mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-22 16:44:32 +01:00
[planetaplay] Add new extractor (closes #3839)
This commit is contained in:
parent
d397c0b3dd
commit
fe506288bd
@ -275,6 +275,7 @@ from .parliamentliveuk import ParliamentLiveUKIE
|
||||
from .patreon import PatreonIE
|
||||
from .pbs import PBSIE
|
||||
from .photobucket import PhotobucketIE
|
||||
from .planetaplay import PlanetaPlayIE
|
||||
from .played import PlayedIE
|
||||
from .playfm import PlayFMIE
|
||||
from .playvid import PlayvidIE
|
||||
|
60
youtube_dl/extractor/planetaplay.py
Normal file
60
youtube_dl/extractor/planetaplay.py
Normal file
@ -0,0 +1,60 @@
|
||||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import ExtractorError
|
||||
|
||||
|
||||
class PlanetaPlayIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?planetaplay\.com/\?sng=(?P<id>[0-9]+)'
|
||||
_API_URL = 'http://planetaplay.com/action/playlist/?sng={0:}'
|
||||
_THUMBNAIL_URL = 'http://planetaplay.com/img/thumb/{thumb:}'
|
||||
_TEST = {
|
||||
'url': 'http://planetaplay.com/?sng=3586',
|
||||
'md5': '9d569dceb7251a4e01355d5aea60f9db',
|
||||
'info_dict': {
|
||||
'id': '3586',
|
||||
'ext': 'flv',
|
||||
'title': 'md5:e829428ee28b1deed00de90de49d1da1',
|
||||
}
|
||||
}
|
||||
|
||||
_SONG_FORMATS = {
|
||||
'lq': (0, 'http://www.planetaplay.com/videoplayback/{med_hash:}'),
|
||||
'hq': (1, 'http://www.planetaplay.com/videoplayback/hi/{med_hash:}'),
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
video_id = mobj.group('id')
|
||||
|
||||
response = self._download_json(
|
||||
self._API_URL.format(video_id), video_id)['response']
|
||||
try:
|
||||
data = response.get('data')[0]
|
||||
except IndexError:
|
||||
raise ExtractorError(
|
||||
'%s: failed to get the playlist' % self.IE_NAME, expected=True)
|
||||
|
||||
title = '{song_artists:} - {sng_name:}'.format(**data)
|
||||
thumbnail = self._THUMBNAIL_URL.format(**data)
|
||||
|
||||
formats = []
|
||||
for format_id, (quality, url_template) in self._SONG_FORMATS.items():
|
||||
formats.append({
|
||||
'format_id': format_id,
|
||||
'url': url_template.format(**data),
|
||||
'quality': quality,
|
||||
'ext': 'flv',
|
||||
})
|
||||
|
||||
self._sort_formats(formats)
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'formats': formats,
|
||||
'thumbnail': thumbnail,
|
||||
}
|
Loading…
Reference in New Issue
Block a user