0
0
Fork 0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-10-31 23:02:40 +00:00
yt-dlp/yt_dlp/extractor/heise.py
pukkandan bfd973ece3 [extractors] Use new framework for existing embeds (#4307)
`Brightcove` is difficult to migrate because it's subclasses may depend
on the signature of the current functions. So it is left as-is for now

Note: Tests have not been migrated
2022-08-02 01:08:16 +05:30

169 lines
6.3 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

from .common import InfoExtractor
from .kaltura import KalturaIE
from .youtube import YoutubeIE
from ..utils import (
determine_ext,
int_or_none,
NO_DEFAULT,
parse_iso8601,
smuggle_url,
xpath_text,
)
class HeiseIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?heise\.de/(?:[^/]+/)+[^/]+-(?P<id>[0-9]+)\.html'
_TESTS = [{
# kaltura embed
'url': 'http://www.heise.de/video/artikel/Podcast-c-t-uplink-3-3-Owncloud-Tastaturen-Peilsender-Smartphone-2404147.html',
'info_dict': {
'id': '1_kkrq94sm',
'ext': 'mp4',
'title': "Podcast: c't uplink 3.3 Owncloud / Tastaturen / Peilsender Smartphone",
'timestamp': 1512734959,
'upload_date': '20171208',
'description': 'md5:c934cbfb326c669c2bcabcbe3d3fcd20',
},
'params': {
'skip_download': True,
},
}, {
# YouTube embed
'url': 'http://www.heise.de/newsticker/meldung/Netflix-In-20-Jahren-vom-Videoverleih-zum-TV-Revolutionaer-3814130.html',
'md5': 'e403d2b43fea8e405e88e3f8623909f1',
'info_dict': {
'id': '6kmWbXleKW4',
'ext': 'mp4',
'title': 'NEU IM SEPTEMBER | Netflix',
'description': 'md5:2131f3c7525e540d5fd841de938bd452',
'upload_date': '20170830',
'uploader': 'Netflix Deutschland, Österreich und Schweiz',
'uploader_id': 'netflixdach',
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.heise.de/video/artikel/nachgehakt-Wie-sichert-das-c-t-Tool-Restric-tor-Windows-10-ab-3700244.html',
'info_dict': {
'id': '1_ntrmio2s',
'ext': 'mp4',
'title': "nachgehakt: Wie sichert das c't-Tool Restric'tor Windows 10 ab?",
'description': 'md5:47e8ffb6c46d85c92c310a512d6db271',
'timestamp': 1512470717,
'upload_date': '20171205',
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.heise.de/ct/artikel/c-t-uplink-20-8-Staubsaugerroboter-Xiaomi-Vacuum-2-AR-Brille-Meta-2-und-Android-rooten-3959893.html',
'info_dict': {
'id': '1_59mk80sf',
'ext': 'mp4',
'title': "c't uplink 20.8: Staubsaugerroboter Xiaomi Vacuum 2, AR-Brille Meta 2 und Android rooten",
'description': 'md5:f50fe044d3371ec73a8f79fcebd74afc',
'timestamp': 1517567237,
'upload_date': '20180202',
},
'params': {
'skip_download': True,
},
}, {
'url': 'http://www.heise.de/ct/artikel/c-t-uplink-3-3-Owncloud-Tastaturen-Peilsender-Smartphone-2403911.html',
'only_matching': True,
}, {
'url': 'http://www.heise.de/newsticker/meldung/c-t-uplink-Owncloud-Tastaturen-Peilsender-Smartphone-2404251.html?wt_mc=rss.ho.beitrag.atom',
'only_matching': True,
}, {
'url': 'http://www.heise.de/ct/ausgabe/2016-12-Spiele-3214137.html',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
def extract_title(default=NO_DEFAULT):
title = self._html_search_meta(
('fulltitle', 'title'), webpage, default=None)
if not title or title == "c't":
title = self._search_regex(
r'<div[^>]+class="videoplayerjw"[^>]+data-title="([^"]+)"',
webpage, 'title', default=None)
if not title:
title = self._html_search_regex(
r'<h1[^>]+\bclass=["\']article_page_title[^>]+>(.+?)<',
webpage, 'title', default=default)
return title
title = extract_title(default=None)
description = self._og_search_description(
webpage, default=None) or self._html_search_meta(
'description', webpage)
def _make_kaltura_result(kaltura_url):
return {
'_type': 'url_transparent',
'url': smuggle_url(kaltura_url, {'source_url': url}),
'ie_key': KalturaIE.ie_key(),
'title': title,
'description': description,
}
kaltura_url = KalturaIE._extract_url(webpage)
if kaltura_url:
return _make_kaltura_result(kaltura_url)
kaltura_id = self._search_regex(
r'entry-id=(["\'])(?P<id>(?:(?!\1).)+)\1', webpage, 'kaltura id',
default=None, group='id')
if kaltura_id:
return _make_kaltura_result('kaltura:2238431:%s' % kaltura_id)
yt_urls = YoutubeIE._extract_embed_urls(url, webpage)
if yt_urls:
return self.playlist_from_matches(
yt_urls, video_id, title, ie=YoutubeIE.ie_key())
title = extract_title()
container_id = self._search_regex(
r'<div class="videoplayerjw"[^>]+data-container="([0-9]+)"',
webpage, 'container ID')
sequenz_id = self._search_regex(
r'<div class="videoplayerjw"[^>]+data-sequenz="([0-9]+)"',
webpage, 'sequenz ID')
doc = self._download_xml(
'http://www.heise.de/videout/feed', video_id, query={
'container': container_id,
'sequenz': sequenz_id,
})
formats = []
for source_node in doc.findall('.//{http://rss.jwpcdn.com/}source'):
label = source_node.attrib['label']
height = int_or_none(self._search_regex(
r'^(.*?_)?([0-9]+)p$', label, 'height', default=None))
video_url = source_node.attrib['file']
ext = determine_ext(video_url, '')
formats.append({
'url': video_url,
'format_note': label,
'format_id': '%s_%s' % (ext, label),
'height': height,
})
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'description': description,
'thumbnail': (xpath_text(doc, './/{http://rss.jwpcdn.com/}image')
or self._og_search_thumbnail(webpage)),
'timestamp': parse_iso8601(
self._html_search_meta('date', webpage)),
'formats': formats,
}