1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-26 18:34:32 +01:00

Merge pull request #9395 from pmrowla/afreecatv

[afreecatv] Add new extractor for afreecatv.com VODs
This commit is contained in:
Yen Chi Hsuan 2016-06-09 17:20:16 +08:00
commit e21c26daf9
2 changed files with 134 additions and 0 deletions

View File

@ -0,0 +1,133 @@
# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import (
compat_urllib_parse_urlparse,
compat_urlparse,
)
from ..utils import (
ExtractorError,
int_or_none,
xpath_element,
xpath_text,
)
class AfreecaTVIE(InfoExtractor):
IE_DESC = 'afreecatv.com'
_VALID_URL = r'''(?x)^
https?://(?:(live|afbbs|www)\.)?afreeca(?:tv)?\.com(?::\d+)?
(?:
/app/(?:index|read_ucc_bbs)\.cgi|
/player/[Pp]layer\.(?:swf|html))
\?.*?\bnTitleNo=(?P<id>\d+)'''
_TESTS = [{
'url': 'http://live.afreecatv.com:8079/app/index.cgi?szType=read_ucc_bbs&szBjId=dailyapril&nStationNo=16711924&nBbsNo=18605867&nTitleNo=36164052&szSkin=',
'md5': 'f72c89fe7ecc14c1b5ce506c4996046e',
'info_dict': {
'id': '36164052',
'ext': 'mp4',
'title': '데일리 에이프릴 요정들의 시상식!',
'thumbnail': 're:^https?://(?:video|st)img.afreecatv.com/.*$',
'uploader': 'dailyapril',
'uploader_id': 'dailyapril',
'upload_date': '20160503',
}
}, {
'url': 'http://afbbs.afreecatv.com:8080/app/read_ucc_bbs.cgi?nStationNo=16711924&nTitleNo=36153164&szBjId=dailyapril&nBbsNo=18605867',
'info_dict': {
'id': '36153164',
'title': "BJ유트루와 함께하는 '팅커벨 메이크업!'",
'thumbnail': 're:^https?://(?:video|st)img.afreecatv.com/.*$',
'uploader': 'dailyapril',
'uploader_id': 'dailyapril',
},
'playlist_count': 2,
'playlist': [{
'md5': 'd8b7c174568da61d774ef0203159bf97',
'info_dict': {
'id': '36153164_1',
'ext': 'mp4',
'title': "BJ유트루와 함께하는 '팅커벨 메이크업!'",
'upload_date': '20160502',
},
}, {
'md5': '58f2ce7f6044e34439ab2d50612ab02b',
'info_dict': {
'id': '36153164_2',
'ext': 'mp4',
'title': "BJ유트루와 함께하는 '팅커벨 메이크업!'",
'upload_date': '20160502',
},
}],
}, {
'url': 'http://www.afreecatv.com/player/Player.swf?szType=szBjId=djleegoon&nStationNo=11273158&nBbsNo=13161095&nTitleNo=36327652',
'only_matching': True,
}]
@staticmethod
def parse_video_key(key):
video_key = {}
m = re.match(r'^(?P<upload_date>\d{8})_\w+_(?P<part>\d+)$', key)
if m:
video_key['upload_date'] = m.group('upload_date')
video_key['part'] = m.group('part')
return video_key
def _real_extract(self, url):
video_id = self._match_id(url)
parsed_url = compat_urllib_parse_urlparse(url)
info_url = compat_urlparse.urlunparse(parsed_url._replace(
netloc='afbbs.afreecatv.com:8080',
path='/api/video/get_video_info.php'))
video_xml = self._download_xml(info_url, video_id)
if xpath_element(video_xml, './track/video/file') is None:
raise ExtractorError('Specified AfreecaTV video does not exist',
expected=True)
title = xpath_text(video_xml, './track/title', 'title')
uploader = xpath_text(video_xml, './track/nickname', 'uploader')
uploader_id = xpath_text(video_xml, './track/bj_id', 'uploader id')
duration = int_or_none(xpath_text(video_xml, './track/duration',
'duration'))
thumbnail = xpath_text(video_xml, './track/titleImage', 'thumbnail')
entries = []
for i, video_file in enumerate(video_xml.findall('./track/video/file')):
video_key = self.parse_video_key(video_file.get('key', ''))
if not video_key:
continue
entries.append({
'id': '%s_%s' % (video_id, video_key.get('part', i + 1)),
'title': title,
'upload_date': video_key.get('upload_date'),
'duration': int_or_none(video_file.get('duration')),
'url': video_file.text,
})
info = {
'id': video_id,
'title': title,
'uploader': uploader,
'uploader_id': uploader_id,
'duration': duration,
'thumbnail': thumbnail,
}
if len(entries) > 1:
info['_type'] = 'multi_video'
info['entries'] = entries
elif len(entries) == 1:
info['url'] = entries[0]['url']
info['upload_date'] = entries[0].get('upload_date')
else:
raise ExtractorError(
'No files found for the specified AfreecaTV video, either'
' the URL is incorrect or the video has been made private.',
expected=True)
return info

View File

@ -21,6 +21,7 @@ from .adobetv import (
) )
from .adultswim import AdultSwimIE from .adultswim import AdultSwimIE
from .aenetworks import AENetworksIE from .aenetworks import AENetworksIE
from .afreecatv import AfreecaTVIE
from .aftonbladet import AftonbladetIE from .aftonbladet import AftonbladetIE
from .airmozilla import AirMozillaIE from .airmozilla import AirMozillaIE
from .aljazeera import AlJazeeraIE from .aljazeera import AlJazeeraIE