mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-14 21:49:39 +00:00
1a830394a2
Authored by: Grub4K
114 lines
3.8 KiB
Python
114 lines
3.8 KiB
Python
#!/usr/bin/env python3
|
|
|
|
# Allow direct execution
|
|
import os
|
|
import sys
|
|
|
|
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
|
|
|
|
|
from inspect import getsource
|
|
|
|
from devscripts.utils import get_filename_args, read_file, write_file
|
|
|
|
NO_ATTR = object()
|
|
STATIC_CLASS_PROPERTIES = [
|
|
'IE_NAME', '_ENABLED', '_VALID_URL', # Used for URL matching
|
|
'_WORKING', 'IE_DESC', '_NETRC_MACHINE', 'SEARCH_KEY', # Used for --extractor-descriptions
|
|
'age_limit', # Used for --age-limit (evaluated)
|
|
'_RETURN_TYPE', # Accessed in CLI only with instance (evaluated)
|
|
]
|
|
CLASS_METHODS = [
|
|
'ie_key', 'suitable', '_match_valid_url', # Used for URL matching
|
|
'working', 'get_temp_id', '_match_id', # Accessed just before instance creation
|
|
'description', # Used for --extractor-descriptions
|
|
'is_suitable', # Used for --age-limit
|
|
'supports_login', 'is_single_video', # Accessed in CLI only with instance
|
|
]
|
|
IE_TEMPLATE = '''
|
|
class {name}({bases}):
|
|
_module = {module!r}
|
|
'''
|
|
MODULE_TEMPLATE = read_file('devscripts/lazy_load_template.py')
|
|
|
|
|
|
def main():
|
|
os.environ['YTDLP_NO_PLUGINS'] = 'true'
|
|
os.environ['YTDLP_NO_LAZY_EXTRACTORS'] = 'true'
|
|
|
|
lazy_extractors_filename = get_filename_args(default_outfile='yt_dlp/extractor/lazy_extractors.py')
|
|
|
|
from yt_dlp.extractor.extractors import _ALL_CLASSES
|
|
from yt_dlp.extractor.common import InfoExtractor, SearchInfoExtractor
|
|
|
|
DummyInfoExtractor = type('InfoExtractor', (InfoExtractor,), {'IE_NAME': NO_ATTR})
|
|
module_src = '\n'.join((
|
|
MODULE_TEMPLATE,
|
|
' _module = None',
|
|
*extra_ie_code(DummyInfoExtractor),
|
|
'\nclass LazyLoadSearchExtractor(LazyLoadExtractor):\n pass\n',
|
|
*build_ies(_ALL_CLASSES, (InfoExtractor, SearchInfoExtractor), DummyInfoExtractor),
|
|
))
|
|
|
|
write_file(lazy_extractors_filename, f'{module_src}\n')
|
|
|
|
|
|
def extra_ie_code(ie, base=None):
|
|
for var in STATIC_CLASS_PROPERTIES:
|
|
val = getattr(ie, var)
|
|
if val != (getattr(base, var) if base else NO_ATTR):
|
|
yield f' {var} = {val!r}'
|
|
yield ''
|
|
|
|
for name in CLASS_METHODS:
|
|
f = getattr(ie, name)
|
|
if not base or f.__func__ != getattr(base, name).__func__:
|
|
yield getsource(f)
|
|
|
|
|
|
def build_ies(ies, bases, attr_base):
|
|
names = []
|
|
for ie in sort_ies(ies, bases):
|
|
yield build_lazy_ie(ie, ie.__name__, attr_base)
|
|
if ie in ies:
|
|
names.append(ie.__name__)
|
|
|
|
yield f'\n_ALL_CLASSES = [{", ".join(names)}]'
|
|
|
|
|
|
def sort_ies(ies, ignored_bases):
|
|
"""find the correct sorting and add the required base classes so that subclasses can be correctly created"""
|
|
classes, returned_classes = ies[:-1], set()
|
|
assert ies[-1].__name__ == 'GenericIE', 'Last IE must be GenericIE'
|
|
while classes:
|
|
for c in classes[:]:
|
|
bases = set(c.__bases__) - {object, *ignored_bases}
|
|
restart = False
|
|
for b in sorted(bases, key=lambda x: x.__name__):
|
|
if b not in classes and b not in returned_classes:
|
|
assert b.__name__ != 'GenericIE', 'Cannot inherit from GenericIE'
|
|
classes.insert(0, b)
|
|
restart = True
|
|
if restart:
|
|
break
|
|
if bases <= returned_classes:
|
|
yield c
|
|
returned_classes.add(c)
|
|
classes.remove(c)
|
|
break
|
|
yield ies[-1]
|
|
|
|
|
|
def build_lazy_ie(ie, name, attr_base):
|
|
bases = ', '.join({
|
|
'InfoExtractor': 'LazyLoadExtractor',
|
|
'SearchInfoExtractor': 'LazyLoadSearchExtractor',
|
|
}.get(base.__name__, base.__name__) for base in ie.__bases__)
|
|
|
|
s = IE_TEMPLATE.format(name=name, module=ie.__module__, bases=bases)
|
|
return s + '\n'.join(extra_ie_code(ie, attr_base))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
main()
|