mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-29 14:45:52 +01:00
252 lines
9.3 KiB
Python
252 lines
9.3 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
clean_html,
|
|
dict_get,
|
|
ExtractorError,
|
|
int_or_none,
|
|
parse_duration,
|
|
try_get,
|
|
update_url_query,
|
|
)
|
|
|
|
|
|
class NaverBaseIE(InfoExtractor):
|
|
_CAPTION_EXT_RE = r'\.(?:ttml|vtt)'
|
|
|
|
def _extract_video_info(self, video_id, vid, key):
|
|
video_data = self._download_json(
|
|
'http://play.rmcnmv.naver.com/vod/play/v2.0/' + vid,
|
|
video_id, query={
|
|
'key': key,
|
|
})
|
|
meta = video_data['meta']
|
|
title = meta['subject']
|
|
formats = []
|
|
get_list = lambda x: try_get(video_data, lambda y: y[x + 's']['list'], list) or []
|
|
|
|
def extract_formats(streams, stream_type, query={}):
|
|
for stream in streams:
|
|
stream_url = stream.get('source')
|
|
if not stream_url:
|
|
continue
|
|
stream_url = update_url_query(stream_url, query)
|
|
encoding_option = stream.get('encodingOption', {})
|
|
bitrate = stream.get('bitrate', {})
|
|
formats.append({
|
|
'format_id': '%s_%s' % (stream.get('type') or stream_type, dict_get(encoding_option, ('name', 'id'))),
|
|
'url': stream_url,
|
|
'width': int_or_none(encoding_option.get('width')),
|
|
'height': int_or_none(encoding_option.get('height')),
|
|
'vbr': int_or_none(bitrate.get('video')),
|
|
'abr': int_or_none(bitrate.get('audio')),
|
|
'filesize': int_or_none(stream.get('size')),
|
|
'protocol': 'm3u8_native' if stream_type == 'HLS' else None,
|
|
})
|
|
|
|
extract_formats(get_list('video'), 'H264')
|
|
for stream_set in video_data.get('streams', []):
|
|
query = {}
|
|
for param in stream_set.get('keys', []):
|
|
query[param['name']] = param['value']
|
|
stream_type = stream_set.get('type')
|
|
videos = stream_set.get('videos')
|
|
if videos:
|
|
extract_formats(videos, stream_type, query)
|
|
elif stream_type == 'HLS':
|
|
stream_url = stream_set.get('source')
|
|
if not stream_url:
|
|
continue
|
|
formats.extend(self._extract_m3u8_formats(
|
|
update_url_query(stream_url, query), video_id,
|
|
'mp4', 'm3u8_native', m3u8_id=stream_type, fatal=False))
|
|
self._sort_formats(formats)
|
|
|
|
replace_ext = lambda x, y: re.sub(self._CAPTION_EXT_RE, '.' + y, x)
|
|
|
|
def get_subs(caption_url):
|
|
if re.search(self._CAPTION_EXT_RE, caption_url):
|
|
return [{
|
|
'url': replace_ext(caption_url, 'ttml'),
|
|
}, {
|
|
'url': replace_ext(caption_url, 'vtt'),
|
|
}]
|
|
else:
|
|
return [{'url': caption_url}]
|
|
|
|
automatic_captions = {}
|
|
subtitles = {}
|
|
for caption in get_list('caption'):
|
|
caption_url = caption.get('source')
|
|
if not caption_url:
|
|
continue
|
|
sub_dict = automatic_captions if caption.get('type') == 'auto' else subtitles
|
|
sub_dict.setdefault(dict_get(caption, ('locale', 'language')), []).extend(get_subs(caption_url))
|
|
|
|
user = meta.get('user', {})
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'subtitles': subtitles,
|
|
'automatic_captions': automatic_captions,
|
|
'thumbnail': try_get(meta, lambda x: x['cover']['source']),
|
|
'view_count': int_or_none(meta.get('count')),
|
|
'uploader_id': user.get('id'),
|
|
'uploader': user.get('name'),
|
|
'uploader_url': user.get('url'),
|
|
}
|
|
|
|
|
|
class NaverIE(NaverBaseIE):
|
|
_VALID_URL = r'https?://(?:m\.)?tv(?:cast)?\.naver\.com/(?:v|embed)/(?P<id>\d+)'
|
|
_GEO_BYPASS = False
|
|
_TESTS = [{
|
|
'url': 'http://tv.naver.com/v/81652',
|
|
'info_dict': {
|
|
'id': '81652',
|
|
'ext': 'mp4',
|
|
'title': '[9월 모의고사 해설강의][수학_김상희] 수학 A형 16~20번',
|
|
'description': '메가스터디 수학 김상희 선생님이 9월 모의고사 수학A형 16번에서 20번까지 해설강의를 공개합니다.',
|
|
'timestamp': 1378200754,
|
|
'upload_date': '20130903',
|
|
'uploader': '메가스터디, 합격불변의 법칙',
|
|
'uploader_id': 'megastudy',
|
|
},
|
|
}, {
|
|
'url': 'http://tv.naver.com/v/395837',
|
|
'md5': '8a38e35354d26a17f73f4e90094febd3',
|
|
'info_dict': {
|
|
'id': '395837',
|
|
'ext': 'mp4',
|
|
'title': '9년이 지나도 아픈 기억, 전효성의 아버지',
|
|
'description': 'md5:eb6aca9d457b922e43860a2a2b1984d3',
|
|
'timestamp': 1432030253,
|
|
'upload_date': '20150519',
|
|
'uploader': '4가지쇼 시즌2',
|
|
'uploader_id': 'wrappinguser29',
|
|
},
|
|
'skip': 'Georestricted',
|
|
}, {
|
|
'url': 'http://tvcast.naver.com/v/81652',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
content = self._download_json(
|
|
'https://tv.naver.com/api/json/v/' + video_id,
|
|
video_id, headers=self.geo_verification_headers())
|
|
player_info_json = content.get('playerInfoJson') or {}
|
|
current_clip = player_info_json.get('currentClip') or {}
|
|
|
|
vid = current_clip.get('videoId')
|
|
in_key = current_clip.get('inKey')
|
|
|
|
if not vid or not in_key:
|
|
player_auth = try_get(player_info_json, lambda x: x['playerOption']['auth'])
|
|
if player_auth == 'notCountry':
|
|
self.raise_geo_restricted(countries=['KR'])
|
|
elif player_auth == 'notLogin':
|
|
self.raise_login_required()
|
|
raise ExtractorError('couldn\'t extract vid and key')
|
|
info = self._extract_video_info(video_id, vid, in_key)
|
|
info.update({
|
|
'description': clean_html(current_clip.get('description')),
|
|
'timestamp': int_or_none(current_clip.get('firstExposureTime'), 1000),
|
|
'duration': parse_duration(current_clip.get('displayPlayTime')),
|
|
'like_count': int_or_none(current_clip.get('recommendPoint')),
|
|
'age_limit': 19 if current_clip.get('adult') else None,
|
|
})
|
|
return info
|
|
|
|
|
|
class NaverLiveIE(InfoExtractor):
|
|
IE_NAME = 'Naver:live'
|
|
_VALID_URL = r'https?://(?:m\.)?tv(?:cast)?\.naver\.com/l/(?P<id>\d+)'
|
|
_GEO_BYPASS = False
|
|
_TESTS = [{
|
|
'url': 'https://tv.naver.com/l/52010',
|
|
'info_dict': {
|
|
'id': '52010',
|
|
'ext': 'm3u8',
|
|
'title': '[LIVE] 뉴스특보 : "수도권 거리두기, 2주간 2단계로 조정"',
|
|
'description': 'md5:df7f0c237a5ed5e786ce5c91efbeaab3',
|
|
'channel_id': 'NTV-ytnnews24-0',
|
|
'start_time': 1597026780000,
|
|
},
|
|
}, {
|
|
'url': 'https://tv.naver.com/l/51549',
|
|
'info_dict': {
|
|
'id': '51549',
|
|
'ext': 'm3u8',
|
|
'title': '연합뉴스TV - 코로나19 뉴스특보',
|
|
'description': 'md5:c655e82091bc21e413f549c0eaccc481',
|
|
'channel_id': 'NTV-yonhapnewstv-0',
|
|
'start_time': 1596406380000,
|
|
},
|
|
}, {
|
|
'url': 'https://tv.naver.com/l/54887',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
page = self._download_webpage(url, video_id, 'Downloading Page', 'Unable to download Page')
|
|
secure_url = self._search_regex(r'sApiF:\s+(?:"|\')([^"\']+)', page, 'secureurl')
|
|
|
|
info = self._extract_video_info(video_id, secure_url)
|
|
info.update({
|
|
'description': self._og_search_description(page)
|
|
})
|
|
|
|
return info
|
|
|
|
def _extract_video_info(self, video_id, url):
|
|
video_data = self._download_json(url, video_id, headers=self.geo_verification_headers())
|
|
meta = video_data.get('meta')
|
|
status = meta.get('status')
|
|
|
|
if status == 'CLOSED':
|
|
raise ExtractorError('Stream is offline.', expected=True)
|
|
elif status != 'OPENED':
|
|
raise ExtractorError('Unknown status %s' % status)
|
|
|
|
title = meta.get('title')
|
|
stream_list = video_data.get('streams')
|
|
|
|
if stream_list is None:
|
|
raise ExtractorError('Could not get stream data.', expected=True)
|
|
|
|
formats = []
|
|
for quality in stream_list:
|
|
if not quality.get('url'):
|
|
continue
|
|
|
|
prop = quality.get('property')
|
|
if prop.get('abr'): # This abr doesn't mean Average audio bitrate.
|
|
continue
|
|
|
|
formats.extend(self._extract_m3u8_formats(
|
|
quality.get('url'), video_id, 'm3u8',
|
|
m3u8_id=quality.get('qualityId'), live=True
|
|
))
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'formats': formats,
|
|
'channel_id': meta.get('channelId'),
|
|
'channel_url': meta.get('channelUrl'),
|
|
'thumbnail': meta.get('imgUrl'),
|
|
'start_time': meta.get('startTime'),
|
|
'categories': [meta.get('categoryId')],
|
|
'is_live': True
|
|
}
|