[InfoExtractor] Misc yt-dlp back-ports, etc
* add _yes_playlist() method * avoid crash using _NETRC_MACHINE * use _search_json() in _search_nextjs_data() * _search_nextjs_data() default is JSON, not text * test for abovepull/29845/head
parent
768ccccd9b
commit
21924742f7
|
@ -153,6 +153,9 @@ class TestInfoExtractor(unittest.TestCase):
|
|||
'''
|
||||
search = self.ie._search_nextjs_data(html, 'testID')
|
||||
self.assertEqual(search['props']['pageProps']['video']['id'], 'testid')
|
||||
search = self.ie._search_nextjs_data(
|
||||
'no next.js data here, move along', 'testID', default={'status': 0})
|
||||
self.assertEqual(search['status'], 0)
|
||||
|
||||
def test_search_nuxt_data(self):
|
||||
html = '''
|
||||
|
|
|
@ -1169,10 +1169,10 @@ class InfoExtractor(object):
|
|||
def _get_netrc_login_info(self, netrc_machine=None):
|
||||
username = None
|
||||
password = None
|
||||
netrc_machine = netrc_machine or self._NETRC_MACHINE
|
||||
|
||||
if self._downloader.params.get('usenetrc', False):
|
||||
try:
|
||||
netrc_machine = netrc_machine or self._NETRC_MACHINE
|
||||
info = netrc.netrc().authenticators(netrc_machine)
|
||||
if info is not None:
|
||||
username = info[0]
|
||||
|
@ -1180,7 +1180,7 @@ class InfoExtractor(object):
|
|||
else:
|
||||
raise netrc.NetrcParseError(
|
||||
'No authenticators for %s' % netrc_machine)
|
||||
except (IOError, netrc.NetrcParseError) as err:
|
||||
except (AttributeError, IOError, netrc.NetrcParseError) as err:
|
||||
self._downloader.report_warning(
|
||||
'parsing .netrc: %s' % error_to_compat_str(err))
|
||||
|
||||
|
@ -1490,14 +1490,18 @@ class InfoExtractor(object):
|
|||
return dict((k, v) for k, v in info.items() if v is not None)
|
||||
|
||||
def _search_nextjs_data(self, webpage, video_id, **kw):
|
||||
nkw = dict((k, v) for k, v in kw.items() if k in ('transform_source', 'fatal'))
|
||||
kw.pop('transform_source', None)
|
||||
next_data = self._search_regex(
|
||||
r'''<script[^>]+\bid\s*=\s*('|")__NEXT_DATA__\1[^>]*>(?P<nd>[^<]+)</script>''',
|
||||
webpage, 'next.js data', group='nd', **kw)
|
||||
if not next_data:
|
||||
return {}
|
||||
return self._parse_json(next_data, video_id, **nkw)
|
||||
# ..., *, transform_source=None, fatal=True, default=NO_DEFAULT
|
||||
|
||||
# TODO: remove this backward compat
|
||||
default = kw.get('default', NO_DEFAULT)
|
||||
if default == '{}':
|
||||
kw['default'] = {}
|
||||
kw = compat_kwargs(kw)
|
||||
|
||||
return self._search_json(
|
||||
r'''<script\s[^>]*?\bid\s*=\s*('|")__NEXT_DATA__\1[^>]*>''',
|
||||
webpage, 'next.js data', video_id, end_pattern='</script>',
|
||||
**kw)
|
||||
|
||||
def _search_nuxt_data(self, webpage, video_id, *args, **kwargs):
|
||||
"""Parses Nuxt.js metadata. This works as long as the function __NUXT__ invokes is a pure function"""
|
||||
|
@ -3296,12 +3300,16 @@ class InfoExtractor(object):
|
|||
return ret
|
||||
|
||||
@classmethod
|
||||
def _merge_subtitles(cls, subtitle_dict1, subtitle_dict2):
|
||||
""" Merge two subtitle dictionaries, language by language. """
|
||||
ret = dict(subtitle_dict1)
|
||||
for lang in subtitle_dict2:
|
||||
ret[lang] = cls._merge_subtitle_items(subtitle_dict1.get(lang, []), subtitle_dict2[lang])
|
||||
return ret
|
||||
def _merge_subtitles(cls, subtitle_dict1, *subtitle_dicts, **kwargs):
|
||||
""" Merge subtitle dictionaries, language by language. """
|
||||
|
||||
# ..., * , target=None
|
||||
target = kwargs.get('target') or dict(subtitle_dict1)
|
||||
|
||||
for subtitle_dict in subtitle_dicts:
|
||||
for lang in subtitle_dict:
|
||||
target[lang] = cls._merge_subtitle_items(target.get(lang, []), subtitle_dict[lang])
|
||||
return target
|
||||
|
||||
def extract_automatic_captions(self, *args, **kwargs):
|
||||
if (self._downloader.params.get('writeautomaticsub', False)
|
||||
|
@ -3334,6 +3342,29 @@ class InfoExtractor(object):
|
|||
def _generic_title(self, url):
|
||||
return compat_urllib_parse_unquote(os.path.splitext(url_basename(url))[0])
|
||||
|
||||
def _yes_playlist(self, playlist_id, video_id, *args, **kwargs):
|
||||
# smuggled_data=None, *, playlist_label='playlist', video_label='video'
|
||||
smuggled_data = args[0] if len(args) == 1 else kwargs.get('smuggled_data')
|
||||
playlist_label = kwargs.get('playlist_label', 'playlist')
|
||||
video_label = kwargs.get('video_label', 'video')
|
||||
|
||||
if not playlist_id or not video_id:
|
||||
return not video_id
|
||||
|
||||
no_playlist = (smuggled_data or {}).get('force_noplaylist')
|
||||
if no_playlist is not None:
|
||||
return not no_playlist
|
||||
|
||||
video_id = '' if video_id is True else ' ' + video_id
|
||||
noplaylist = self.get_param('noplaylist')
|
||||
self.to_screen(
|
||||
'Downloading just the {0}{1} because of --no-playlist'.format(video_label, video_id)
|
||||
if noplaylist else
|
||||
'Downloading {0}{1} - add --no-playlist to download just the {2}{3}'.format(
|
||||
playlist_label, '' if playlist_id is True else ' ' + playlist_id,
|
||||
video_label, video_id))
|
||||
return not noplaylist
|
||||
|
||||
|
||||
class SearchInfoExtractor(InfoExtractor):
|
||||
"""
|
||||
|
|
Loading…
Reference in New Issue