Telethon/telethon_generator/parsers/errors.py
2020-12-04 20:43:47 +01:00

85 lines
2.9 KiB
Python

import csv
import re
from ..utils import snake_to_camel_case
# Core base classes depending on the integer error code
KNOWN_BASE_CLASSES = {
303: 'InvalidDCError',
400: 'BadRequestError',
401: 'UnauthorizedError',
403: 'ForbiddenError',
404: 'NotFoundError',
406: 'AuthKeyError',
420: 'FloodError',
500: 'ServerError',
503: 'TimedOutError'
}
def _get_class_name(error_code):
"""
Gets the corresponding class name for the given error code,
this either being an integer (thus base error name) or str.
"""
if isinstance(error_code, int):
return KNOWN_BASE_CLASSES.get(
abs(error_code), 'RPCError' + str(error_code).replace('-', 'Neg')
)
if error_code.startswith('2'):
error_code = re.sub(r'2', 'TWO_', error_code, count=1)
if re.match(r'\d+', error_code):
raise RuntimeError('error code starting with a digit cannot have valid Python name: {}'.format(error_code))
return snake_to_camel_case(
error_code.replace('FIRSTNAME', 'FIRST_NAME')\
.replace('SLOWMODE', 'SLOW_MODE').lower(), suffix='Error')
class Error:
def __init__(self, codes, name, description):
# TODO Some errors have the same name but different integer codes
# Should these be split into different files or doesn't really matter?
# Telegram isn't exactly consistent with returned errors anyway.
self.int_code = codes[0]
self.str_code = name
self.subclass = _get_class_name(codes[0])
self.subclass_exists = abs(codes[0]) in KNOWN_BASE_CLASSES
self.description = description
self.has_captures = '_X' in name
if self.has_captures:
self.name = _get_class_name(name.replace('_X', ''))
self.pattern = name.replace('_X', r'_(\d+)')
self.capture_name = re.search(r'{(\w+)}', description).group(1)
else:
self.name = _get_class_name(name)
self.pattern = name
self.capture_name = None
def parse_errors(csv_file):
"""
Parses the input CSV file with columns (name, error codes, description)
and yields `Error` instances as a result.
"""
with csv_file.open(newline='') as f:
f = csv.reader(f)
next(f, None) # header
for line, tup in enumerate(f, start=2):
try:
name, codes, description = tup
except ValueError:
raise ValueError('Columns count mismatch, unquoted comma in '
'desc? (line {})'.format(line)) from None
try:
codes = [int(x) for x in codes.split()] or [400]
except ValueError:
raise ValueError('Not all codes are integers '
'(line {})'.format(line)) from None
yield Error([int(x) for x in codes], name, description)