PlexKodiConnect/resources/lib/PlexFunctions.py
2016-03-24 18:52:02 +01:00

491 lines
14 KiB
Python

# -*- coding: utf-8 -*-
from urllib import urlencode
from ast import literal_eval
from urlparse import urlparse, parse_qs
import re
from copy import deepcopy
import requests
from xbmcaddon import Addon
import downloadutils
from utils import logMsg, settings
# Disable requests logging
from requests.packages.urllib3.exceptions import InsecureRequestWarning
requests.packages.urllib3.disable_warnings(InsecureRequestWarning)
addonName = Addon().getAddonInfo('name')
title = "%s %s" % (addonName, __name__)
def PlexToKodiTimefactor():
"""
Kodi measures time in seconds, but Plex in milliseconds
"""
return 1.0 / 1000.0
def ConvertPlexToKodiTime(plexTime):
"""
Converts Plextime to Koditime. Returns an int (in seconds).
"""
if plexTime is None:
return None
return int(float(plexTime) * PlexToKodiTimefactor())
def GetItemClassFromType(itemType):
classes = {
'movie': 'Movies',
'season': 'TVShows',
'episode': 'TVShows',
'show': 'TVShows',
'artist': 'Music',
'album': 'Music',
'track': 'Music'
}
return classes[itemType]
def GetItemClassFromNumber(itemType):
classes = {
1: 'Movies',
4: 'TVShows',
}
return classes[itemType]
def GetKodiTypeFromPlex(plexItemType):
"""
As used in playlist.item here: http://kodi.wiki/view/JSON-RPC_API
"""
classes = {
'movie': 'movie',
'episode': 'episode',
'track': 'song',
'artist': 'artist',
'album': 'album',
'XXXXXX': 'musicvideo',
'XXXXXXX': 'genre'
}
return classes[plexItemType.lower()]
def GetPlexKeyNumber(plexKey):
"""
Deconstructs e.g. '/library/metadata/xxxx' to the tuple
('library/metadata', 'xxxx')
Returns ('','') if nothing is found
"""
regex = re.compile(r'''/(.+)/(\d+)$''')
try:
result = regex.findall(plexKey)[0]
except IndexError:
result = ('', '')
return result
def ParseContainerKey(containerKey):
"""
Parses e.g. /playQueues/3045?own=1&repeat=0&window=200 to:
'playQueues', '3045', {'window': ['200'], 'own': ['1'], 'repeat': ['0']}
Output hence: library, key, query (query as a special dict)
"""
result = urlparse(containerKey)
library, key = GetPlexKeyNumber(result.path)
query = parse_qs(result.query)
return library, key, query
def LiteralEval(string):
"""
Turns a string e.g. in a dict, safely :-)
"""
return literal_eval(string)
def GetMethodFromPlexType(plexType):
methods = {
'movie': 'add_update',
'episode': 'add_updateEpisode',
'show': 'add_update',
'season': 'add_updateSeason',
'track': 'add_updateSong',
'album': 'add_updateAlbum',
'artist': 'add_updateArtist'
}
return methods[plexType]
def XbmcItemtypes():
return ['photo', 'video', 'audio']
def PlexItemtypes():
return ['photo', 'video', 'audio']
def PlexLibraryItemtypes():
return ['movie', 'show']
# later add: 'artist', 'photo'
def EmbyItemtypes():
return ['Movie', 'Series', 'Season', 'Episode']
def SelectStreams(url, args):
"""
Does a PUT request to tell the PMS what audio and subtitle streams we have
chosen.
"""
downloadutils.DownloadUtils().downloadUrl(
url + '?' + urlencode(args), type='PUT')
def GetPlayQueue(playQueueID):
"""
Fetches the PMS playqueue with the playQueueID as an XML
Returns None if something went wrong
"""
url = "{server}/playQueues/%s" % playQueueID
args = {'Accept': 'application/xml'}
xml = downloadutils.DownloadUtils().downloadUrl(url, headerOptions=args)
try:
xml.attrib['playQueueID']
except (AttributeError, KeyError):
return None
return xml
def GetPlexMetadata(key):
"""
Returns raw API metadata for key as an etree XML.
Can be called with either Plex key '/library/metadata/xxxx'metadata
OR with the digits 'xxxx' only.
Returns None if something went wrong
"""
key = str(key)
if '/library/metadata/' in key:
url = "{server}" + key
else:
url = "{server}/library/metadata/" + key
arguments = {
'checkFiles': 1, # No idea
'includeExtras': 1, # Trailers and Extras => Extras
# 'includeRelated': 1, # Similar movies => Video -> Related
# 'includeRelatedCount': 5,
# 'includeOnDeck': 1,
'includeChapters': 1,
'includePopularLeaves': 1,
'includeConcerts': 1
}
url = url + '?' + urlencode(arguments)
xml = downloadutils.DownloadUtils().downloadUrl(url)
# Did we receive a valid XML?
try:
xml.attrib
# Nope we did not receive a valid XML
except AttributeError:
logMsg(title, "Error retrieving metadata for %s" % url, -1)
xml = None
return xml
def GetAllPlexChildren(key, containerSize=None):
"""
Returns a list (raw xml API dump) of all Plex children for the key.
(e.g. /library/metadata/194853/children pointing to a season)
Input:
key Key to a Plex item, e.g. 12345
"""
url = "{server}/library/metadata/%s/children?" % key
return DownloadChunks(url, containerSize)
def GetPlexSectionResults(viewId, args=None, containerSize=None):
"""
Returns a list (XML API dump) of all Plex items in the Plex
section with key = viewId.
Input:
args: optional dict to be urlencoded
Returns None if something went wrong
"""
url = "{server}/library/sections/%s/all?" % viewId
if args:
url += urlencode(args) + '&'
return DownloadChunks(url, containerSize)
def DownloadChunks(url, containerSize):
"""
Downloads PMS url in chunks of containerSize (int).
If containerSize is None: ONE xml is fetched directly
url MUST end with '?' (if no other url encoded args are present) or '&'
Returns a stitched-together xml or None.
"""
if containerSize is None:
# Get rid of '?' or '&' at the end of url
xml = downloadutils.DownloadUtils().downloadUrl(url[:-1])
try:
xml.attrib
except AttributeError:
# Nope, not an XML, abort
logMsg(title, "Error getting url %s" % url[:-1], -1)
return None
else:
return xml
xml = None
pos = 0
errorCounter = 0
while errorCounter < 10:
args = {
'X-Plex-Container-Size': containerSize,
'X-Plex-Container-Start': pos
}
xmlpart = downloadutils.DownloadUtils().downloadUrl(
url + urlencode(args))
# If something went wrong - skip in the hope that it works next time
try:
xmlpart.attrib
except AttributeError:
logMsg(title, 'Error while downloading chunks: %s'
% (url + urlencode(args)), -1)
pos += containerSize
errorCounter += 1
continue
# Very first run: starting xml (to retain data in xml's root!)
if xml is None:
xml = deepcopy(xmlpart)
if len(xmlpart) < containerSize:
break
else:
pos += containerSize
continue
# Build answer xml - containing the entire library
for child in xmlpart:
xml.append(child)
# Done as soon as we don't receive a full complement of items
if len(xmlpart) < containerSize:
break
pos += containerSize
if errorCounter == 10:
logMsg(title, 'Fatal error while downloading chunks for %s' % url, -1)
return None
return xml
def GetAllPlexLeaves(viewId, lastViewedAt=None, updatedAt=None,
containerSize=None):
"""
Returns a list (raw XML API dump) of all Plex subitems for the key.
(e.g. /library/sections/2/allLeaves pointing to all TV shows)
Input:
viewId Id of Plex library, e.g. '2'
lastViewedAt Unix timestamp; only retrieves PMS items viewed
since that point of time until now.
updatedAt Unix timestamp; only retrieves PMS items updated
by the PMS since that point of time until now.
containerSize Number of items simultaneously fetched from PMS
If lastViewedAt and updatedAt=None, ALL PMS items are returned.
Warning: lastViewedAt and updatedAt are combined with AND by the PMS!
Relevant "master time": PMS server. I guess this COULD lead to problems,
e.g. when server and client are in different time zones.
"""
args = []
url = "{server}/library/sections/%s/allLeaves" % viewId
if lastViewedAt:
args.append('lastViewedAt>=%s' % lastViewedAt)
if updatedAt:
args.append('updatedAt>=%s' % updatedAt)
if args:
url += '?' + '&'.join(args) + '&'
else:
url += '?'
return DownloadChunks(url, containerSize)
def GetPlexOnDeck(viewId, containerSize=None):
"""
"""
url = "{server}/library/sections/%s/onDeck?" % viewId
return DownloadChunks(url, containerSize)
def GetPlexCollections(mediatype):
"""
Input:
mediatype String or list of strings with possible values
'movie', 'show', 'artist', 'photo'
Output:
List with an entry of the form:
{
'name': xxx Plex title for the media section
'type': xxx Plex type: 'movie', 'show', 'artist', 'photo'
'id': xxx Plex unique key for the section (1, 2, 3...)
'uuid': xxx Other unique Plex key, e.g.
74aec9f2-a312-4723-9436-de2ea43843c1
}
Returns an empty list if nothing is found.
"""
collections = []
url = "{server}/library/sections"
xml = downloadutils.DownloadUtils().downloadUrl(url)
try:
xml.attrib
except AttributeError:
logMsg(title, 'Could not download PMS sections for %s' % url, -1)
return {}
for item in xml:
contentType = item['type']
if contentType in mediatype:
name = item['title']
contentId = item['key']
uuid = item['uuid']
collections.append({
'name': name,
'type': contentType,
'id': str(contentId),
'uuid': uuid
})
return collections
def GetPlexPlaylist(itemid, librarySectionUUID, mediatype='movie'):
"""
Returns raw API metadata XML dump for a playlist with e.g. trailers.
"""
trailerNumber = settings('trailerNumber')
if not trailerNumber:
trailerNumber = '3'
url = "{server}/playQueues"
args = {
'type': mediatype,
'uri': 'library://' + librarySectionUUID +
'/item/%2Flibrary%2Fmetadata%2F' + itemid,
'includeChapters': '1',
'extrasPrefixCount': trailerNumber,
'shuffle': '0',
'repeat': '0'
}
xml = downloadutils.DownloadUtils().downloadUrl(
url + '?' + urlencode(args), type="POST")
try:
xml[0].tag
except (IndexError, TypeError, AttributeError):
logMsg(title, "Error retrieving metadata for %s" % url, -1)
return None
return xml
def getPlexRepeat(kodiRepeat):
plexRepeat = {
'off': '0',
'one': '1',
'all': '2' # does this work?!?
}
return plexRepeat.get(kodiRepeat)
def PMSHttpsEnabled(url):
"""
Returns True if the PMS wants to talk https, False otherwise. None if error
occured, e.g. the connection timed out
With with e.g. url=192.168.0.1:32400 (NO http/https)
This is done by GET /identity (returns an error if https is enabled and we
are trying to use http)
Prefers HTTPS over HTTP
"""
# True if https, False if http
answer = True
try:
# Don't use downloadutils here, otherwise we may get un-authorized!
res = requests.get('https://%s/identity' % url,
headers={},
verify=False,
timeout=(3, 10))
# Don't verify SSL since we can connect for sure then!
except requests.exceptions.ConnectionError as e:
# Might have SSL deactivated. Try with http
try:
res = requests.get('http://%s/identity' % url,
headers={},
timeout=(3, 10))
except requests.exceptions.ConnectionError as e:
logMsg(title, "Server is offline or cannot be reached. Url: %s"
", Error message: %s" % (url, e), -1)
return None
except requests.exceptions.ReadTimeout:
logMsg(title, "Server timeout reached for Url %s" % url, -1)
return None
else:
answer = False
except requests.exceptions.ReadTimeout:
logMsg(title, "Server timeout reached for Url %s" % url, -1)
return None
if res.status_code == requests.codes.ok:
return answer
else:
return None
def GetMachineIdentifier(url):
"""
Returns the unique PMS machine identifier of url
Returns None if something went wrong
"""
xml = downloadutils.DownloadUtils().downloadUrl(
url + '/identity', type="GET")
try:
xml.attrib
except:
logMsg(title, 'Could not get the PMS machineIdentifier for %s'
% url, -1)
return None
machineIdentifier = xml.attrib.get('machineIdentifier')
logMsg(title, 'Found machineIdentifier %s for %s'
% (machineIdentifier, url), 1)
return machineIdentifier
def scrobble(ratingKey, state):
"""
Tells the PMS to set an item's watched state to state="watched" or
state="unwatched"
"""
args = {
'key': ratingKey,
'identifier': 'com.plexapp.plugins.library'
}
if state == "watched":
url = "{server}/:/scrobble?" + urlencode(args)
elif state == "unwatched":
url = "{server}/:/unscrobble?" + urlencode(args)
else:
return
downloadutils.DownloadUtils().downloadUrl(url, type="GET")
logMsg(title, "Toggled watched state for Plex item %s" % ratingKey, 1)