1
0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-05 02:32:44 +01:00
yt-dlp/devscripts/make_lazy_extractors.py
2021-08-23 05:31:55 +05:30

115 lines
3.3 KiB
Python

#!/usr/bin/env python3
from __future__ import unicode_literals, print_function
from inspect import getsource
import io
import os
from os.path import dirname as dirn
import sys
print('WARNING: Lazy loading extractors is an experimental feature that may not always work', file=sys.stderr)
sys.path.insert(0, dirn(dirn((os.path.abspath(__file__)))))
lazy_extractors_filename = sys.argv[1]
if os.path.exists(lazy_extractors_filename):
os.remove(lazy_extractors_filename)
# Block plugins from loading
plugins_dirname = 'ytdlp_plugins'
plugins_blocked_dirname = 'ytdlp_plugins_blocked'
if os.path.exists(plugins_dirname):
os.rename(plugins_dirname, plugins_blocked_dirname)
from yt_dlp.extractor import _ALL_CLASSES
from yt_dlp.extractor.common import InfoExtractor, SearchInfoExtractor
if os.path.exists(plugins_blocked_dirname):
os.rename(plugins_blocked_dirname, plugins_dirname)
with open('devscripts/lazy_load_template.py', 'rt') as f:
module_template = f.read()
CLASS_PROPERTIES = ['ie_key', 'working', '_match_valid_url', 'suitable', '_match_id', 'get_temp_id']
module_contents = [
module_template,
*[getsource(getattr(InfoExtractor, k)) for k in CLASS_PROPERTIES],
'\nclass LazyLoadSearchExtractor(LazyLoadExtractor):\n pass\n']
ie_template = '''
class {name}({bases}):
_module = '{module}'
'''
make_valid_template = '''
@classmethod
def _make_valid_url(cls):
return {valid_url!r}
'''
def get_base_name(base):
if base is InfoExtractor:
return 'LazyLoadExtractor'
elif base is SearchInfoExtractor:
return 'LazyLoadSearchExtractor'
else:
return base.__name__
def build_lazy_ie(ie, name):
s = ie_template.format(
name=name,
bases=', '.join(map(get_base_name, ie.__bases__)),
module=ie.__module__)
valid_url = getattr(ie, '_VALID_URL', None)
if valid_url:
s += f' _VALID_URL = {valid_url!r}\n'
if not ie._WORKING:
s += ' _WORKING = False\n'
if ie.suitable.__func__ is not InfoExtractor.suitable.__func__:
s += f'\n{getsource(ie.suitable)}'
if hasattr(ie, '_make_valid_url'):
# search extractors
s += make_valid_template.format(valid_url=ie._make_valid_url())
return s
# find the correct sorting and add the required base classes so that subclasses
# can be correctly created
classes = _ALL_CLASSES[:-1]
ordered_cls = []
while classes:
for c in classes[:]:
bases = set(c.__bases__) - set((object, InfoExtractor, SearchInfoExtractor))
stop = False
for b in bases:
if b not in classes and b not in ordered_cls:
if b.__name__ == 'GenericIE':
exit()
classes.insert(0, b)
stop = True
if stop:
break
if all(b in ordered_cls for b in bases):
ordered_cls.append(c)
classes.remove(c)
break
ordered_cls.append(_ALL_CLASSES[-1])
names = []
for ie in ordered_cls:
name = ie.__name__
src = build_lazy_ie(ie, name)
module_contents.append(src)
if ie in _ALL_CLASSES:
names.append(name)
module_contents.append(
'\n_ALL_CLASSES = [{0}]'.format(', '.join(names)))
module_src = '\n'.join(module_contents) + '\n'
with io.open(lazy_extractors_filename, 'wt', encoding='utf-8') as f:
f.write(module_src)