1
0
mirror of https://github.com/l1ving/youtube-dl synced 2024-11-30 11:52:52 +08:00
youtube-dl/youtube_dl/extractor/vrt.py

178 lines
7.1 KiB
Python
Raw Normal View History

2014-10-26 17:48:11 +08:00
# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
determine_ext,
float_or_none,
)
2014-10-26 17:48:11 +08:00
class VRTIE(InfoExtractor):
2016-01-05 03:58:25 +08:00
_VALID_URL = r'https?://(?:deredactie|sporza|cobra(?:\.canvas)?)\.be/cm/(?:[^/]+/)+(?P<id>[^/]+)/*'
2014-10-26 17:48:11 +08:00
_TESTS = [
# deredactie.be
{
'url': 'http://deredactie.be/cm/vrtnieuws/videozone/programmas/journaal/EP_141025_JOL',
'md5': '4cebde1eb60a53782d4f3992cbd46ec8',
'info_dict': {
'id': '2129880',
'ext': 'flv',
'title': 'Het journaal L - 25/10/14',
'description': None,
'timestamp': 1414271750.949,
'upload_date': '20141025',
'duration': 929,
2016-06-28 21:02:03 +08:00
},
'skip': 'HTTP Error 404: Not Found',
2014-10-26 17:48:11 +08:00
},
# sporza.be
{
'url': 'http://sporza.be/cm/sporza/videozone/programmas/extratime/EP_141020_Extra_time',
'md5': '11f53088da9bf8e7cfc42456697953ff',
'info_dict': {
'id': '2124639',
'ext': 'flv',
'title': 'Bekijk Extra Time van 20 oktober',
'description': 'md5:83ac5415a4f1816c6a93f8138aef2426',
'timestamp': 1413835980.560,
'upload_date': '20141020',
'duration': 3238,
2016-06-28 21:02:03 +08:00
},
'skip': 'HTTP Error 404: Not Found',
2014-10-26 17:48:11 +08:00
},
# cobra.be
{
'url': 'http://cobra.be/cm/cobra/videozone/rubriek/film-videozone/141022-mv-ellis-cafecorsari',
'md5': '78a2b060a5083c4f055449a72477409d',
'info_dict': {
'id': '2126050',
'ext': 'flv',
'title': 'Bret Easton Ellis in Café Corsari',
'description': 'md5:f699986e823f32fd6036c1855a724ee9',
'timestamp': 1413967500.494,
'upload_date': '20141022',
'duration': 661,
2016-06-28 21:02:03 +08:00
},
'skip': 'HTTP Error 404: Not Found',
2014-10-26 17:48:11 +08:00
},
{
# YouTube video
'url': 'http://deredactie.be/cm/vrtnieuws/videozone/nieuws/cultuurenmedia/1.2622957',
2016-06-28 21:02:03 +08:00
'md5': 'b8b93da1df1cea6c8556255a796b7d61',
'info_dict': {
'id': 'Wji-BZ0oCwg',
'ext': 'mp4',
'title': 'ROGUE ONE: A STAR WARS STORY Official Teaser Trailer',
'description': 'md5:8e468944dce15567a786a67f74262583',
'uploader': 'Star Wars',
'uploader_id': 'starwars',
'upload_date': '20160407',
},
'add_ie': ['Youtube'],
},
2016-01-05 03:58:25 +08:00
{
'url': 'http://cobra.canvas.be/cm/cobra/videozone/rubriek/film-videozone/1.2377055',
2016-06-28 21:02:03 +08:00
'md5': '',
'info_dict': {
'id': '2377055',
'ext': 'mp4',
'title': 'Cafe Derby',
'description': 'Lenny Van Wesemael debuteert met de langspeelfilm Café Derby. Een waar gebeurd maar ook verzonnen verhaal.',
'upload_date': '20150626',
'timestamp': 1435305240.769,
},
'params': {
# m3u8 download
'skip_download': True,
}
2016-01-05 03:58:25 +08:00
}
2014-10-26 17:48:11 +08:00
]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
video_id = self._search_regex(
r'data-video-id="([^"]+)_[^"]+"', webpage, 'video id', fatal=False)
src = self._search_regex(
r'data-video-src="([^"]+)"', webpage, 'video src', default=None)
video_type = self._search_regex(
r'data-video-type="([^"]+)"', webpage, 'video type', default=None)
if video_type == 'YouTubeVideo':
return self.url_result(src, 'Youtube')
2014-10-26 17:48:11 +08:00
formats = []
2014-10-26 17:48:11 +08:00
mobj = re.search(
r'data-video-iphone-server="(?P<server>[^"]+)"\s+data-video-iphone-path="(?P<path>[^"]+)"',
webpage)
if mobj:
formats.extend(self._extract_m3u8_formats(
'%s/%s' % (mobj.group('server'), mobj.group('path')),
video_id, 'mp4', m3u8_id='hls', fatal=False))
if src:
if determine_ext(src) == 'm3u8':
formats.extend(self._extract_m3u8_formats(
src, video_id, 'mp4', entry_protocol='m3u8_native',
m3u8_id='hls', fatal=False))
2016-06-28 21:02:03 +08:00
formats.extend(self._extract_f4m_formats(
src.replace('playlist.m3u8', 'manifest.f4m'),
video_id, f4m_id='hds', fatal=False))
if 'data-video-geoblocking="true"' not in webpage:
rtmp_formats = self._extract_smil_formats(
src.replace('playlist.m3u8', 'jwplayer.smil'),
video_id, fatal=False)
formats.extend(rtmp_formats)
for rtmp_format in rtmp_formats:
rtmp_format_c = rtmp_format.copy()
rtmp_format_c['url'] = '%s/%s' % (rtmp_format['url'], rtmp_format['play_path'])
del rtmp_format_c['play_path']
del rtmp_format_c['ext']
http_format = rtmp_format_c.copy()
http_format.update({
'url': rtmp_format_c['url'].replace('rtmp://', 'http://').replace('vod.', 'download.').replace('/_definst_/', '/').replace('mp4:', ''),
'format_id': rtmp_format['format_id'].replace('rtmp', 'http'),
'protocol': 'http',
})
rtsp_format = rtmp_format_c.copy()
rtsp_format.update({
'url': rtsp_format['url'].replace('rtmp://', 'rtsp://'),
'format_id': rtmp_format['format_id'].replace('rtmp', 'rtsp'),
'protocol': 'rtsp',
})
formats.extend([http_format, rtsp_format])
else:
formats.extend(self._extract_f4m_formats(
'%s/manifest.f4m' % src, video_id, f4m_id='hds', fatal=False))
2016-02-18 01:28:41 +08:00
if not formats and 'data-video-geoblocking="true"' in webpage:
self.raise_geo_restricted('This video is only available in Belgium')
2014-10-26 17:48:11 +08:00
self._sort_formats(formats)
title = self._og_search_title(webpage)
description = self._og_search_description(webpage, default=None)
thumbnail = self._og_search_thumbnail(webpage)
timestamp = float_or_none(self._search_regex(
r'data-video-sitestat-pubdate="(\d+)"', webpage, 'timestamp', fatal=False), 1000)
duration = float_or_none(self._search_regex(
r'data-video-duration="(\d+)"', webpage, 'duration', fatal=False), 1000)
return {
'id': video_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'timestamp': timestamp,
'duration': duration,
'formats': formats,
2014-11-24 03:41:03 +08:00
}