# api: streamtunter2
# title: Live365
# description: Around 5000 categorized internet radio streams, some paid ad-free ones.
# version: 0.3
# type: channel
# category: radio
# url: http://www.live365.com/
# config: -
# priority: optional
# png:
# iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAABHNCSVQICAgIfAhkiAAAAzlJREFUOI1lk0tMXGUUx3/fvTP3zgyPGR5SKIQODdDGEtoCJdFopA9SJTFMjInGaJDEulWTGg2JApqoO+PShaAmdVEXQAxoCtgKNjYFWrQgMD6KDAylkde8Lnfunfu5IJii
# Z3POSc7//M/i/AT/ichrbUE1NyckfN5WV0kpUjpYS4s4sdhAZnOrP9jTv/jgvHiw+ftSb5f2UEGnfuQYCIG9fg8pJa5AIdJIEf9hhNTkze5gz9ddexrXXrE1+E2v7/TZl9lcx/hxBCu5DUikzCDTFjiSrMYnyGp8vPOuaR+quNTXDqACRLve7cp5suV1YaSwpyewYts4
# dgbhcuGprkFoOnZ0BTM8j9A85J576sQr96Pik4Xfr6l/nj8b9D7yaJ8rz8/O2FWctEXGTKP6/eQ924ZeUY21sowViSDTaaylJbSKSrCsplfNnS9cMiNC3ppayEhyX2wHAdt9l8k+cx6h68SGhzCmJ9CP1hAIPYdjGGQ21lF82cSHR0MuxzBbFU0js7FOOhzGU1tL9ulm
# 3IUHAMhtbsFbc5yd2TsAJEavIC0b78k6sGWrohUfxJybJzE8gpoXAMBeWwPAWAqTvD6Ou6QULXgYgOwzzRi3bpMcv47n4WMo6B4yWzHM+TDGz7suyN2k5x9Er6zCSSSIDQ6x+tabKD4f7tIynHgcK3oPZf3mDfSqStSCQnZmf91V2hmiFy9iLixgLa+wMTmIOTeHq6gY
# AOuvCHp1NcmFMK6ERx0wEvGmou5OtLJSrJUV7n/wEWY4TOLKCHsnlfV8hq+hgc3eL4l9O4y7sYGErgyIofqaYEHdqbvljzWx+v6HaI5ERSCE2Pen0pE4Wh22NUXxOx1Exq+yOTVRobRMzSxGZqa6zUI/3uefYcOrsq0JEqokhcRAkhSSuBti+gw5F9qxSw+wuRrpbp6e
# XVQBLi+vXatfngseeuGlE/5TjUQX7pCSNoZbkNIUUrqCWRCg8u0OKC9h/L2Oz0OjP73xP5g+PXeyq+R4Q+eR5qfJ8maz/ds8AIGqoySNBMs3xvhjbKT7wvfT/8K0bwHAxy31QcWRIaA1v/ww3kA+0V8mAQb8+UX9bV99tw/nfwAe2WTAAcikxQAAAABJRU5ErkJggg==
#
# Live365 lists around 5000 radio stations. Some are paid
# entries and require a logon. This plugins tries to filter
# thoise out.
# streamtuner2 modules
from config import conf
from uikit import uikit
import ahttp as http
from channels import *
from config import __print__, dbg
import action
# python modules
import re
import xml.dom.minidom
from xml.sax.saxutils import unescape as entity_decode, escape as xmlentities
import gtk
import copy
import urllib
from itertools import groupby
from time import time
from xml.dom.minidom import parseString
# channel live365
#
# We're currently extracting from the JavaScript;
#
# stn.set("param", "value");
#
# And using a HTML5 player direct URL now:
#
# /cgi-bin/play.pls?stationid=%s&direct=1&file=%s.pls
#
class live365(ChannelPlugin):
# desc
module = "live365"
title = "Live365"
homepage = "http://www.live365.com/"
base_url = "http://www.live365.com/"
has_search = True
listformat = "pls"
mediatype = "audio/mpeg"
has_search = False
# content
categories = ['Alternative', 'Blues', 'Classical', 'Country', 'Easy Listening', 'Electronic/Dance', 'Folk', 'Freeform', 'Hip-Hop/Rap', 'Inspirational', 'International', 'Jazz', 'Latin', 'Metal', 'New Age', 'Oldies', 'Pop', 'R&B/Urban', 'Reggae', 'Rock', 'Seasonal/Holiday', 'Soundtracks', 'Talk']
current = "Alternative"
default = "Pop"
empty = None
# redefine
streams = {}
def __init__(self, parent=None):
# override datamap fields //@todo: might need a cleaner method, also: do we really want the stream data in channels to be different/incompatible?
self.datamap = copy.deepcopy(self.datamap)
self.datamap[5][0] = "Rating"
self.datamap[5][2][0] = "rating"
# superclass
ChannelPlugin.__init__(self, parent)
# fixed for now
def update_categories(self):
pass
# extract stream infos
def update_streams(self, cat):
# Retrieve genere index pages
html = ""
for i in [1, 17, 33, 49]:
url = "http://www.live365.com/cgi-bin/directory.cgi?first=%i&site=web&mode=3&genre=%s&charset=UTF-8&target=content" % (i, cat.lower())
html += http.get(url, feedback=self.parent.status)
# Extract from JavaScript
rx = re.compile(r"""
stn.set\( " (\w+) ", \s+ " ((?:[^"\\]+|\\.)*) "\); \s+
""", re.X|re.I|re.S|re.M)
# Group entries before adding them
ls = []
for i,row in groupby(rx.findall(html), self.group_by_station):
row = dict(row)
ls.append({
"status": (None if row["listenerAccess"] == "PUBLIC" else gtk.STOCK_STOP),
"deleted": row["status"] != "OK",
"name": row["stationName"],
"title": row["title"],
"playing": "n/a",
"id": row["id"],
"access": row["listenerAccess"],
"status": row["status"],
"mode": row["serverMode"],
"rating": int(row["rating"]),
#"rating": row["ratingCount"],
"listeners": int(row["tlh"]),
"location": row["location"],
"favicon": row["imgUrl"],
"format": self.mediatype,
"url": "%scgi-bin/play.pls?stationid=%s&direct=1&file=%s.pls" % (self.base_url, row["id"], row["stationName"])
})
return ls
# itertools.groupby filter
gi = 0
def group_by_station(self, kv):
if kv[0] == "stationName":
self.gi += 1
return self.gi