('{', '').replace('}', '')
representation = re.sub(
r' \w+:flags\.\d+\?true',
r'',
representation
)
return crc32(representation.encode('ascii'))
class TLArg:
def __init__(self, name, arg_type, generic_definition):
"""
Initializes a new .tl argument
:param name: The name of the .tl argument
:param arg_type: The type of the .tl argument
:param generic_definition: Is the argument a generic definition?
(i.e. {X:Type})
"""
self.name = 'is_self' if name == 'self' else name
# Default values
self.is_vector = False
self.is_flag = False
self.skip_constructor_id = False
self.flag_index = -1
# Special case: some types can be inferred, which makes it
# less annoying to type. Currently the only type that can
# be inferred is if the name is 'random_id', to which a
# random ID will be assigned if left as None (the default)
self.can_be_inferred = name == 'random_id'
# The type can be an indicator that other arguments will be flags
if arg_type == '#':
self.flag_indicator = True
self.type = None
self.is_generic = False
else:
self.flag_indicator = False
self.is_generic = arg_type.startswith('!')
# Strip the exclamation mark always to have only the name
self.type = arg_type.lstrip('!')
# The type may be a flag (flags.IDX?REAL_TYPE)
# Note that 'flags' is NOT the flags name; this
# is determined by a previous argument
# However, we assume that the argument will always be called 'flags'
flag_match = re.match(r'flags.(\d+)\?([\w<>.]+)', self.type)
if flag_match:
self.is_flag = True
self.flag_index = int(flag_match.group(1))
# Update the type to match the exact type, not the "flagged" one
self.type = flag_match.group(2)
# Then check if the type is a Vector<REAL_TYPE>
vector_match = re.match(r'[Vv]ector<([\w\d.]+)>', self.type)
if vector_match:
self.is_vector = True
# If the type's first letter is not uppercase, then
# it is a constructor and we use (read/write) its ID
# as pinpointed on issue #81.
self.use_vector_id = self.type[0] == 'V'
# Update the type to match the one inside the vector
self.type = vector_match.group(1)
# See use_vector_id. An example of such case is ipPort in
# help.configSpecial
if self.type.split('.')[-1][0].islower():
self.skip_constructor_id = True
# The name may contain "date" in it, if this is the case and the type is "int",
# we can safely assume that this should be treated as a "date" object.
# Note that this is not a valid Telegram object, but it's easier to work with
if self.type == 'int' and (
re.search(r'(\b|_)date\b', name) or
name in ('expires', 'expires_at', 'was_online')):
self.type = 'date'
self.generic_definition = generic_definition
def type_hint(self):
type = self.type
if '.' in type:
type = type.split('.')[1]
result = {
'int': 'int',
'long': 'int',
'int128': 'int',
'int256': 'int',
'string': 'str',
'date': 'Optional[datetime]', # None date = 0 timestamp
'bytes': 'bytes',
'true': 'bool',
}.get(type, "Type{}".format(type))
if self.is_vector:
result = 'List[{}]'.format(result)
if self.is_flag and type != 'date':
result = 'Optional[{}]'.format(result)
return result
def __str__(self):
# Find the real type representation by updating it as required
real_type = self.type
if self.flag_indicator:
real_type = '#'
if self.is_vector:
if self.use_vector_id:
real_type = 'Vector<{}>'.format(real_type)
else:
real_type = 'vector<{}>'.format(real_type)
if self.is_generic:
real_type = '!{}'.format(real_type)
if self.is_flag:
real_type = 'flags.{}?{}'.format(self.flag_index, real_type)
if self.generic_definition:
return '{{{}:{}}}'.format(self.name, real_type)
else:
return '{}:{}'.format(self.name, real_type)
def __repr__(self):
return str(self).replace(':date', ':int').replace('?date', '?int')
def _from_line(line, is_function, layer):
match = re.match(
r'^([\w.]+)' # 'name'
r'(?:#([0-9a-fA-F]+))?' # '#optionalcode'
r'(?:\s{?\w+:[\w\d<>#.?!]+}?)*' # '{args:.0?type}'
r'\s=\s' # ' = '
r'([\w\d<>#.?]+);$', # '<result.type>;'
line
)
if match is None:
# Probably "vector#1cb5c415 {t:Type} # [ t ] = Vector t;"
raise ValueError('Cannot parse TLObject {}'.format(line))
args_match = re.findall(
r'({)?'
r'(\w+)'
r':'
r'([\w\d<>#.?!]+)'
r'}?',
line
)
return TLObject(
fullname=match.group(1),
object_id=match.group(2),
result=match.group(3),
is_function=is_function,
layer=layer,
args=[TLArg(name, arg_type, brace != '')
for brace, name, arg_type in args_match]
)
def parse_tl(file_path, layer, ignore_core=False):
"""This method yields TLObjects from a given .tl file."""
with open(file_path, encoding='utf-8') as file:
is_function = False
for line in file:
comment_index = line.find('//')
if comment_index != -1:
line = line[:comment_index]
line = line.strip()
if not line:
continue
match = re.match('---(\w+)---', line)
if match:
following_types = match.group(1)
is_function = following_types == 'functions'
continue
try:
result = _from_line(line, is_function, layer=layer)
if not ignore_core or result.id not in CORE_TYPES:
yield result
except ValueError as e:
if 'vector#1cb5c415' not in str(e):
raise
def find_layer(file_path):
"""Finds the layer used on the specified scheme.tl file."""
layer_regex = re.compile(r'^//\s*LAYER\s*(\d+)$')
with open(file_path, encoding='utf-8') as file:
for line in file:
match = layer_regex.match(line)
if match:
return int(match.group(1))