youtube-dl/youtube_dl/extractor/playtvak.py

182 lines
6.6 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..compat import (
compat_urlparse,
compat_urllib_parse,
)
2015-08-15 21:07:05 +00:00
from ..utils import (
ExtractorError,
int_or_none,
parse_iso8601,
qualities,
)
class PlaytvakIE(InfoExtractor):
2015-08-15 21:07:05 +00:00
IE_DESC = 'Playtvak.cz, iDNES.cz and Lidovky.cz'
_VALID_URL = r'https?://(?:.+?\.)?(?:playtvak|idnes|lidovky|metro)\.cz/.*\?(?:c|idvideo)=(?P<id>[^&]+)'
_TESTS = [{
'url': 'http://www.playtvak.cz/vyzente-vosy-a-srsne-ze-zahrady-dn5-/hodinovy-manzel.aspx?c=A150730_150323_hodinovy-manzel_kuko',
'md5': '4525ae312c324b4be2f4603cc78ceb4a',
'info_dict': {
'id': 'A150730_150323_hodinovy-manzel_kuko',
'ext': 'mp4',
'title': 'Vyžeňte vosy a sršně ze zahrady',
2015-08-15 21:07:05 +00:00
'description': 'md5:f93d398691044d303bc4a3de62f3e976',
'thumbnail': 're:(?i)^https?://.*\.(?:jpg|png)$',
'duration': 279,
'timestamp': 1438732860,
'upload_date': '20150805',
'is_live': False,
}
}, { # live video test
'url': 'http://slowtv.playtvak.cz/planespotting-0pr-/planespotting.aspx?c=A150624_164934_planespotting_cat',
'info_dict': {
'id': 'A150624_164934_planespotting_cat',
'ext': 'flv',
'title': 're:^Přímý přenos iDNES.cz [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
'description': 'Sledujte provoz na ranveji Letiště Václava Havla v Praze',
2015-08-15 21:07:05 +00:00
'thumbnail': 're:(?i)^https?://.*\.(?:jpg|png)$',
'is_live': True,
},
'params': {
'skip_download': True, # requires rtmpdump
},
}, { # idnes.cz
'url': 'http://zpravy.idnes.cz/pes-zavreny-v-aute-rozbijeni-okynek-v-aute-fj5-/domaci.aspx?c=A150809_104116_domaci_pku',
'md5': '819832ba33cd7016e58a6658577fe289',
'info_dict': {
'id': 'A150809_104116_domaci_pku',
'ext': 'mp4',
'title': 'Zavřeli jsme mraženou pizzu do auta. Upekla se',
2015-08-15 21:07:05 +00:00
'description': 'md5:01e73f02329e2e5760bd5eed4d42e3c2',
'thumbnail': 're:(?i)^https?://.*\.(?:jpg|png)$',
'duration': 39,
'timestamp': 1438969140,
'upload_date': '20150807',
'is_live': False,
}
}, { # lidovky.cz
'url': 'http://www.lidovky.cz/dalsi-demonstrace-v-praze-o-migraci-duq-/video.aspx?c=A150808_214044_ln-video_ELE',
'md5': 'c7209ac4ba9d234d4ad5bab7485bcee8',
'info_dict': {
'id': 'A150808_214044_ln-video_ELE',
'ext': 'mp4',
'title': 'Táhni! Demonstrace proti imigrantům budila emoce',
2015-08-15 21:07:05 +00:00
'description': 'md5:97c81d589a9491fbfa323c9fa3cca72c',
'thumbnail': 're:(?i)^https?://.*\.(?:jpg|png)$',
'timestamp': 1439052180,
'upload_date': '20150808',
'is_live': False,
}
}, { # metro.cz
'url': 'http://www.metro.cz/video-pod-billboardem-se-na-vltavske-roztocil-kolotoc-deti-vozil-jen-par-hodin-1hx-/metro-extra.aspx?c=A141111_173251_metro-extra_row',
'md5': '84fc1deedcac37b7d4a6ccae7c716668',
'info_dict': {
'id': 'A141111_173251_metro-extra_row',
'ext': 'mp4',
'title': 'Recesisté udělali z billboardu kolotoč',
'description': 'md5:7369926049588c3989a66c9c1a043c4c',
'thumbnail': 're:(?i)^https?://.*\.(?:jpg|png)$',
'timestamp': 1415725500,
'upload_date': '20141111',
'is_live': False,
}
2015-08-15 21:07:05 +00:00
}, {
'url': 'http://www.playtvak.cz/embed.aspx?idvideo=V150729_141549_play-porad_kuko',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
2015-08-15 21:07:05 +00:00
webpage = self._download_webpage(url, video_id)
2015-08-15 21:07:05 +00:00
info_url = self._html_search_regex(
r'Misc\.videoFLV\(\s*{\s*data\s*:\s*"([^"]+)"', webpage, 'info url')
parsed_url = compat_urlparse.urlparse(info_url)
qs = compat_urlparse.parse_qs(parsed_url.query)
qs.update({
'reklama': ['0'],
'type': ['js'],
})
info_url = compat_urlparse.urlunparse(
parsed_url._replace(query = compat_urllib_parse.urlencode(qs, True)))
json_info = self._download_json(
info_url, video_id,
transform_source=lambda s: s[s.index('{'):s.rindex('}') + 1])
item = None
2015-08-15 21:07:05 +00:00
for i in json_info['items']:
if i.get('type') == 'video' or i.get('type') == 'stream':
item = i
break
2015-08-15 21:07:05 +00:00
if not item:
raise ExtractorError('No suitable stream found')
2015-08-15 21:07:05 +00:00
quality = qualities(['low', 'middle', 'high'])
formats = []
for fmt in item['video']:
2015-08-15 21:07:05 +00:00
video_url = fmt.get('file')
if not video_url:
continue
format_ = fmt['format']
format_id = '%s_%s' % (format_, fmt['quality'])
preference = None
if format_ in ['mp4', 'webm']:
ext = format_
elif format_ == 'rtmp':
ext = 'flv'
elif format_ == 'apple':
ext = 'mp4'
# Some streams have mp3 audio which does not play
# well with ffmpeg filter aac_adtstoasc
2015-08-15 21:07:05 +00:00
preference = -1
elif format_ == 'adobe': # f4m manifest fails with 404 in 80% of requests
continue
else: # Other formats not supported yet
continue
2015-08-15 21:07:05 +00:00
formats.append({
'url': video_url,
'ext': ext,
'format_id': format_id,
'quality': quality(fmt.get('quality')),
'preference': preference,
})
self._sort_formats(formats)
2015-08-15 21:07:05 +00:00
title = item['title']
is_live = item['type'] == 'stream'
if is_live:
title = self._live_title(title)
description = self._og_search_description(webpage, default=None) or self._html_search_meta(
'description', webpage, 'description')
2015-08-15 21:07:05 +00:00
timestamp = None
duration = None
if not is_live:
duration = int_or_none(item.get('length'))
timestamp = item.get('published')
if timestamp:
timestamp = parse_iso8601(timestamp[:-5])
return {
'id': video_id,
'title': title,
'description': description,
2015-08-15 21:07:05 +00:00
'thumbnail': item.get('image'),
'duration': duration,
'timestamp': timestamp,
'is_live': is_live,
'formats': formats,
}