mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-22 02:01:50 +00:00
[22tracks] Remove extractor (closes #11024)
This commit is contained in:
parent
44cca168cc
commit
cd670befc4
2 changed files with 0 additions and 90 deletions
|
@ -1110,10 +1110,6 @@ from .tvplayer import TVPlayerIE
|
|||
from .tweakers import TweakersIE
|
||||
from .twentyfourvideo import TwentyFourVideoIE
|
||||
from .twentymin import TwentyMinutenIE
|
||||
from .twentytwotracks import (
|
||||
TwentyTwoTracksIE,
|
||||
TwentyTwoTracksGenreIE
|
||||
)
|
||||
from .twitch import (
|
||||
TwitchVideoIE,
|
||||
TwitchChapterIE,
|
||||
|
|
|
@ -1,86 +0,0 @@
|
|||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import int_or_none
|
||||
|
||||
# 22Tracks regularly replace the audio tracks that can be streamed on their
|
||||
# site. The tracks usually expire after 1 months, so we can't add tests.
|
||||
|
||||
|
||||
class TwentyTwoTracksIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://22tracks\.com/(?P<city>[a-z]+)/(?P<genre>[\da-z]+)/(?P<id>\d+)'
|
||||
IE_NAME = '22tracks:track'
|
||||
|
||||
_API_BASE = 'http://22tracks.com/api'
|
||||
|
||||
def _extract_info(self, city, genre_name, track_id=None):
|
||||
item_id = track_id if track_id else genre_name
|
||||
|
||||
cities = self._download_json(
|
||||
'%s/cities' % self._API_BASE, item_id,
|
||||
'Downloading cities info',
|
||||
'Unable to download cities info')
|
||||
city_id = [x['id'] for x in cities if x['slug'] == city][0]
|
||||
|
||||
genres = self._download_json(
|
||||
'%s/genres/%s' % (self._API_BASE, city_id), item_id,
|
||||
'Downloading %s genres info' % city,
|
||||
'Unable to download %s genres info' % city)
|
||||
genre = [x for x in genres if x['slug'] == genre_name][0]
|
||||
genre_id = genre['id']
|
||||
|
||||
tracks = self._download_json(
|
||||
'%s/tracks/%s' % (self._API_BASE, genre_id), item_id,
|
||||
'Downloading %s genre tracks info' % genre_name,
|
||||
'Unable to download track info')
|
||||
|
||||
return [x for x in tracks if x['id'] == item_id][0] if track_id else [genre['title'], tracks]
|
||||
|
||||
def _get_track_url(self, filename, track_id):
|
||||
token = self._download_json(
|
||||
'http://22tracks.com/token.php?desktop=true&u=/128/%s' % filename,
|
||||
track_id, 'Downloading token', 'Unable to download token')
|
||||
return 'http://audio.22tracks.com%s?st=%s&e=%d' % (token['filename'], token['st'], token['e'])
|
||||
|
||||
def _extract_track_info(self, track_info, track_id):
|
||||
download_url = self._get_track_url(track_info['filename'], track_id)
|
||||
title = '%s - %s' % (track_info['artist'].strip(), track_info['title'].strip())
|
||||
return {
|
||||
'id': track_id,
|
||||
'url': download_url,
|
||||
'ext': 'mp3',
|
||||
'title': title,
|
||||
'duration': int_or_none(track_info.get('duration')),
|
||||
'timestamp': int_or_none(track_info.get('published_at') or track_info.get('created'))
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
|
||||
city = mobj.group('city')
|
||||
genre = mobj.group('genre')
|
||||
track_id = mobj.group('id')
|
||||
|
||||
track_info = self._extract_info(city, genre, track_id)
|
||||
return self._extract_track_info(track_info, track_id)
|
||||
|
||||
|
||||
class TwentyTwoTracksGenreIE(TwentyTwoTracksIE):
|
||||
_VALID_URL = r'https?://22tracks\.com/(?P<city>[a-z]+)/(?P<genre>[\da-z]+)/?$'
|
||||
IE_NAME = '22tracks:genre'
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
|
||||
city = mobj.group('city')
|
||||
genre = mobj.group('genre')
|
||||
|
||||
genre_title, tracks = self._extract_info(city, genre)
|
||||
|
||||
entries = [
|
||||
self._extract_track_info(track_info, track_info['id'])
|
||||
for track_info in tracks]
|
||||
|
||||
return self.playlist_result(entries, genre, genre_title)
|
Loading…
Reference in a new issue