mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-02 17:22:31 +01:00
82d020804d
and refactor lazy extractors accordingly. This reduces the need to create extractor instances
29 lines
1.0 KiB
Python
29 lines
1.0 KiB
Python
import importlib
|
|
import random
|
|
import re
|
|
|
|
from ..utils import bug_reports_message, classproperty, write_string
|
|
|
|
|
|
class LazyLoadMetaClass(type):
|
|
def __getattr__(cls, name):
|
|
# "is_suitable" requires "_TESTS". However, they bloat the lazy_extractors
|
|
if '_real_class' not in cls.__dict__ and name not in ('is_suitable', 'get_testcases'):
|
|
write_string(
|
|
'WARNING: Falling back to normal extractor since lazy extractor '
|
|
f'{cls.__name__} does not have attribute {name}{bug_reports_message()}\n')
|
|
return getattr(cls.real_class, name)
|
|
|
|
|
|
class LazyLoadExtractor(metaclass=LazyLoadMetaClass):
|
|
@classproperty
|
|
def real_class(cls):
|
|
if '_real_class' not in cls.__dict__:
|
|
cls._real_class = getattr(importlib.import_module(cls._module), cls.__name__)
|
|
return cls._real_class
|
|
|
|
def __new__(cls, *args, **kwargs):
|
|
instance = cls.real_class.__new__(cls.real_class)
|
|
instance.__init__(*args, **kwargs)
|
|
return instance
|