# encoding: UTF-8
# api: streamtuner2
# title: SurfMusik
# description: User collection of streams categorized by region and genre.
# version: 0.5
# type: channel
# category: radio
# author: gorgonz123
# source: http://forum.ubuntuusers.de/topic/streamtuner2-zwei-internet-radios-anhoeren-au/3/
# recognizes: max_streams
#
# This plugin comes in German (SurfMusik) and English (SurfMusic) variations.
# It provides a vast collection of international stations and genres.
# While it's not an open source project, most entries are user contributed.
#
# They do have a Windows client, hencewhy it's even more important for
# streamtuner2 to support it on other plattforms.
#
# TV stations don't seem to work mostly. And loading the webtv/ pages would
# be somewhat slow (for querying the actual mms:// streams).
#
#
#
import re
import ahttp as http
from config import conf, dbg, __print__
from channels import *
# Surfmusik sharing site
class surfmusik (ChannelPlugin):
# description
title = "SurfMusik"
module = "surfmusik"
homepage = "http://www.surfmusik.de/"
listformat = "audio/x-scpls"
lang = "DE" # last configured categories
base = {
"DE": ("http://www.surfmusik.de/", "genre/", "land/"),
"EN": ("http://www.surfmusic.de/", "format/", "country/"),
}
categories = []
titles = dict( genre="Genre", title="Station", playing="Location", bitrate=False, listeners=False )
config = [
{
"name": "surfmusik_lang",
"value": "EN",
"type": "select",
"select":"DE=German|EN=English",
"description": "Switching to a new category title language requires reloading the category tree.",
"category": "language",
}
]
# Set channel title
def __init__(self, parent=None):
self.title = ("SurfMusik", "SurfMusic")[conf.get("surfmusik_lang", "EN") == "EN"]
ChannelPlugin.__init__(self, parent)
# just a static list for now
def update_categories(self):
lang = conf.surfmusik_lang
(base_url, path_genre, path_country) = self.base[lang]
cats = {
"DE": ["Genres", "Deutschland", "Europa", "USA", "Kanada", "Amerika", "Afrika", "Asien", "Ozeanien", "MusicTV", "NewsTV", "Poli", "Flug"],
"EN": ["Genres", "Europe", "Germany", "USA", "Canada", "America", "Africa", "Asia", "Oceania", "MusicTV", "NewsTV", "Poli", "Flug"],
}
map = {
"Genres": "genres.htm",
"Europe": "euro.htm", "Europa": "euro.htm",
"Germany": "bundesland.htm", "Deutschland": "bundesland.htm",
"Africa": "africa.htm", "Afrika": "africa.htm",
"America": "amerika.htm", "Amerika": "amerika.htm",
"Asia": "asien.htm", "Asien": "asien.htm",
"Oceania": "ozean.htm", "Ozeanien": "ozean.htm",
"Canada": "canadian-radio-stations.htm", "Kanada": "kanada-online-radio.htm",
"USA": "staaten.htm",
}
rx_links = re.compile(r"""
<a\b [^>]+ \b href="
(?:(?:https?:)?//www.surfmusi[kc].de)? /?
(?:land|country|genre|format)/
([\-+\w\d\s%]+) \.html"
""", re.X)
r = []
# Add main categories, and fetch subentries (genres or country names)
for cat in cats[lang]:
r.append(cat)
if map.get(cat):
subcats = rx_links.findall( http.get(base_url + map[cat]) )
subcats = [x.replace("+", " ").title() for x in subcats]
r.append(sorted(subcats))
self.categories = r
# summarize links from surfmusik
def update_streams(self, cat):
(base_url, path_genre, path_country) = self.base[conf.surfmusik_lang]
entries = []
i = 0
max = int(conf.max_streams)
is_tv = 0
# placeholder category
if cat in ["Genres"]:
path = None
# separate
elif cat in ["Poli", "Flug"]:
path = ""
# tv
elif cat in ["SurfTV", "MusikTV", "NewsTV"]:
path = ""
is_tv = 1
# genre
elif cat in self.categories[self.categories.index("Genres") + 1]:
path = path_genre
# country
else:
path = path_country
if path is not None:
ucat = cat.replace(" ", "+").lower()
html = http.get(base_url + path + ucat + ".html")
html = re.sub("&#x?\d+;", "", html)
rx_radio = re.compile(r"""
<td\s+class="home1"><a[^>]*\s+href="(.+?)"[^>]*> .*?
<a\s+class="navil"\s+href="([^"]+)"[^>]*>([^<>]+)</a></td>
<td\s+class="ort">(.*?)</td>.*?
<td\s+class="ort">(.*?)</td>.*?
""", re.X|re.I)
rx_video = re.compile(r"""
<a[^>]+href="([^"]+)"[^>]*>(?:<[^>]+>)*Externer
""", re.X|re.I)
# per-country list
for uu in rx_radio.findall(html):
(url, homepage, name, genre, city) = uu
# find mms:// for webtv stations
if is_tv:
m = rx_video.search(http.get(url))
if m:
url = m.group(1)
# just convert /radio/ into /m3u/ link
else:
url = "http://www.surfmusik.de/m3u/" + url[30:-5] + ".m3u"
entries.append({
"title": name,
"homepage": homepage,
"url": url,
"playing": city,
"genre": genre,
"format": ("video/html" if is_tv else "audio/mpeg"),
})
# limit result list
if i > max:
break
if i % 10 == 0:
self.parent.status(float(i)/float(max+5))
i += 1
# done
return entries