|
|
- # coding: utf-8
- from __future__ import unicode_literals
-
- import re
-
- from .common import InfoExtractor
- from ..utils import (
- determine_ext,
- int_or_none,
- js_to_json,
- )
-
-
- class SportBoxEmbedIE(InfoExtractor):
- _VALID_URL = r'https?://news\.sportbox\.ru/vdl/player(?:/[^/]+/|\?.*?\bn?id=)(?P<id>\d+)'
- _TESTS = [{
- 'url': 'http://news.sportbox.ru/vdl/player/ci/211355',
- 'info_dict': {
- 'id': '211355',
- 'ext': 'mp4',
- 'title': '211355',
- 'thumbnail': r're:^https?://.*\.jpg$',
- 'duration': 292,
- 'view_count': int,
- },
- 'params': {
- # m3u8 download
- 'skip_download': True,
- },
- }, {
- 'url': 'http://news.sportbox.ru/vdl/player?nid=370908&only_player=1&autostart=false&playeri=2&height=340&width=580',
- 'only_matching': True,
- }, {
- 'url': 'https://news.sportbox.ru/vdl/player/media/193095',
- 'only_matching': True,
- }]
-
- @staticmethod
- def _extract_urls(webpage):
- return re.findall(
- r'<iframe[^>]+src="(https?://news\.sportbox\.ru/vdl/player[^"]+)"',
- webpage)
-
- def _real_extract(self, url):
- video_id = self._match_id(url)
-
- webpage = self._download_webpage(url, video_id)
-
- wjplayer_data = self._parse_json(
- self._search_regex(
- r'(?s)wjplayer\(({.+?})\);', webpage, 'wjplayer settings'),
- video_id, transform_source=js_to_json)
-
- formats = []
- for source in wjplayer_data['sources']:
- src = source.get('src')
- if not src:
- continue
- if determine_ext(src) == 'm3u8':
- formats.extend(self._extract_m3u8_formats(
- src, video_id, 'mp4', entry_protocol='m3u8_native',
- m3u8_id='hls', fatal=False))
- else:
- formats.append({
- 'url': src,
- })
- self._sort_formats(formats)
-
- view_count = int_or_none(self._search_regex(
- r'Просмотров\s*:\s*(\d+)', webpage, 'view count', default=None))
-
- return {
- 'id': video_id,
- 'title': video_id,
- 'thumbnail': wjplayer_data.get('poster'),
- 'duration': int_or_none(wjplayer_data.get('duration')),
- 'view_count': view_count,
- 'formats': formats,
- }
|