2017-02-04 19:46:51 +00:00
|
|
|
|
# -*- coding: utf-8 -*-
|
|
|
|
|
import re
|
2017-02-06 04:52:10 +00:00
|
|
|
|
from plexapi import log, utils
|
2017-02-04 19:46:51 +00:00
|
|
|
|
from plexapi.compat import urlencode
|
2017-02-09 06:54:38 +00:00
|
|
|
|
from plexapi.exceptions import BadRequest, NotFound
|
|
|
|
|
from plexapi.exceptions import UnknownType, Unsupported
|
|
|
|
|
|
|
|
|
|
OPERATORS = {
|
|
|
|
|
'exact': lambda v,q: v == q,
|
|
|
|
|
'iexact': lambda v,q: v.lower() == q.lower(),
|
|
|
|
|
'contains': lambda v,q: q in v,
|
|
|
|
|
'icontains': lambda v,q: q.lower() in v.lower(),
|
|
|
|
|
'in': lambda v,q: v in q,
|
|
|
|
|
'gt': lambda v,q: v > q,
|
|
|
|
|
'gte': lambda v,q: v >= q,
|
|
|
|
|
'lt': lambda v,q: v < q,
|
|
|
|
|
'lte': lambda v,q: v <= q,
|
|
|
|
|
'startswith': lambda v,q: v.startswith(q),
|
|
|
|
|
'istartswith': lambda v,q: v.lower().startswith(q),
|
|
|
|
|
'endswith': lambda v,q: v.endswith(q),
|
|
|
|
|
'iendswith': lambda v,q: v.lower().endswith(q),
|
|
|
|
|
'ismissing': None, # special case in _checkAttrs
|
|
|
|
|
'regex': lambda v,q: re.match(q, v),
|
2017-02-09 06:59:14 +00:00
|
|
|
|
'iregex': lambda v,q: re.match(q, v, flags=re.IGNORECASE),
|
2017-02-09 06:54:38 +00:00
|
|
|
|
}
|
2017-02-04 19:46:51 +00:00
|
|
|
|
|
|
|
|
|
|
2017-02-07 06:20:49 +00:00
|
|
|
|
class PlexObject(object):
|
|
|
|
|
""" Base class for all Plex objects.
|
|
|
|
|
TODO: Finish documenting this.
|
|
|
|
|
"""
|
|
|
|
|
key = None
|
|
|
|
|
|
|
|
|
|
def __init__(self, root, data, initpath=None):
|
2017-02-09 04:13:54 +00:00
|
|
|
|
self._server = root # Root MyPlexAccount or PlexServer
|
2017-02-07 06:20:49 +00:00
|
|
|
|
self._data = data # XML data needed to build object
|
|
|
|
|
self._initpath = initpath or self.key # Request path used to fetch data
|
|
|
|
|
self._loadData(data)
|
|
|
|
|
|
|
|
|
|
def __repr__(self):
|
|
|
|
|
return '<%s>' % ':'.join([p for p in [
|
|
|
|
|
self.__class__.__name__,
|
2017-02-07 06:58:29 +00:00
|
|
|
|
self.__firstattr('_baseurl', 'key', 'id', 'playQueueID', 'uri'),
|
2017-02-08 07:00:43 +00:00
|
|
|
|
self.__firstattr('title', 'name', 'username', 'product', 'tag')
|
2017-02-07 06:20:49 +00:00
|
|
|
|
] if p])
|
|
|
|
|
|
|
|
|
|
def __setattr__(self, attr, value):
|
2017-02-08 05:36:22 +00:00
|
|
|
|
if value is not None or attr.startswith('_') or attr not in self.__dict__:
|
2017-02-07 06:20:49 +00:00
|
|
|
|
self.__dict__[attr] = value
|
|
|
|
|
|
|
|
|
|
def __firstattr(self, *attrs):
|
|
|
|
|
for attr in attrs:
|
2017-02-08 07:00:43 +00:00
|
|
|
|
value = self.__dict__.get(attr)
|
|
|
|
|
if value:
|
|
|
|
|
value = str(value).replace(' ','-')
|
|
|
|
|
value = value.replace('/library/metadata/','')
|
|
|
|
|
value = value.replace('/children','')
|
|
|
|
|
return value[:20]
|
2017-02-07 06:20:49 +00:00
|
|
|
|
|
2017-02-07 06:58:29 +00:00
|
|
|
|
def _buildItem(self, elem, cls=None, initpath=None, bytag=False):
|
2017-02-07 06:20:49 +00:00
|
|
|
|
""" Factory function to build objects based on registered LIBRARY_TYPES. """
|
2017-02-07 06:58:29 +00:00
|
|
|
|
initpath = initpath or self._initpath
|
2017-02-07 06:20:49 +00:00
|
|
|
|
libtype = elem.tag if bytag else elem.attrib.get('type')
|
|
|
|
|
if libtype == 'photo' and elem.tag == 'Directory':
|
|
|
|
|
libtype = 'photoalbum'
|
2017-02-08 05:36:22 +00:00
|
|
|
|
if cls and libtype == cls.TYPE:
|
2017-02-09 04:13:54 +00:00
|
|
|
|
return cls(self._server, elem, initpath)
|
2017-02-07 06:20:49 +00:00
|
|
|
|
if libtype in utils.LIBRARY_TYPES:
|
2017-02-07 06:58:29 +00:00
|
|
|
|
cls = utils.LIBRARY_TYPES[libtype]
|
2017-02-09 04:13:54 +00:00
|
|
|
|
return cls(self._server, elem, initpath)
|
2017-02-07 06:20:49 +00:00
|
|
|
|
raise UnknownType("Unknown library type <%s type='%s'../>" % (elem.tag, libtype))
|
|
|
|
|
|
2017-02-07 06:58:29 +00:00
|
|
|
|
def _buildItemOrNone(self, elem, cls=None, initpath=None, bytag=False):
|
2017-02-07 06:20:49 +00:00
|
|
|
|
""" Calls :func:`~plexapi.base.PlexObject._buildItem()` but returns
|
|
|
|
|
None if elem is an unknown type.
|
|
|
|
|
"""
|
|
|
|
|
try:
|
2017-02-07 06:58:29 +00:00
|
|
|
|
return self._buildItem(elem, cls, initpath, bytag)
|
2017-02-07 06:20:49 +00:00
|
|
|
|
except UnknownType:
|
|
|
|
|
return None
|
|
|
|
|
|
2017-02-07 06:58:29 +00:00
|
|
|
|
def _buildItems(self, data, cls=None, initpath=None, bytag=False):
|
2017-02-07 06:20:49 +00:00
|
|
|
|
""" Build and return a list of items (optionally filtered by tag).
|
|
|
|
|
|
|
|
|
|
Parameters:
|
|
|
|
|
data (ElementTree): XML data to search for items.
|
|
|
|
|
cls (:class:`plexapi.base.PlexObject`): Optionally specify the PlexObject
|
|
|
|
|
to be built. If not specified _buildItem will be called and the best
|
|
|
|
|
guess item will be built.
|
|
|
|
|
"""
|
|
|
|
|
items = []
|
|
|
|
|
for elem in data:
|
2017-02-07 06:58:29 +00:00
|
|
|
|
items.append(self._buildItemOrNone(elem, cls, initpath, bytag))
|
2017-02-07 06:20:49 +00:00
|
|
|
|
return [item for item in items if item]
|
|
|
|
|
|
2017-02-09 04:08:25 +00:00
|
|
|
|
def fetchItem(self, key, cls=None, bytag=False, tag=None, **kwargs):
|
2017-02-07 06:20:49 +00:00
|
|
|
|
""" Load the specified key to find and build the first item with the
|
|
|
|
|
specified tag and attrs. If no tag or attrs are specified then
|
|
|
|
|
the first item in the result set is returned.
|
2017-02-09 06:54:38 +00:00
|
|
|
|
|
|
|
|
|
Parameters:
|
|
|
|
|
key (str or int): Path in Plex to fetch items from. If an int is passed
|
|
|
|
|
in, the key will be translated to /library/metadata/<key>. This allows
|
|
|
|
|
fetching an item only knowing its key-id.
|
|
|
|
|
cls (:class:`~plexapi.base.PlexObject`): If you know the class of the
|
|
|
|
|
items to be fetched, passing this in will help the parser ensure
|
|
|
|
|
it only returns those items. By default we convert the xml elements
|
|
|
|
|
to the best guess PlexObjects based on the type attr or tag.
|
|
|
|
|
bytag (bool): Setting this to True tells the build-items function to guess
|
|
|
|
|
the PlexObject to build from the tag (instead of the type attr).
|
|
|
|
|
tag (str): Only fetch items with the specified tag.
|
|
|
|
|
**kwargs (dict): Optionally add attribute filters on the items to fetch. For
|
|
|
|
|
example, passing in viewCount=0 will only return matching items. Filtering
|
|
|
|
|
is done before the Python objects are built to help keep things speedy.
|
|
|
|
|
Note: Because some attribute names are already used as arguments to this
|
|
|
|
|
function, such as 'tag', you may still reference the attr tag byappending
|
|
|
|
|
an underscore. For example, passing in _tag='foobar' will return all items
|
|
|
|
|
where tag='foobar'. Also Note: Case very much matters when specifying kwargs
|
|
|
|
|
-- Optionally, operators can be specified by append it
|
|
|
|
|
to the end of the attribute name for more complex lookups. For example,
|
|
|
|
|
passing in viewCount__gte=0 will return all items where viewCount >= 0.
|
|
|
|
|
Available operations include:
|
|
|
|
|
|
|
|
|
|
* __exact: Value matches specified arg.
|
|
|
|
|
* __iexact: Case insensative value matches specified arg.
|
|
|
|
|
* __contains: Value contains specified arg.
|
|
|
|
|
* __icontains: Case insensative value contains specified arg.
|
|
|
|
|
* __in: Value is in a specified list or tuple.
|
|
|
|
|
* __gt: Value is greater than specified arg.
|
|
|
|
|
* __gte: Value is greater than or equal to specified arg.
|
|
|
|
|
* __lt: Value is less than specified arg.
|
|
|
|
|
* __lte: Value is less than or equal to specified arg.
|
|
|
|
|
* __startswith: Value starts with specified arg.
|
|
|
|
|
* __istartswith: Case insensative value starts with specified arg.
|
|
|
|
|
* __endswith: Value ends with specified arg.
|
|
|
|
|
* __iendswith: Case insensative value ends with specified arg.
|
|
|
|
|
* __ismissing (bool): Value is or is not present in the attrs.
|
|
|
|
|
* __regex: Value matches the specified regular expression.
|
|
|
|
|
* __iregex: Case insensative value matches the specified regular expression.
|
2017-02-07 06:20:49 +00:00
|
|
|
|
"""
|
2017-02-09 06:54:38 +00:00
|
|
|
|
if isinstance(key, int):
|
|
|
|
|
key = '/library/metadata/%s' % key
|
2017-02-09 04:29:17 +00:00
|
|
|
|
for elem in self._server.query(key):
|
2017-02-09 04:08:25 +00:00
|
|
|
|
if tag and elem.tag != tag or not self._checkAttrs(elem, **kwargs):
|
2017-02-07 06:20:49 +00:00
|
|
|
|
continue
|
2017-02-07 06:58:29 +00:00
|
|
|
|
return self._buildItem(elem, cls, key, bytag)
|
2017-02-09 04:08:25 +00:00
|
|
|
|
raise NotFound('Unable to find elem: tag=%s, attrs=%s' % (tag, kwargs))
|
2017-02-07 06:20:49 +00:00
|
|
|
|
|
2017-02-09 04:08:25 +00:00
|
|
|
|
def fetchItems(self, key, cls=None, bytag=False, tag=None, **kwargs):
|
2017-02-09 06:54:38 +00:00
|
|
|
|
""" Load the specified key to find and build all items with the specified tag
|
|
|
|
|
and attrs. See :func:`~plexapi.base.PlexObject.fetchItem` for more details
|
|
|
|
|
on how this is used.
|
2017-02-07 06:20:49 +00:00
|
|
|
|
"""
|
|
|
|
|
items = []
|
2017-02-09 04:29:17 +00:00
|
|
|
|
for elem in self._server.query(key):
|
2017-02-09 04:08:25 +00:00
|
|
|
|
if tag and elem.tag != tag or not self._checkAttrs(elem, **kwargs):
|
2017-02-07 06:20:49 +00:00
|
|
|
|
continue
|
2017-02-07 06:58:29 +00:00
|
|
|
|
items.append(self._buildItemOrNone(elem, cls, key, bytag))
|
2017-02-07 06:20:49 +00:00
|
|
|
|
return [item for item in items if item]
|
|
|
|
|
|
|
|
|
|
def reload(self, safe=False):
|
|
|
|
|
""" Reload the data for this object from self.key. """
|
|
|
|
|
if not self.key:
|
|
|
|
|
if safe: return None
|
|
|
|
|
raise Unsupported('Cannot reload an object not built from a URL.')
|
|
|
|
|
self._initpath = self.key
|
2017-02-09 04:29:17 +00:00
|
|
|
|
data = self._server.query(self.key)
|
2017-02-07 06:20:49 +00:00
|
|
|
|
self._loadData(data[0])
|
2017-02-08 07:00:43 +00:00
|
|
|
|
return self
|
2017-02-07 06:20:49 +00:00
|
|
|
|
|
2017-02-09 06:54:38 +00:00
|
|
|
|
def _checkAttrs(self, elem, **kwargs):
|
2017-02-09 21:29:23 +00:00
|
|
|
|
attrsFound = {}
|
|
|
|
|
for attr, query in kwargs.items():
|
|
|
|
|
attr, op, operator = self._getAttrOperator(attr)
|
|
|
|
|
values = self._getAttrValue(elem, attr)
|
|
|
|
|
# special case ismissing operator
|
2017-02-09 06:54:38 +00:00
|
|
|
|
if op == 'ismissing':
|
|
|
|
|
if query not in (True, False):
|
|
|
|
|
raise BadRequest('Value when using __ismissing must be in (True, False).')
|
2017-02-09 21:29:23 +00:00
|
|
|
|
if (query is True and values) or (query is False and not values):
|
2017-02-09 06:54:38 +00:00
|
|
|
|
return False
|
2017-02-09 21:29:23 +00:00
|
|
|
|
# special case query in (None,0,'') to include missing attr
|
|
|
|
|
if op == 'exact' and query in (None, 0, '') and not values:
|
2017-02-09 06:54:38 +00:00
|
|
|
|
return True
|
|
|
|
|
# return if attr were looking for is missing
|
2017-02-09 21:29:23 +00:00
|
|
|
|
attrsFound[attr] = False
|
|
|
|
|
for value in values:
|
|
|
|
|
if isinstance(query, int): value = int(value)
|
|
|
|
|
if isinstance(query, float): value = float(value)
|
|
|
|
|
if isinstance(query, bool): value = bool(int(value))
|
|
|
|
|
if operator(value, query):
|
|
|
|
|
attrsFound[attr] = True
|
|
|
|
|
break
|
|
|
|
|
return all(attrsFound.values())
|
|
|
|
|
|
|
|
|
|
def _getAttrOperator(self, attr):
|
|
|
|
|
attr = attr.lstrip('_')
|
|
|
|
|
for op, operator in OPERATORS.items():
|
|
|
|
|
if attr.endswith('__%s' % op):
|
|
|
|
|
attr = attr.rsplit('__', 1)[0]
|
|
|
|
|
return attr, op, operator
|
|
|
|
|
# default to exact match
|
|
|
|
|
return attr, 'exact', OPERATORS['exact']
|
|
|
|
|
|
|
|
|
|
def _getAttrValue(self, elem, attrstr, results=None):
|
|
|
|
|
#log.debug('Fetching %s in %s', attrstr, elem.tag)
|
|
|
|
|
parts = attrstr.split('__', 1)
|
|
|
|
|
attr = parts[0]
|
|
|
|
|
attrstr = parts[1] if len(parts) == 2 else None
|
|
|
|
|
if attrstr:
|
|
|
|
|
results = [] if results is None else results
|
|
|
|
|
for child in [c for c in elem if c.tag.lower() == attr.lower()]:
|
|
|
|
|
results += self._getAttrValue(child, attrstr, results)
|
|
|
|
|
return [r for r in results if r is not None]
|
|
|
|
|
# loop through attrs so we can perform case-insensative match
|
|
|
|
|
for _attr, value in elem.attrib.items():
|
|
|
|
|
if attr.lower() == _attr.lower():
|
|
|
|
|
return [value]
|
|
|
|
|
return []
|
2017-02-09 06:54:38 +00:00
|
|
|
|
|
|
|
|
|
def _loadData(self, data):
|
2017-02-09 06:59:14 +00:00
|
|
|
|
raise NotImplementedError('Abstract method not implemented.')
|
2017-02-09 06:54:38 +00:00
|
|
|
|
|
2017-02-07 06:20:49 +00:00
|
|
|
|
|
|
|
|
|
class PlexPartialObject(PlexObject):
|
|
|
|
|
""" Not all objects in the Plex listings return the complete list of elements
|
|
|
|
|
for the object. This object will allow you to assume each object is complete,
|
|
|
|
|
and if the specified value you request is None it will fetch the full object
|
|
|
|
|
automatically and update itself.
|
|
|
|
|
|
|
|
|
|
Attributes:
|
|
|
|
|
data (ElementTree): Response from PlexServer used to build this object (optional).
|
|
|
|
|
initpath (str): Relative path requested when retrieving specified `data` (optional).
|
|
|
|
|
server (:class:`~plexapi.server.PlexServer`): PlexServer object this is from.
|
|
|
|
|
"""
|
|
|
|
|
def __eq__(self, other):
|
|
|
|
|
return other is not None and self.key == other.key
|
|
|
|
|
|
|
|
|
|
def __getattribute__(self, attr):
|
2017-02-08 05:36:22 +00:00
|
|
|
|
# Dragons inside.. :-/
|
|
|
|
|
value = utils.getattributeOrNone(PlexPartialObject, self, attr)
|
2017-02-07 06:20:49 +00:00
|
|
|
|
# Check a few cases where we dont want to reload
|
|
|
|
|
if attr == 'key' or attr.startswith('_'): return value
|
|
|
|
|
if value not in (None, []): return value
|
|
|
|
|
if self.isFullObject(): return value
|
|
|
|
|
# Log warning that were reloading the object
|
|
|
|
|
clsname = self.__class__.__name__
|
|
|
|
|
title = self.__dict__.get('title', self.__dict__.get('name'))
|
|
|
|
|
objname = "%s '%s'" % (clsname, title) if title else clsname
|
|
|
|
|
log.warn("Reloading %s for attr '%s'" % (objname, attr))
|
|
|
|
|
# Reload and return the value
|
|
|
|
|
self.reload()
|
2017-02-08 05:36:22 +00:00
|
|
|
|
return utils.getattributeOrNone(PlexPartialObject, self, attr)
|
2017-02-07 06:20:49 +00:00
|
|
|
|
|
2017-02-09 20:01:23 +00:00
|
|
|
|
def analyze(self):
|
|
|
|
|
""" Tell Plex Media Server to performs analysis on it this item to gather
|
|
|
|
|
information. Analysis includes:
|
|
|
|
|
|
|
|
|
|
* Gather Media Properties: All of the media you add to a Library has
|
|
|
|
|
properties that are useful to know–whether it's a video file, a
|
|
|
|
|
music track, or one of your photos (container, codec, resolution, etc).
|
|
|
|
|
* Generate Default Artwork: Artwork will automatically be grabbed from a
|
|
|
|
|
video file. A background image will be pulled out as well as a
|
|
|
|
|
smaller image to be used for poster/thumbnail type purposes.
|
|
|
|
|
* Generate Video Preview Thumbnails: Video preview thumbnails are created,
|
|
|
|
|
if you have that feature enabled. Video preview thumbnails allow
|
|
|
|
|
graphical seeking in some Apps. It's also used in the Plex Web App Now
|
|
|
|
|
Playing screen to show a graphical representation of where playback
|
|
|
|
|
is. Video preview thumbnails creation is a CPU-intensive process akin
|
|
|
|
|
to transcoding the file.
|
|
|
|
|
"""
|
|
|
|
|
key = '/%s/analyze' % self.key.lstrip('/')
|
|
|
|
|
self._server.query(key, method=self._server._session.put)
|
|
|
|
|
|
2017-02-07 06:20:49 +00:00
|
|
|
|
def isFullObject(self):
|
|
|
|
|
""" Retruns True if this is already a full object. A full object means all attributes
|
|
|
|
|
were populated from the api path representing only this item. For example, the
|
|
|
|
|
search result for a movie often only contain a portion of the attributes a full
|
|
|
|
|
object (main url) for that movie contain.
|
|
|
|
|
"""
|
|
|
|
|
return not self.key or self.key == self._initpath
|
|
|
|
|
|
|
|
|
|
def isPartialObject(self):
|
|
|
|
|
""" Returns True if this is not a full object. """
|
|
|
|
|
return not self.isFullObject()
|
|
|
|
|
|
2017-02-09 20:01:23 +00:00
|
|
|
|
def refresh(self):
|
|
|
|
|
""" Refreshing a Library or individual item causes the metadata for the item to be
|
|
|
|
|
refreshed, even if it already has metadata. You can think of refreshing as
|
|
|
|
|
"update metadata for the requested item even if it already has some". You should
|
|
|
|
|
refresh a Library or individual item if:
|
|
|
|
|
|
|
|
|
|
* You've changed the Library Metadata Agent.
|
|
|
|
|
* You've added "Local Media Assets" (such as artwork, theme music, external
|
|
|
|
|
subtitle files, etc.)
|
|
|
|
|
* You want to freshen the item posters, summary, etc.
|
|
|
|
|
* There's a problem with the poster image that's been downloaded.
|
|
|
|
|
* Items are missing posters or other downloaded information. This is possible if
|
|
|
|
|
the refresh process is interrupted (the Server is turned off, internet
|
|
|
|
|
connection dies, etc).
|
|
|
|
|
"""
|
|
|
|
|
key = '%s/refresh' % self.key
|
|
|
|
|
self._server.query(key, method=self._server._session.put)
|
|
|
|
|
|
|
|
|
|
def section(self):
|
|
|
|
|
""" Returns the :class:`~plexapi.library.LibrarySection` this item belongs to. """
|
|
|
|
|
return self._server.library.sectionByID(self.librarySectionID)
|
|
|
|
|
|
2017-02-07 06:20:49 +00:00
|
|
|
|
|
2017-02-04 19:46:51 +00:00
|
|
|
|
class Playable(object):
|
|
|
|
|
""" This is a general place to store functions specific to media that is Playable.
|
|
|
|
|
Things were getting mixed up a bit when dealing with Shows, Season, Artists,
|
|
|
|
|
Albums which are all not playable.
|
|
|
|
|
|
|
|
|
|
Attributes:
|
|
|
|
|
player (:class:`~plexapi.client.PlexClient`): Client object playing this item (for active sessions).
|
|
|
|
|
playlistItemID (int): Playlist item ID (only populated for :class:`~plexapi.playlist.Playlist` items).
|
|
|
|
|
sessionKey (int): Active session key.
|
|
|
|
|
transcodeSession (:class:`~plexapi.media.TranscodeSession`): Transcode Session object
|
|
|
|
|
if item is being transcoded (None otherwise).
|
|
|
|
|
username (str): Username of the person playing this item (for active sessions).
|
|
|
|
|
viewedAt (datetime): Datetime item was last viewed (history).
|
|
|
|
|
"""
|
|
|
|
|
def _loadData(self, data):
|
|
|
|
|
# Load data for active sessions (/status/sessions)
|
|
|
|
|
self.sessionKey = utils.cast(int, data.attrib.get('sessionKey'))
|
|
|
|
|
self.username = utils.findUsername(data)
|
2017-02-09 04:13:54 +00:00
|
|
|
|
self.player = utils.findPlayer(self._server, data)
|
|
|
|
|
self.transcodeSession = utils.findTranscodeSession(self._server, data)
|
2017-02-04 19:46:51 +00:00
|
|
|
|
# Load data for history details (/status/sessions/history/all)
|
|
|
|
|
self.viewedAt = utils.toDatetime(data.attrib.get('viewedAt'))
|
|
|
|
|
# Load data for playlist items
|
|
|
|
|
self.playlistItemID = utils.cast(int, data.attrib.get('playlistItemID'))
|
|
|
|
|
|
|
|
|
|
def getStreamURL(self, **params):
|
|
|
|
|
""" Returns a stream url that may be used by external applications such as VLC.
|
|
|
|
|
|
|
|
|
|
Parameters:
|
|
|
|
|
**params (dict): optional parameters to manipulate the playback when accessing
|
|
|
|
|
the stream. A few known parameters include: maxVideoBitrate, videoResolution
|
|
|
|
|
offset, copyts, protocol, mediaIndex, platform.
|
|
|
|
|
|
|
|
|
|
Raises:
|
|
|
|
|
Unsupported: When the item doesn't support fetching a stream URL.
|
|
|
|
|
"""
|
|
|
|
|
if self.TYPE not in ('movie', 'episode', 'track'):
|
|
|
|
|
raise Unsupported('Fetching stream URL for %s is unsupported.' % self.TYPE)
|
|
|
|
|
mvb = params.get('maxVideoBitrate')
|
|
|
|
|
vr = params.get('videoResolution', '')
|
|
|
|
|
params = {
|
|
|
|
|
'path': self.key,
|
|
|
|
|
'offset': params.get('offset', 0),
|
|
|
|
|
'copyts': params.get('copyts', 1),
|
|
|
|
|
'protocol': params.get('protocol'),
|
|
|
|
|
'mediaIndex': params.get('mediaIndex', 0),
|
|
|
|
|
'X-Plex-Platform': params.get('platform', 'Chrome'),
|
|
|
|
|
'maxVideoBitrate': max(mvb, 64) if mvb else None,
|
|
|
|
|
'videoResolution': vr if re.match('^\d+x\d+$', vr) else None
|
|
|
|
|
}
|
|
|
|
|
# remove None values
|
|
|
|
|
params = {k: v for k, v in params.items() if v is not None}
|
|
|
|
|
streamtype = 'audio' if self.TYPE in ('track', 'album') else 'video'
|
|
|
|
|
# sort the keys since the randomness fucks with my tests..
|
|
|
|
|
sorted_params = sorted(params.items(), key=lambda val: val[0])
|
2017-02-09 04:29:17 +00:00
|
|
|
|
return self._server.url('/%s/:/transcode/universal/start.m3u8?%s' %
|
2017-02-04 19:46:51 +00:00
|
|
|
|
(streamtype, urlencode(sorted_params)))
|
|
|
|
|
|
|
|
|
|
def iterParts(self):
|
|
|
|
|
""" Iterates over the parts of this media item. """
|
|
|
|
|
for item in self.media:
|
|
|
|
|
for part in item.parts:
|
|
|
|
|
yield part
|
|
|
|
|
|
|
|
|
|
def play(self, client):
|
|
|
|
|
""" Start playback on the specified client.
|
|
|
|
|
|
|
|
|
|
Parameters:
|
|
|
|
|
client (:class:`~plexapi.client.PlexClient`): Client to start playing on.
|
|
|
|
|
"""
|
|
|
|
|
client.playMedia(self)
|
|
|
|
|
|
|
|
|
|
def download(self, savepath=None, keep_orginal_name=False, **kwargs):
|
|
|
|
|
""" Downloads this items media to the specified location. Returns a list of
|
|
|
|
|
filepaths that have been saved to disk.
|
|
|
|
|
|
|
|
|
|
Parameters:
|
|
|
|
|
savepath (str): Title of the track to return.
|
|
|
|
|
keep_orginal_name (bool): Set True to keep the original filename as stored in
|
|
|
|
|
the Plex server. False will create a new filename with the format
|
|
|
|
|
"<Atrist> - <Album> <Track>".
|
|
|
|
|
kwargs (dict): If specified, a :func:`~plexapi.audio.Track.getStreamURL()` will
|
|
|
|
|
be returned and the additional arguments passed in will be sent to that
|
|
|
|
|
function. If kwargs is not specified, the media items will be downloaded
|
|
|
|
|
and saved to disk.
|
|
|
|
|
"""
|
|
|
|
|
filepaths = []
|
|
|
|
|
locations = [i for i in self.iterParts() if i]
|
|
|
|
|
for location in locations:
|
|
|
|
|
filename = location.file
|
|
|
|
|
if keep_orginal_name is False:
|
|
|
|
|
filename = '%s.%s' % (self._prettyfilename(), location.container)
|
|
|
|
|
# So this seems to be a alot slower but allows transcode.
|
|
|
|
|
if kwargs:
|
|
|
|
|
download_url = self.getStreamURL(**kwargs)
|
|
|
|
|
else:
|
2017-02-09 04:29:17 +00:00
|
|
|
|
download_url = self._server.url('%s?download=1' % location.key)
|
2017-02-04 19:46:51 +00:00
|
|
|
|
filepath = utils.download(download_url, filename=filename,
|
2017-02-09 04:13:54 +00:00
|
|
|
|
savepath=savepath, session=self._server._session)
|
2017-02-04 19:46:51 +00:00
|
|
|
|
if filepath:
|
|
|
|
|
filepaths.append(filepath)
|
|
|
|
|
return filepaths
|