mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-05 12:00:21 +00:00
47 lines
1.6 KiB
Python
47 lines
1.6 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
|
class InternazionaleIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?internazionale\.it/video/(?P<id>.*)'
|
|
_TESTS = [{
|
|
'url': 'https://www.internazionale.it/video/2015/02/19/richard-linklater-racconta-una-scena-di-boyhood',
|
|
'md5': '11b54a3d3333e455c00684e50a65c58e',
|
|
'info_dict': {
|
|
'id': '265968',
|
|
'ext': 'mp4',
|
|
'description': 'md5:efb7e5bbfb1a54ae2ed5a4a015f0e665',
|
|
'title': 'Richard Linklater racconta una scena di Boyhood',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
}
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
data_job_id = self._html_search_regex(r'data-job-id="([^"]+)"', webpage, 'data-job-id')
|
|
data_video_path = self._html_search_regex(r'data-video-path="([^"]+)"', webpage, 'data-video-path')
|
|
|
|
formats = []
|
|
|
|
formats.extend(self._extract_m3u8_formats(
|
|
'https://video.internazionale.it/%s/%s.m3u8' % (data_video_path, data_job_id),
|
|
video_id))
|
|
|
|
formats.extend(self._extract_mpd_formats(
|
|
'https://video.internazionale.it/%s/%s.mpd' % (data_video_path, data_job_id),
|
|
video_id))
|
|
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': data_job_id,
|
|
'title': self._og_search_title(webpage),
|
|
'thumbnail': self._og_search_thumbnail(webpage),
|
|
'description': self._og_search_description(webpage),
|
|
'formats': formats,
|
|
}
|