1
1
mirror of https://github.com/ytdl-org/youtube-dl synced 2024-12-15 01:37:48 +01:00
youtube-dl/youtube_dl/extractor/soundgasm.py
Philipp Hagemeister 591ab1dff9 [soundgasm] PEP8
2015-02-23 16:51:21 +01:00

65 lines
2.0 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
class SoundgasmIE(InfoExtractor):
IE_NAME = 'soundgasm'
_VALID_URL = r'https?://(?:www\.)?soundgasm\.net/u/(?P<user>[0-9a-zA-Z_\-]+)/(?P<title>[0-9a-zA-Z_\-]+)'
_TEST = {
'url': 'http://soundgasm.net/u/ytdl/Piano-sample',
'md5': '010082a2c802c5275bb00030743e75ad',
'info_dict': {
'id': '88abd86ea000cafe98f96321b23cc1206cbcbcc9',
'ext': 'm4a',
'title': 'ytdl_Piano-sample',
'description': 'Royalty Free Sample Music'
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
display_id = mobj.group('title')
audio_title = mobj.group('user') + '_' + mobj.group('title')
webpage = self._download_webpage(url, display_id)
audio_url = self._html_search_regex(
r'(?s)m4a\:\s"([^"]+)"', webpage, 'audio URL')
audio_id = re.split('\/|\.', audio_url)[-2]
description = self._html_search_regex(
r'(?s)<li>Description:\s(.*?)<\/li>', webpage, 'description',
fatal=False)
return {
'id': audio_id,
'display_id': display_id,
'url': audio_url,
'title': audio_title,
'description': description
}
class SoundgasmProfileIE(InfoExtractor):
IE_NAME = 'soundgasm:profile'
_VALID_URL = r'https?://(?:www\.)?soundgasm\.net/u/(?P<id>[^/]+)/?(?:\#.*)?$'
_TEST = {
'url': 'http://soundgasm.net/u/ytdl',
'info_dict': {
'id': 'ytdl',
},
'playlist_count': 1,
}
def _real_extract(self, url):
profile_id = self._match_id(url)
webpage = self._download_webpage(url, profile_id)
entries = [
self.url_result(audio_url, 'Soundgasm')
for audio_url in re.findall(r'href="([^"]+/u/%s/[^"]+)' % profile_id, webpage)]
return self.playlist_result(entries, profile_id)