[ZDF] Overhaul ZDF extractors

* pull some yt-dlp changes into ZDFBaseIE._extract_format()
* add test cases from yt-dlp to ZDFIE
* fix crash in ZDFIE._extract_mobile() when object had no `formitaeten`
* improve title extraction in ZDFChannelIE (remove trailing station ident)
* avoid extracting non-video playlist items (fixes #31149)
pull/20291/merge
dirkf 2022-08-09 19:37:58 +01:00
parent 2ced5a7912
commit c282e5f8d7
1 changed files with 97 additions and 72 deletions

View File

@ -8,13 +8,14 @@ from ..compat import compat_str
from ..utils import ( from ..utils import (
determine_ext, determine_ext,
ExtractorError, ExtractorError,
extract_attributes,
float_or_none, float_or_none,
int_or_none, int_or_none,
merge_dicts, merge_dicts,
NO_DEFAULT, NO_DEFAULT,
orderedSet,
parse_codecs, parse_codecs,
qualities, qualities,
str_or_none,
try_get, try_get,
unified_timestamp, unified_timestamp,
update_url_query, update_url_query,
@ -57,28 +58,39 @@ class ZDFBaseIE(InfoExtractor):
format_urls.add(format_url) format_urls.add(format_url)
mime_type = meta.get('mimeType') mime_type = meta.get('mimeType')
ext = determine_ext(format_url) ext = determine_ext(format_url)
join_nonempty = lambda s, l: s.join(filter(None, l))
meta_map = lambda t: map(lambda x: str_or_none(meta.get(x)), t)
if mime_type == 'application/x-mpegURL' or ext == 'm3u8': if mime_type == 'application/x-mpegURL' or ext == 'm3u8':
formats.extend(self._extract_m3u8_formats( new_formats = self._extract_m3u8_formats(
format_url, video_id, 'mp4', m3u8_id='hls', format_url, video_id, 'mp4', m3u8_id='hls',
entry_protocol='m3u8_native', fatal=False)) entry_protocol='m3u8_native', fatal=False)
elif mime_type == 'application/f4m+xml' or ext == 'f4m': elif mime_type == 'application/f4m+xml' or ext == 'f4m':
formats.extend(self._extract_f4m_formats( new_formats = self._extract_f4m_formats(
update_url_query(format_url, {'hdcore': '3.7.0'}), video_id, f4m_id='hds', fatal=False)) update_url_query(format_url, {'hdcore': '3.7.0'}), video_id, f4m_id='hds', fatal=False)
else: else:
f = parse_codecs(meta.get('mimeCodec')) f = parse_codecs(meta.get('mimeCodec'))
if not f:
data = meta.get('type', '').split('_')
if try_get(data, lambda x: x[2]) == ext:
f = dict(zip(('vcodec', 'acodec'), data[1]))
format_id = ['http'] format_id = ['http']
for p in (meta.get('type'), meta.get('quality')): format_id.extend(join_nonempty('-', meta_map(('type', 'quality'))))
if p and isinstance(p, compat_str):
format_id.append(p)
f.update({ f.update({
'url': format_url, 'url': format_url,
'format_id': '-'.join(format_id), 'format_id': '-'.join(format_id),
'format_note': meta.get('quality'), 'tbr': int_or_none(self._search_regex(r'_(\d+)k_', format_url, 'tbr', default=None))
'language': meta.get('language'),
'quality': qualities(self._QUALITIES)(meta.get('quality')),
'preference': -10,
}) })
formats.append(f) new_formats = [f]
formats.extend(merge_dicts(f, {
'format_note': join_nonempty(',', meta_map(('quality', 'class'))),
'language': meta.get('language'),
'language_preference': 10 if meta.get('class') == 'main' else -10 if meta.get('class') == 'ad' else -1,
'quality': qualities(self._QUALITIES)(meta.get('quality')),
}) for f in new_formats)
def _extract_ptmd(self, ptmd_url, video_id, api_token, referrer): def _extract_ptmd(self, ptmd_url, video_id, api_token, referrer):
ptmd = self._call_api( ptmd = self._call_api(
@ -107,6 +119,7 @@ class ZDFBaseIE(InfoExtractor):
'type': f.get('type'), 'type': f.get('type'),
'mimeType': f.get('mimeType'), 'mimeType': f.get('mimeType'),
'quality': quality.get('quality'), 'quality': quality.get('quality'),
'class': track.get('class'),
'language': track.get('language'), 'language': track.get('language'),
}) })
self._sort_formats(formats) self._sort_formats(formats)
@ -171,6 +184,20 @@ class ZDFIE(ZDFBaseIE):
'duration': 2615, 'duration': 2615,
'timestamp': 1465021200, 'timestamp': 1465021200,
'upload_date': '20160604', 'upload_date': '20160604',
'thumbnail': 'https://www.zdf.de/assets/mauve-im-labor-100~768x432?cb=1464909117806',
},
}, {
'url': 'https://www.zdf.de/funk/druck-11790/funk-alles-ist-verzaubert-102.html',
'md5': '1b93bdec7d02fc0b703c5e7687461628',
'info_dict': {
'ext': 'mp4',
'id': 'video_funk_1770473',
'duration': 1278,
'description': 'Die Neue an der Schule verdreht Ismail den Kopf.',
'title': 'Alles ist verzaubert',
'timestamp': 1635520560,
'upload_date': '20211029',
'thumbnail': 'https://www.zdf.de/assets/teaser-funk-alles-ist-verzaubert-100~1920x1080?cb=1636466431799',
}, },
}, { }, {
# Same as https://www.phoenix.de/sendungen/dokumentationen/gesten-der-maechtigen-i-a-89468.html?ref=suche # Same as https://www.phoenix.de/sendungen/dokumentationen/gesten-der-maechtigen-i-a-89468.html?ref=suche
@ -204,6 +231,19 @@ class ZDFIE(ZDFBaseIE):
'timestamp': 1641355200, 'timestamp': 1641355200,
'upload_date': '20220105', 'upload_date': '20220105',
}, },
'skip': 'No longer available "Diese Seite wurde leider nicht gefunden"'
}, {
'url': 'https://www.zdf.de/serien/soko-stuttgart/das-geld-anderer-leute-100.html',
'info_dict': {
'id': '191205_1800_sendung_sok8',
'ext': 'mp4',
'title': 'Das Geld anderer Leute',
'description': 'md5:cb6f660850dc5eb7d1ab776ea094959d',
'duration': 2581.0,
'timestamp': 1654790700,
'upload_date': '20220609',
'thumbnail': 'https://epg-image.zdf.de/fotobase-webdelivery/images/e2d7e55a-09f0-424e-ac73-6cac4dd65f35?layout=2400x1350',
},
}] }]
def _extract_entry(self, url, player, content, video_id): def _extract_entry(self, url, player, content, video_id):
@ -265,14 +305,15 @@ class ZDFIE(ZDFBaseIE):
'https://zdf-cdn.live.cellular.de/mediathekV2/document/%s' % video_id, 'https://zdf-cdn.live.cellular.de/mediathekV2/document/%s' % video_id,
video_id) video_id)
document = video['document'] formats = []
formitaeten = try_get(video, lambda x: x['document']['formitaeten'], list)
document = formitaeten and video['document']
if formitaeten:
title = document['titel'] title = document['titel']
content_id = document['basename'] content_id = document['basename']
formats = []
format_urls = set() format_urls = set()
for f in document['formitaeten']: for f in formitaeten or []:
self._extract_format(content_id, formats, format_urls, f) self._extract_format(content_id, formats, format_urls, f)
self._sort_formats(formats) self._sort_formats(formats)
@ -320,9 +361,9 @@ class ZDFChannelIE(ZDFBaseIE):
'url': 'https://www.zdf.de/sport/das-aktuelle-sportstudio', 'url': 'https://www.zdf.de/sport/das-aktuelle-sportstudio',
'info_dict': { 'info_dict': {
'id': 'das-aktuelle-sportstudio', 'id': 'das-aktuelle-sportstudio',
'title': 'das aktuelle sportstudio | ZDF', 'title': 'das aktuelle sportstudio',
}, },
'playlist_mincount': 23, 'playlist_mincount': 18,
}, { }, {
'url': 'https://www.zdf.de/dokumentation/planet-e', 'url': 'https://www.zdf.de/dokumentation/planet-e',
'info_dict': { 'info_dict': {
@ -330,6 +371,14 @@ class ZDFChannelIE(ZDFBaseIE):
'title': 'planet e.', 'title': 'planet e.',
}, },
'playlist_mincount': 50, 'playlist_mincount': 50,
}, {
'url': 'https://www.zdf.de/gesellschaft/aktenzeichen-xy-ungeloest',
'info_dict': {
'id': 'aktenzeichen-xy-ungeloest',
'title': 'Aktenzeichen XY... ungelöst',
'entries': "lambda x: not any('xy580-fall1-kindermoerder-gesucht-100' in e['url'] for e in x)",
},
'playlist_mincount': 2,
}, { }, {
'url': 'https://www.zdf.de/filme/taunuskrimi/', 'url': 'https://www.zdf.de/filme/taunuskrimi/',
'only_matching': True, 'only_matching': True,
@ -339,60 +388,36 @@ class ZDFChannelIE(ZDFBaseIE):
def suitable(cls, url): def suitable(cls, url):
return False if ZDFIE.suitable(url) else super(ZDFChannelIE, cls).suitable(url) return False if ZDFIE.suitable(url) else super(ZDFChannelIE, cls).suitable(url)
def _og_search_title(self, webpage, fatal=False):
title = super(ZDFChannelIE, self)._og_search_title(webpage, fatal=fatal)
return re.split(r'\s+[-|]\s+ZDF(?:mediathek)?$', title or '')[0] or None
def _real_extract(self, url): def _real_extract(self, url):
channel_id = self._match_id(url) channel_id = self._match_id(url)
webpage = self._download_webpage(url, channel_id) webpage = self._download_webpage(url, channel_id)
entries = [ matches = re.finditer(
self.url_result(item_url, ie=ZDFIE.ie_key()) r'''<div\b[^>]*?\sdata-plusbar-id\s*=\s*(["'])(?P<p_id>[\w-]+)\1[^>]*?\sdata-plusbar-url=\1(?P<url>%s)\1''' % ZDFIE._VALID_URL,
for item_url in orderedSet(re.findall( webpage)
r'data-plusbar-url=["\'](http.+?\.html)', webpage))]
return self.playlist_result( if self._downloader.params.get('noplaylist', False):
entries, channel_id, self._og_search_title(webpage, fatal=False)) entry = next(
(self.url_result(m.group('url'), ie=ZDFIE.ie_key()) for m in matches),
None)
self.to_screen('Downloading just the main video because of --no-playlist')
if entry:
return entry
else:
self.to_screen('Downloading playlist %s - add --no-playlist to download just the main video' % (channel_id, ))
r""" def check_video(m):
player = self._extract_player(webpage, channel_id) v_ref = self._search_regex(
r'''(<a\b[^>]*?\shref\s*=[^>]+?\sdata-target-id\s*=\s*(["'])%s\2[^>]*>)''' % (m.group('p_id'), ),
webpage, 'check id', default='')
v_ref = extract_attributes(v_ref)
return v_ref.get('data-target-video-type') != 'novideo'
channel_id = self._search_regex( return self.playlist_from_matches(
r'docId\s*:\s*(["\'])(?P<id>(?!\1).+?)\1', webpage, (m.group('url') for m in matches if check_video(m)),
'channel id', group='id') channel_id, self._og_search_title(webpage, fatal=False))
channel = self._call_api(
'https://api.zdf.de/content/documents/%s.json' % channel_id,
player, url, channel_id)
items = []
for module in channel['module']:
for teaser in try_get(module, lambda x: x['teaser'], list) or []:
t = try_get(
teaser, lambda x: x['http://zdf.de/rels/target'], dict)
if not t:
continue
items.extend(try_get(
t,
lambda x: x['resultsWithVideo']['http://zdf.de/rels/search/results'],
list) or [])
items.extend(try_get(
module,
lambda x: x['filterRef']['resultsWithVideo']['http://zdf.de/rels/search/results'],
list) or [])
entries = []
entry_urls = set()
for item in items:
t = try_get(item, lambda x: x['http://zdf.de/rels/target'], dict)
if not t:
continue
sharing_url = t.get('http://zdf.de/rels/sharing-url')
if not sharing_url or not isinstance(sharing_url, compat_str):
continue
if sharing_url in entry_urls:
continue
entry_urls.add(sharing_url)
entries.append(self.url_result(
sharing_url, ie=ZDFIE.ie_key(), video_id=t.get('id')))
return self.playlist_result(entries, channel_id, channel.get('title'))
"""