mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 02:35:12 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
45 lines
1.5 KiB
Python
45 lines
1.5 KiB
Python
from .common import InfoExtractor
|
|
from ..utils import parse_duration
|
|
|
|
|
|
class HistoricFilmsIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?historicfilms\.com/(?:tapes/|play)(?P<id>\d+)'
|
|
_TEST = {
|
|
'url': 'http://www.historicfilms.com/tapes/4728',
|
|
'md5': 'd4a437aec45d8d796a38a215db064e9a',
|
|
'info_dict': {
|
|
'id': '4728',
|
|
'ext': 'mov',
|
|
'title': 'Historic Films: GP-7',
|
|
'description': 'md5:1a86a0f3ac54024e419aba97210d959a',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'duration': 2096,
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
tape_id = self._search_regex(
|
|
[r'class="tapeId"[^>]*>([^<]+)<', r'tapeId\s*:\s*"([^"]+)"'],
|
|
webpage, 'tape id')
|
|
|
|
title = self._og_search_title(webpage)
|
|
description = self._og_search_description(webpage)
|
|
thumbnail = self._html_search_meta(
|
|
'thumbnailUrl', webpage, 'thumbnails') or self._og_search_thumbnail(webpage)
|
|
duration = parse_duration(self._html_search_meta(
|
|
'duration', webpage, 'duration'))
|
|
|
|
video_url = f'http://www.historicfilms.com/video/{tape_id}_{video_id}_web.mov'
|
|
|
|
return {
|
|
'id': video_id,
|
|
'url': video_url,
|
|
'title': title,
|
|
'description': description,
|
|
'thumbnail': thumbnail,
|
|
'duration': duration,
|
|
}
|