mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-25 00:00:43 +01:00
[tagesschau] Separate player extractor
This commit is contained in:
parent
9cf79e8f4b
commit
6a0f9a24d0
@ -724,7 +724,10 @@ from .svt import (
|
|||||||
from .swrmediathek import SWRMediathekIE
|
from .swrmediathek import SWRMediathekIE
|
||||||
from .syfy import SyfyIE
|
from .syfy import SyfyIE
|
||||||
from .sztvhu import SztvHuIE
|
from .sztvhu import SztvHuIE
|
||||||
from .tagesschau import TagesschauIE
|
from .tagesschau import (
|
||||||
|
TagesschauPlayerIE,
|
||||||
|
TagesschauIE,
|
||||||
|
)
|
||||||
from .tapely import TapelyIE
|
from .tapely import TapelyIE
|
||||||
from .tass import TassIE
|
from .tass import TassIE
|
||||||
from .tdslifeway import TDSLifewayIE
|
from .tdslifeway import TDSLifewayIE
|
||||||
|
@ -6,10 +6,124 @@ import re
|
|||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
determine_ext,
|
determine_ext,
|
||||||
|
js_to_json,
|
||||||
|
parse_iso8601,
|
||||||
parse_filesize,
|
parse_filesize,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
|
class TagesschauPlayerIE(InfoExtractor):
|
||||||
|
IE_NAME = 'tagesschau:player'
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?tagesschau\.de/multimedia/(?P<kind>audio|video)/(?P=kind)-(?P<id>\d+)~player(?:_[^/?#&]+)?\.html'
|
||||||
|
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'http://www.tagesschau.de/multimedia/video/video-179517~player.html',
|
||||||
|
'md5': '8d09548d5c15debad38bee3a4d15ca21',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '179517',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'Marie Kristin Boese, ARD Berlin, über den zukünftigen Kurs der AfD',
|
||||||
|
'thumbnail': 're:^https?:.*\.jpg$',
|
||||||
|
'formats': 'mincount:6',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
'url': 'https://www.tagesschau.de/multimedia/audio/audio-29417~player.html',
|
||||||
|
'md5': '76e6eec6ebd40740671cf0a2c88617e5',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '29417',
|
||||||
|
'ext': 'mp3',
|
||||||
|
'title': 'Trabi - Bye, bye Rennpappe',
|
||||||
|
'thumbnail': 're:^https?:.*\.jpg$',
|
||||||
|
'formats': 'mincount:2',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
'url': 'http://www.tagesschau.de/multimedia/audio/audio-29417~player_autoplay-true.html',
|
||||||
|
'only_matching': True,
|
||||||
|
}]
|
||||||
|
|
||||||
|
_FORMATS = {
|
||||||
|
'xs': {'quality': 0},
|
||||||
|
's': {'width': 320, 'height': 180, 'quality': 1},
|
||||||
|
'm': {'width': 512, 'height': 288, 'quality': 2},
|
||||||
|
'l': {'width': 960, 'height': 540, 'quality': 3},
|
||||||
|
'xl': {'width': 1280, 'height': 720, 'quality': 4},
|
||||||
|
'xxl': {'quality': 5},
|
||||||
|
}
|
||||||
|
|
||||||
|
def _extract_via_api(self, kind, video_id):
|
||||||
|
info = self._download_json(
|
||||||
|
'https://www.tagesschau.de/api/multimedia/{0}/{0}-{1}.json'.format(kind, video_id),
|
||||||
|
video_id)
|
||||||
|
title = info['headline']
|
||||||
|
formats = []
|
||||||
|
for media in info['mediadata']:
|
||||||
|
for format_id, format_url in media.items():
|
||||||
|
if determine_ext(format_url) == 'm3u8':
|
||||||
|
formats.extend(self._extract_m3u8_formats(
|
||||||
|
format_url, video_id, 'mp4',
|
||||||
|
entry_protocol='m3u8_native', m3u8_id='hls'))
|
||||||
|
else:
|
||||||
|
formats.append({
|
||||||
|
'url': format_url,
|
||||||
|
'format_id': format_id,
|
||||||
|
'vcodec': 'none' if kind == 'audio' else None,
|
||||||
|
})
|
||||||
|
self._sort_formats(formats)
|
||||||
|
timestamp = parse_iso8601(info.get('date'))
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': title,
|
||||||
|
'timestamp': timestamp,
|
||||||
|
'formats': formats,
|
||||||
|
}
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
mobj = re.match(self._VALID_URL, url)
|
||||||
|
video_id = mobj.group('id')
|
||||||
|
|
||||||
|
# kind = mobj.group('kind').lower()
|
||||||
|
# if kind == 'video':
|
||||||
|
# return self._extract_via_api(kind, video_id)
|
||||||
|
|
||||||
|
# JSON api does not provide some audio formats (e.g. ogg) thus
|
||||||
|
# extractiong audio via webpage
|
||||||
|
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
|
||||||
|
title = self._og_search_title(webpage).strip()
|
||||||
|
formats = []
|
||||||
|
|
||||||
|
for media_json in re.findall(r'({src\s*:\s*["\']http[^}]+type\s*:[^}]+})', webpage):
|
||||||
|
media = self._parse_json(js_to_json(media_json), video_id, fatal=False)
|
||||||
|
if not media:
|
||||||
|
continue
|
||||||
|
src = media.get('src')
|
||||||
|
if not src:
|
||||||
|
return
|
||||||
|
quality = media.get('quality')
|
||||||
|
kind = media.get('type', '').split('/')[0]
|
||||||
|
ext = determine_ext(src)
|
||||||
|
f = {
|
||||||
|
'url': src,
|
||||||
|
'format_id': '%s_%s' % (quality, ext) if quality else ext,
|
||||||
|
'ext': ext,
|
||||||
|
'vcodec': 'none' if kind == 'audio' else None,
|
||||||
|
}
|
||||||
|
f.update(self._FORMATS.get(quality, {}))
|
||||||
|
formats.append(f)
|
||||||
|
|
||||||
|
self._sort_formats(formats)
|
||||||
|
|
||||||
|
thumbnail = self._og_search_thumbnail(webpage)
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': title,
|
||||||
|
'thumbnail': thumbnail,
|
||||||
|
'formats': formats,
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
class TagesschauIE(InfoExtractor):
|
class TagesschauIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://(?:www\.)?tagesschau\.de/[^/]+/(?:[^/]+/)*?[^/#?]+?(?P<id>-?[0-9]+)(?:~_?[^/#?]+?)?\.html'
|
_VALID_URL = r'https?://(?:www\.)?tagesschau\.de/[^/]+/(?:[^/]+/)*?[^/#?]+?(?P<id>-?[0-9]+)(?:~_?[^/#?]+?)?\.html'
|
||||||
|
|
||||||
@ -20,7 +134,7 @@ class TagesschauIE(InfoExtractor):
|
|||||||
'id': '102143',
|
'id': '102143',
|
||||||
'ext': 'mp4',
|
'ext': 'mp4',
|
||||||
'title': 'Regierungsumbildung in Athen: Neue Minister in Griechenland vereidigt',
|
'title': 'Regierungsumbildung in Athen: Neue Minister in Griechenland vereidigt',
|
||||||
'description': 'md5:171feccd9d9b3dd54d05d501568f6359',
|
'description': '18.07.2015 20:10 Uhr',
|
||||||
'thumbnail': 're:^https?:.*\.jpg$',
|
'thumbnail': 're:^https?:.*\.jpg$',
|
||||||
},
|
},
|
||||||
}, {
|
}, {
|
||||||
@ -29,18 +143,30 @@ class TagesschauIE(InfoExtractor):
|
|||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '5727',
|
'id': '5727',
|
||||||
'ext': 'mp4',
|
'ext': 'mp4',
|
||||||
'description': 'md5:695c01bfd98b7e313c501386327aea59',
|
|
||||||
'title': 'Sendung: tagesschau \t04.12.2014 20:00 Uhr',
|
'title': 'Sendung: tagesschau \t04.12.2014 20:00 Uhr',
|
||||||
|
'description': 'md5:695c01bfd98b7e313c501386327aea59',
|
||||||
'thumbnail': 're:^https?:.*\.jpg$',
|
'thumbnail': 're:^https?:.*\.jpg$',
|
||||||
},
|
},
|
||||||
}, {
|
}, {
|
||||||
'url': 'http://www.tagesschau.de/multimedia/politikimradio/audio-18407.html',
|
# exclusive audio
|
||||||
'md5': 'aef45de271c4bf0a5db834aa40bf774c',
|
'url': 'http://www.tagesschau.de/multimedia/audio/audio-29417.html',
|
||||||
|
'md5': '76e6eec6ebd40740671cf0a2c88617e5',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '18407',
|
'id': '29417',
|
||||||
'ext': 'mp3',
|
'ext': 'mp3',
|
||||||
'title': 'Flüchtlingsdebatte: Hitzig, aber wenig hilfreich',
|
'title': 'Trabi - Bye, bye Rennpappe',
|
||||||
'description': 'Flüchtlingsdebatte: Hitzig, aber wenig hilfreich',
|
'description': 'md5:8687dda862cbbe2cfb2df09b56341317',
|
||||||
|
'thumbnail': 're:^https?:.*\.jpg$',
|
||||||
|
},
|
||||||
|
}, {
|
||||||
|
# audio in article
|
||||||
|
'url': 'http://www.tagesschau.de/inland/bnd-303.html',
|
||||||
|
'md5': 'e0916c623e85fc1d2b26b78f299d3958',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '303',
|
||||||
|
'ext': 'mp3',
|
||||||
|
'title': 'Viele Baustellen für neuen BND-Chef',
|
||||||
|
'description': 'md5:1e69a54be3e1255b2b07cdbce5bcd8b4',
|
||||||
'thumbnail': 're:^https?:.*\.jpg$',
|
'thumbnail': 're:^https?:.*\.jpg$',
|
||||||
},
|
},
|
||||||
}, {
|
}, {
|
||||||
@ -71,19 +197,11 @@ class TagesschauIE(InfoExtractor):
|
|||||||
}, {
|
}, {
|
||||||
'url': 'http://www.tagesschau.de/multimedia/video/video-102303~_bab-sendung-211.html',
|
'url': 'http://www.tagesschau.de/multimedia/video/video-102303~_bab-sendung-211.html',
|
||||||
'only_matching': True,
|
'only_matching': True,
|
||||||
}, {
|
|
||||||
'url': 'http://www.tagesschau.de/multimedia/video/video-179517~player.html',
|
|
||||||
'only_matching': True,
|
|
||||||
}]
|
}]
|
||||||
|
|
||||||
_FORMATS = {
|
@classmethod
|
||||||
'xs': {'quality': 0},
|
def suitable(cls, url):
|
||||||
's': {'width': 320, 'height': 180, 'quality': 1},
|
return False if TagesschauPlayerIE.suitable(url) else super(TagesschauIE, cls).suitable(url)
|
||||||
'm': {'width': 512, 'height': 288, 'quality': 2},
|
|
||||||
'l': {'width': 960, 'height': 540, 'quality': 3},
|
|
||||||
'xl': {'width': 1280, 'height': 720, 'quality': 4},
|
|
||||||
'xxl': {'quality': 5},
|
|
||||||
}
|
|
||||||
|
|
||||||
def _extract_formats(self, download_text, media_kind):
|
def _extract_formats(self, download_text, media_kind):
|
||||||
links = re.finditer(
|
links = re.finditer(
|
||||||
@ -140,39 +258,12 @@ class TagesschauIE(InfoExtractor):
|
|||||||
def _real_extract(self, url):
|
def _real_extract(self, url):
|
||||||
video_id = self._match_id(url)
|
video_id = self._match_id(url)
|
||||||
display_id = video_id.lstrip('-')
|
display_id = video_id.lstrip('-')
|
||||||
|
|
||||||
webpage = self._download_webpage(url, display_id)
|
webpage = self._download_webpage(url, display_id)
|
||||||
|
|
||||||
player_url = self._html_search_meta(
|
|
||||||
'twitter:player', webpage, 'player URL', default=None)
|
|
||||||
if player_url:
|
|
||||||
playerpage = self._download_webpage(
|
|
||||||
player_url, display_id, 'Downloading player page')
|
|
||||||
|
|
||||||
formats = []
|
|
||||||
for media in re.finditer(
|
|
||||||
r'''(?x)
|
|
||||||
(?P<q_url>["\'])(?P<url>http://media.+?)(?P=q_url)
|
|
||||||
,\s*type:(?P<q_type>["\'])(?P<type>video|audio)/(?P<ext>.+?)(?P=q_type)
|
|
||||||
(?:,\s*quality:(?P<q_quality>["\'])(?P<quality>.+?)(?P=q_quality))?
|
|
||||||
''', playerpage):
|
|
||||||
url = media.group('url')
|
|
||||||
webpage_type = media.group('type')
|
|
||||||
ext = media.group('ext')
|
|
||||||
res = media.group('quality')
|
|
||||||
f = {
|
|
||||||
'format_id': '%s_%s' % (res, ext) if res else ext,
|
|
||||||
'url': url,
|
|
||||||
'ext': ext,
|
|
||||||
'vcodec': 'none' if webpage_type == 'audio' else None,
|
|
||||||
}
|
|
||||||
f.update(self._FORMATS.get(res, {}))
|
|
||||||
formats.append(f)
|
|
||||||
thumbnail = self._og_search_thumbnail(playerpage)
|
|
||||||
title = self._og_search_title(webpage).strip()
|
|
||||||
description = self._og_search_description(webpage).strip()
|
|
||||||
else:
|
|
||||||
title = self._html_search_regex(
|
title = self._html_search_regex(
|
||||||
r'<span class="headline".*?>(.*?)</span>', webpage, 'title')
|
r'<span[^>]*class="headline"[^>]*>(.+?)</span>',
|
||||||
|
webpage, 'title', default=None) or self._og_search_title(webpage)
|
||||||
|
|
||||||
DOWNLOAD_REGEX = r'(?s)<p>Wir bieten dieses (?P<kind>Video|Audio) in folgenden Formaten zum Download an:</p>\s*<div class="controls">(?P<links>.*?)</div>\s*<p>'
|
DOWNLOAD_REGEX = r'(?s)<p>Wir bieten dieses (?P<kind>Video|Audio) in folgenden Formaten zum Download an:</p>\s*<div class="controls">(?P<links>.*?)</div>\s*<p>'
|
||||||
|
|
||||||
@ -187,12 +278,14 @@ class TagesschauIE(InfoExtractor):
|
|||||||
'title': '%s' % entry_title,
|
'title': '%s' % entry_title,
|
||||||
'formats': self._extract_formats(download_text, media_kind),
|
'formats': self._extract_formats(download_text, media_kind),
|
||||||
})
|
})
|
||||||
|
if len(entries) > 1:
|
||||||
return self.playlist_result(entries, display_id, title)
|
return self.playlist_result(entries, display_id, title)
|
||||||
|
formats = entries[0]['formats']
|
||||||
else: # Assume single video
|
else: # Assume single video
|
||||||
download_text = self._search_regex(
|
download_text = self._search_regex(
|
||||||
DOWNLOAD_REGEX, webpage, 'download links', group='links')
|
DOWNLOAD_REGEX, webpage, 'download links', group='links')
|
||||||
media_kind = self._search_regex(
|
media_kind = self._search_regex(
|
||||||
DOWNLOAD_REGEX, webpage, 'media kind', default='Video', group='links')
|
DOWNLOAD_REGEX, webpage, 'media kind', default='Video', group='kind')
|
||||||
formats = self._extract_formats(download_text, media_kind)
|
formats = self._extract_formats(download_text, media_kind)
|
||||||
thumbnail = self._og_search_thumbnail(webpage)
|
thumbnail = self._og_search_thumbnail(webpage)
|
||||||
description = self._html_search_regex(
|
description = self._html_search_regex(
|
||||||
|
Loading…
Reference in New Issue
Block a user