Telethon/telethon_generator/generators/errors.py

53 lines
1.7 KiB
Python
Raw Normal View History

import itertools
def generate_errors(errors, f):
# Exact/regex match to create {CODE: ErrorClassName}
exact_match = []
regex_match = []
# Find out what subclasses to import and which to create
import_base, create_base = set(), {}
for error in errors:
if error.subclass_exists:
import_base.add(error.subclass)
else:
create_base[error.subclass] = error.int_code
if error.has_captures:
regex_match.append(error)
else:
exact_match.append(error)
# Imports and new subclass creation
f.write('from .rpc_base_errors import RPCError, {}\n'
.format(", ".join(sorted(import_base))))
for cls, int_code in sorted(create_base.items(), key=lambda t: t[1]):
f.write('\n\nclass {}(RPCError):\n code = {}\n'
.format(cls, int_code))
# Error classes generation
for error in errors:
f.write('\n\nclass {}({}):\n def __init__(self, **kwargs):\n'
' '.format(error.name, error.subclass))
if error.has_captures:
f.write("self.{} = int(kwargs.get('capture', 0))\n "
.format(error.capture_name))
f.write('super(Exception, self).__init__({}'
.format(repr(error.description)))
if error.has_captures:
f.write('.format(self.{})'.format(error.capture_name))
f.write(')\n')
# Create the actual {CODE: ErrorClassName} dict once classes are defined
# TODO Actually make a difference between regex/exact
f.write('\n\nrpc_errors_all = {\n')
for error in itertools.chain(regex_match, exact_match):
f.write(' {}: {},\n'.format(repr(error.pattern), error.name))
f.write('}\n')