[audiomack] Better titles, simplify code

This commit is contained in:
Philipp Hagemeister 2015-01-10 02:24:46 +01:00
parent 51897bb77c
commit 904fffffeb

View File

@ -1,11 +1,15 @@
# coding: utf-8 # coding: utf-8
from __future__ import unicode_literals from __future__ import unicode_literals
import itertools
import time
from .common import InfoExtractor from .common import InfoExtractor
from .soundcloud import SoundcloudIE from .soundcloud import SoundcloudIE
from ..utils import ExtractorError from ..utils import (
ExtractorError,
import time url_basename,
)
class AudiomackIE(InfoExtractor): class AudiomackIE(InfoExtractor):
@ -19,7 +23,7 @@ class AudiomackIE(InfoExtractor):
{ {
'id': '310086', 'id': '310086',
'ext': 'mp3', 'ext': 'mp3',
'artist': 'Roosh Williams', 'uploader': 'Roosh Williams',
'title': 'Extraordinary' 'title': 'Extraordinary'
} }
}, },
@ -38,17 +42,6 @@ class AudiomackIE(InfoExtractor):
}, },
] ]
@staticmethod
def create_song_dictionary(api_response, album_url_tag, track_no=0):
# All keys are the same in audiomack api and InfoExtractor format
entry = {key: api_response[key] for key in ['title', 'artist', 'id', 'url'] if key in api_response}
# Fudge values in the face of missing metadata
if 'id' not in entry:
entry['id'] = track_no
if 'title' not in entry:
entry['title'] = album_url_tag
return entry
def _real_extract(self, url): def _real_extract(self, url):
# URLs end with [uploader name]/[uploader title] # URLs end with [uploader name]/[uploader title]
# this title is whatever the user types in, and is rarely # this title is whatever the user types in, and is rarely
@ -70,7 +63,12 @@ def _real_extract(self, url):
if SoundcloudIE.suitable(api_response['url']): if SoundcloudIE.suitable(api_response['url']):
return {'_type': 'url', 'url': api_response['url'], 'ie_key': 'Soundcloud'} return {'_type': 'url', 'url': api_response['url'], 'ie_key': 'Soundcloud'}
return self.create_song_dictionary(api_response, album_url_tag) return {
'id': api_response.get('id', album_url_tag),
'uploader': api_response.get('artist'),
'title': api_response.get('title'),
'url': api_response['url'],
}
class AudiomackAlbumIE(InfoExtractor): class AudiomackAlbumIE(InfoExtractor):
@ -90,7 +88,17 @@ class AudiomackAlbumIE(InfoExtractor):
# Album playlist ripped from fakeshoredrive with no metadata # Album playlist ripped from fakeshoredrive with no metadata
{ {
'url': 'http://www.audiomack.com/album/fakeshoredrive/ppp-pistol-p-project', 'url': 'http://www.audiomack.com/album/fakeshoredrive/ppp-pistol-p-project',
'playlist_count': 10 'playlist': [{
'info_dict': {
'title': '9.-heaven-or-hell-chimaca-ft-zuse-prod-by-dj-fu',
'id': '9.-heaven-or-hell-chimaca-ft-zuse-prod-by-dj-fu',
'ext': 'mp3',
}
}],
'params': {
'playliststart': 8,
'playlistend': 8,
}
} }
] ]
@ -102,11 +110,12 @@ def _real_extract(self, url):
result = {'_type': 'playlist', 'entries': []} result = {'_type': 'playlist', 'entries': []}
# There is no one endpoint for album metadata - instead it is included/repeated in each song's metadata # There is no one endpoint for album metadata - instead it is included/repeated in each song's metadata
# Therefore we don't know how many songs the album has and must infi-loop until failure # Therefore we don't know how many songs the album has and must infi-loop until failure
track_no = 0 for track_no in itertools.count():
while True:
# Get song's metadata # Get song's metadata
api_response = self._download_json('http://www.audiomack.com/api/music/url/album/%s/%d?extended=1&_=%d' api_response = self._download_json(
% (album_url_tag, track_no, time.time()), album_url_tag) 'http://www.audiomack.com/api/music/url/album/%s/%d?extended=1&_=%d'
% (album_url_tag, track_no, time.time()), album_url_tag,
note='Querying song information (%d)' % (track_no + 1))
# Total failure, only occurs when url is totally wrong # Total failure, only occurs when url is totally wrong
# Won't happen in middle of valid playlist (next case) # Won't happen in middle of valid playlist (next case)
@ -120,6 +129,11 @@ def _real_extract(self, url):
for resultkey, apikey in [('id', 'album_id'), ('title', 'album_title')]: for resultkey, apikey in [('id', 'album_id'), ('title', 'album_title')]:
if apikey in api_response and resultkey not in result: if apikey in api_response and resultkey not in result:
result[resultkey] = api_response[apikey] result[resultkey] = api_response[apikey]
result['entries'].append(AudiomackIE.create_song_dictionary(api_response, album_url_tag, track_no)) song_id = url_basename(api_response['url']).rpartition('.')[0]
track_no += 1 result['entries'].append({
'id': api_response.get('id', song_id),
'uploader': api_response.get('artist'),
'title': api_response.get('title', song_id),
'url': api_response['url'],
})
return result return result