mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 15:16:52 +01:00
[hse] Add extractors (#1906)
Original PR: https://github.com/ytdl-org/youtube-dl/pull/27060 Authored by: cypheron, pukkandan
This commit is contained in:
parent
ca5db158ae
commit
ae43a4b986
@ -565,6 +565,10 @@
|
||||
HRTiIE,
|
||||
HRTiPlaylistIE,
|
||||
)
|
||||
from .hse import (
|
||||
HSEShowIE,
|
||||
HSEProductIE,
|
||||
)
|
||||
from .huajiao import HuajiaoIE
|
||||
from .huffpost import HuffPostIE
|
||||
from .hungama import (
|
||||
|
95
yt_dlp/extractor/hse.py
Normal file
95
yt_dlp/extractor/hse.py
Normal file
@ -0,0 +1,95 @@
|
||||
# coding: utf-8
|
||||
from .common import InfoExtractor
|
||||
from ..utils import (
|
||||
ExtractorError,
|
||||
traverse_obj,
|
||||
unified_timestamp,
|
||||
)
|
||||
|
||||
|
||||
class HSEShowBaseInfoExtractor(InfoExtractor):
|
||||
_GEO_COUNTRIES = ['DE']
|
||||
|
||||
def _extract_redux_data(self, url, video_id):
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
redux = self._html_search_regex(
|
||||
r'window\.__REDUX_DATA__\s*=\s*({.*});?', webpage, 'redux data')
|
||||
return self._parse_json(redux.replace('\n', ''), video_id)
|
||||
|
||||
def _extract_formats_and_subtitles(self, sources, video_id):
|
||||
if not sources:
|
||||
raise ExtractorError('No video found', expected=True, video_id=video_id)
|
||||
formats, subtitles = [], {}
|
||||
for src in sources:
|
||||
if src['mimetype'] != 'application/x-mpegURL':
|
||||
continue
|
||||
fmts, subs = self._extract_m3u8_formats_and_subtitles(src['url'], video_id, ext='mp4')
|
||||
formats.extend(fmts)
|
||||
subtitles = self._merge_subtitles(subtitles, subs)
|
||||
self._sort_formats(formats)
|
||||
return formats, subtitles
|
||||
|
||||
|
||||
class HSEShowIE(HSEShowBaseInfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?hse\.de/dpl/c/tv-shows/(?P<id>[0-9]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://www.hse.de/dpl/c/tv-shows/505350',
|
||||
'info_dict': {
|
||||
'id': '505350',
|
||||
'ext': 'mp4',
|
||||
'title': 'Pfeffinger Mode & Accessoires',
|
||||
'timestamp': 1638810000,
|
||||
'upload_date': '20211206',
|
||||
'channel': 'HSE24',
|
||||
'uploader': 'Arina Pirayesh'
|
||||
},
|
||||
'params': {'skip_download': 'm3u8'},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
json_data = self._extract_redux_data(url, video_id)
|
||||
formats, subtitles = self._extract_formats_and_subtitles(
|
||||
traverse_obj(json_data, ('tvShowPage', 'tvShowVideo', 'sources')), video_id)
|
||||
|
||||
show = traverse_obj(json_data, ('tvShowPage', 'tvShow')) or {}
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': show.get('title') or video_id,
|
||||
'formats': formats,
|
||||
'timestamp': unified_timestamp(f'{show.get("date")} {show.get("hour")}:00'),
|
||||
'thumbnail': traverse_obj(json_data, ('tvShowVideo', 'poster')),
|
||||
'channel': self._search_regex(
|
||||
r'tvShow \| ([A-Z0-9]+)_', show.get('actionFieldText') or '', video_id, fatal=False),
|
||||
'uploader': show.get('presenter'),
|
||||
'subtitles': subtitles,
|
||||
}
|
||||
|
||||
|
||||
class HSEProductIE(HSEShowBaseInfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?hse\.de/dpl/p/product/(?P<id>[0-9]+)'
|
||||
_TESTS = [{
|
||||
'url': 'https://www.hse.de/dpl/p/product/408630',
|
||||
'info_dict': {
|
||||
'id': '408630',
|
||||
'ext': 'mp4',
|
||||
'title': 'Hose im Ponte-Mix',
|
||||
'uploader': 'Judith Williams'
|
||||
},
|
||||
'params': {'skip_download': 'm3u8'},
|
||||
}]
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
json_data = self._extract_redux_data(url, video_id)
|
||||
video = traverse_obj(json_data, ('productContent', 'productContent', 'videos', 0)) or {}
|
||||
formats, subtitles = self._extract_formats_and_subtitles(video.get('sources'), video_id)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': traverse_obj(json_data, ('productDetail', 'product', 'name', 'short')) or video_id,
|
||||
'formats': formats,
|
||||
'subtitles': subtitles,
|
||||
'thumbnail': video.get('poster'),
|
||||
'uploader': traverse_obj(json_data, ('productDetail', 'product', 'brand', 'brandName')),
|
||||
}
|
Loading…
Reference in New Issue
Block a user