mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-20 18:27:46 +01:00
91 lines
3.8 KiB
Python
91 lines
3.8 KiB
Python
import json
|
||
import re
|
||
|
||
from .common import InfoExtractor
|
||
from ..utils import (
|
||
int_or_none,
|
||
traverse_obj,
|
||
try_call,
|
||
unescapeHTML,
|
||
unified_timestamp,
|
||
)
|
||
|
||
|
||
class HRFernsehenIE(InfoExtractor):
|
||
IE_NAME = 'hrfernsehen'
|
||
_VALID_URL = r'^https?://www\.(?:hr-fernsehen|hessenschau)\.de/.*,video-(?P<id>[0-9]{6})\.html'
|
||
_TESTS = [{
|
||
'url': 'https://www.hessenschau.de/tv-sendung/hessenschau-vom-26082020,video-130546.html',
|
||
'md5': '5c4e0ba94677c516a2f65a84110fc536',
|
||
'info_dict': {
|
||
'id': '130546',
|
||
'ext': 'mp4',
|
||
'description': 'Sturmtief Kirsten fegt über Hessen / Die Corona-Pandemie – eine Chronologie / '
|
||
'Sterbehilfe: Die Lage in Hessen / Miss Hessen leitet zwei eigene Unternehmen / '
|
||
'Pop-Up Museum zeigt Schwarze Unterhaltung und Black Music',
|
||
'subtitles': {'de': [{
|
||
'url': 'https://hr-a.akamaihd.net/video/as/hessenschau/2020_08/hrLogo_200826200407_L385592_512x288-25p-500kbit.vtt'
|
||
}]},
|
||
'timestamp': 1598400000,
|
||
'upload_date': '20200826',
|
||
'thumbnail': 'https://www.hessenschau.de/tv-sendung/hs_ganz-1554~_t-1598465545029_v-16to9.jpg',
|
||
'title': 'hessenschau vom 26.08.2020',
|
||
'duration': 1654
|
||
}
|
||
}, {
|
||
'url': 'https://www.hr-fernsehen.de/sendungen-a-z/mex/sendungen/fair-und-gut---was-hinter-aldis-eigenem-guetesiegel-steckt,video-130544.html',
|
||
'only_matching': True
|
||
}]
|
||
|
||
_GEO_COUNTRIES = ['DE']
|
||
|
||
def extract_formats(self, loader_data):
|
||
stream_formats = []
|
||
data = loader_data['mediaCollection']['streams'][0]['media']
|
||
for inner in data[1:]:
|
||
stream_format = {
|
||
'format_id': try_call(lambda: f'{inner["maxHResolutionPx"]}p'),
|
||
'height': inner.get('maxHResolutionPx'),
|
||
'url': inner['url'],
|
||
}
|
||
|
||
quality_information = re.search(r'([0-9]{3,4})x([0-9]{3,4})-([0-9]{2})p-([0-9]{3,4})kbit',
|
||
inner['url'])
|
||
if quality_information:
|
||
stream_format['width'] = int_or_none(quality_information.group(1))
|
||
stream_format['height'] = int_or_none(quality_information.group(2))
|
||
stream_format['fps'] = int_or_none(quality_information.group(3))
|
||
stream_format['tbr'] = int_or_none(quality_information.group(4))
|
||
|
||
stream_formats.append(stream_format)
|
||
return stream_formats
|
||
|
||
def _real_extract(self, url):
|
||
video_id = self._match_id(url)
|
||
webpage = self._download_webpage(url, video_id)
|
||
|
||
title = self._html_search_meta(
|
||
['og:title', 'twitter:title', 'name'], webpage)
|
||
description = self._html_search_meta(
|
||
['description'], webpage)
|
||
|
||
loader_str = unescapeHTML(self._search_regex(r"data-(?:new-)?hr-mediaplayer-loader='([^']*)'", webpage, 'ardloader'))
|
||
loader_data = json.loads(loader_str)
|
||
|
||
subtitle = traverse_obj(loader_data, ('mediaCollection', 'subTitles', 0, 'sources', 0, 'url'))
|
||
|
||
info = {
|
||
'id': video_id,
|
||
'title': title,
|
||
'description': description,
|
||
'formats': self.extract_formats(loader_data),
|
||
'subtitles': {'de': [{'url': subtitle}]},
|
||
'timestamp': unified_timestamp(self._search_regex(
|
||
r'<time\sdatetime="(\d{4}\W\d{1,2}\W\d{1,2})', webpage, 'datetime', fatal=False)),
|
||
'duration': int_or_none(traverse_obj(
|
||
loader_data, ('playerConfig', 'pluginData', 'trackingAti@all', 'richMedia', 'duration'))),
|
||
'thumbnail': self._search_regex(r'thumbnailUrl\W*([^"]+)', webpage, 'thumbnail', default=None),
|
||
}
|
||
|
||
return info
|