mirror of
https://github.com/moparisthebest/SickRage
synced 2024-11-11 11:55:03 -05:00
2dcd26e69c
Fixed invalid indexer id issues for TVRage shows. Fixed issues for getting posters and backdrops for TVRage shows. We now convert XML straight to a dict object for Indexer APIs, improved overall performance api's Fixed issues with TVRage shows and displaying genre's properly.
981 lines
41 KiB
Python
981 lines
41 KiB
Python
"""
|
|
imdb package.
|
|
|
|
This package can be used to retrieve information about a movie or
|
|
a person from the IMDb database.
|
|
It can fetch data through different media (e.g.: the IMDb web pages,
|
|
a SQL database, etc.)
|
|
|
|
Copyright 2004-2014 Davide Alberani <da@erlug.linux.it>
|
|
|
|
This program is free software; you can redistribute it and/or modify
|
|
it under the terms of the GNU General Public License as published by
|
|
the Free Software Foundation; either version 2 of the License, or
|
|
(at your option) any later version.
|
|
|
|
This program is distributed in the hope that it will be useful,
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with this program; if not, write to the Free Software
|
|
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
|
"""
|
|
|
|
__all__ = ['IMDb', 'IMDbError', 'Movie', 'Person', 'Character', 'Company',
|
|
'available_access_systems']
|
|
__version__ = VERSION = '5.0'
|
|
|
|
# Import compatibility module (importing it is enough).
|
|
import _compat
|
|
|
|
import sys, os, ConfigParser, logging
|
|
from types import MethodType
|
|
|
|
from imdb import Movie, Person, Character, Company
|
|
import imdb._logging
|
|
from imdb._exceptions import IMDbError, IMDbDataAccessError, IMDbParserError
|
|
from imdb.utils import build_title, build_name, build_company_name
|
|
|
|
_aux_logger = logging.getLogger('imdbpy.aux')
|
|
|
|
|
|
# URLs of the main pages for movies, persons, characters and queries.
|
|
imdbURL_base = 'http://akas.imdb.com/'
|
|
|
|
# NOTE: the urls below will be removed in a future version.
|
|
# please use the values in the 'urls' attribute
|
|
# of the IMDbBase subclass instance.
|
|
# http://akas.imdb.com/title/
|
|
imdbURL_movie_base = '%stitle/' % imdbURL_base
|
|
# http://akas.imdb.com/title/tt%s/
|
|
imdbURL_movie_main = imdbURL_movie_base + 'tt%s/'
|
|
# http://akas.imdb.com/name/
|
|
imdbURL_person_base = '%sname/' % imdbURL_base
|
|
# http://akas.imdb.com/name/nm%s/
|
|
imdbURL_person_main = imdbURL_person_base + 'nm%s/'
|
|
# http://akas.imdb.com/character/
|
|
imdbURL_character_base = '%scharacter/' % imdbURL_base
|
|
# http://akas.imdb.com/character/ch%s/
|
|
imdbURL_character_main = imdbURL_character_base + 'ch%s/'
|
|
# http://akas.imdb.com/company/
|
|
imdbURL_company_base = '%scompany/' % imdbURL_base
|
|
# http://akas.imdb.com/company/co%s/
|
|
imdbURL_company_main = imdbURL_company_base + 'co%s/'
|
|
# http://akas.imdb.com/keyword/%s/
|
|
imdbURL_keyword_main = imdbURL_base + 'keyword/%s/'
|
|
# http://akas.imdb.com/chart/top
|
|
imdbURL_top250 = imdbURL_base + 'chart/top'
|
|
# http://akas.imdb.com/chart/bottom
|
|
imdbURL_bottom100 = imdbURL_base + 'chart/bottom'
|
|
# http://akas.imdb.com/find?%s
|
|
imdbURL_find = imdbURL_base + 'find?%s'
|
|
|
|
# Name of the configuration file.
|
|
confFileName = 'imdbpy.cfg'
|
|
|
|
class ConfigParserWithCase(ConfigParser.ConfigParser):
|
|
"""A case-sensitive parser for configuration files."""
|
|
def __init__(self, defaults=None, confFile=None, *args, **kwds):
|
|
"""Initialize the parser.
|
|
|
|
*defaults* -- defaults values.
|
|
*confFile* -- the file (or list of files) to parse."""
|
|
ConfigParser.ConfigParser.__init__(self, defaults=defaults)
|
|
if confFile is None:
|
|
dotFileName = '.' + confFileName
|
|
# Current and home directory.
|
|
confFile = [os.path.join(os.getcwd(), confFileName),
|
|
os.path.join(os.getcwd(), dotFileName),
|
|
os.path.join(os.path.expanduser('~'), confFileName),
|
|
os.path.join(os.path.expanduser('~'), dotFileName)]
|
|
if os.name == 'posix':
|
|
sep = getattr(os.path, 'sep', '/')
|
|
# /etc/ and /etc/conf.d/
|
|
confFile.append(os.path.join(sep, 'etc', confFileName))
|
|
confFile.append(os.path.join(sep, 'etc', 'conf.d',
|
|
confFileName))
|
|
else:
|
|
# etc subdirectory of sys.prefix, for non-unix systems.
|
|
confFile.append(os.path.join(sys.prefix, 'etc', confFileName))
|
|
for fname in confFile:
|
|
try:
|
|
self.read(fname)
|
|
except (ConfigParser.MissingSectionHeaderError,
|
|
ConfigParser.ParsingError), e:
|
|
_aux_logger.warn('Troubles reading config file: %s' % e)
|
|
# Stop at the first valid file.
|
|
if self.has_section('imdbpy'):
|
|
break
|
|
|
|
def optionxform(self, optionstr):
|
|
"""Option names are case sensitive."""
|
|
return optionstr
|
|
|
|
def _manageValue(self, value):
|
|
"""Custom substitutions for values."""
|
|
if not isinstance(value, (str, unicode)):
|
|
return value
|
|
vlower = value.lower()
|
|
if vlower in self._boolean_states:
|
|
return self._boolean_states[vlower]
|
|
elif vlower == 'none':
|
|
return None
|
|
return value
|
|
|
|
def get(self, section, option, *args, **kwds):
|
|
"""Return the value of an option from a given section."""
|
|
value = ConfigParser.ConfigParser.get(self, section, option,
|
|
*args, **kwds)
|
|
return self._manageValue(value)
|
|
|
|
def items(self, section, *args, **kwds):
|
|
"""Return a list of (key, value) tuples of items of the
|
|
given section."""
|
|
if section != 'DEFAULT' and not self.has_section(section):
|
|
return []
|
|
keys = ConfigParser.ConfigParser.options(self, section)
|
|
return [(k, self.get(section, k, *args, **kwds)) for k in keys]
|
|
|
|
def getDict(self, section):
|
|
"""Return a dictionary of items of the specified section."""
|
|
return dict(self.items(section))
|
|
|
|
|
|
def IMDb(accessSystem=None, *arguments, **keywords):
|
|
"""Return an instance of the appropriate class.
|
|
The accessSystem parameter is used to specify the kind of
|
|
the preferred access system."""
|
|
if accessSystem is None or accessSystem in ('auto', 'config'):
|
|
try:
|
|
cfg_file = ConfigParserWithCase(*arguments, **keywords)
|
|
# Parameters set by the code take precedence.
|
|
kwds = cfg_file.getDict('imdbpy')
|
|
if 'accessSystem' in kwds:
|
|
accessSystem = kwds['accessSystem']
|
|
del kwds['accessSystem']
|
|
else:
|
|
accessSystem = 'http'
|
|
kwds.update(keywords)
|
|
keywords = kwds
|
|
except Exception, e:
|
|
import logging
|
|
logging.getLogger('imdbpy').warn('Unable to read configuration' \
|
|
' file; complete error: %s' % e)
|
|
# It just LOOKS LIKE a bad habit: we tried to read config
|
|
# options from some files, but something is gone horribly
|
|
# wrong: ignore everything and pretend we were called with
|
|
# the 'http' accessSystem.
|
|
accessSystem = 'http'
|
|
if 'loggingLevel' in keywords:
|
|
imdb._logging.setLevel(keywords['loggingLevel'])
|
|
del keywords['loggingLevel']
|
|
if 'loggingConfig' in keywords:
|
|
logCfg = keywords['loggingConfig']
|
|
del keywords['loggingConfig']
|
|
try:
|
|
import logging.config
|
|
logging.config.fileConfig(os.path.expanduser(logCfg))
|
|
except Exception, e:
|
|
logging.getLogger('imdbpy').warn('unable to read logger ' \
|
|
'config: %s' % e)
|
|
if accessSystem in ('httpThin', 'webThin', 'htmlThin'):
|
|
logging.warn('httpThin was removed since IMDbPY 4.8')
|
|
accessSystem = 'http'
|
|
if accessSystem in ('http', 'web', 'html'):
|
|
from parser.http import IMDbHTTPAccessSystem
|
|
return IMDbHTTPAccessSystem(*arguments, **keywords)
|
|
elif accessSystem in ('mobile',):
|
|
from parser.mobile import IMDbMobileAccessSystem
|
|
return IMDbMobileAccessSystem(*arguments, **keywords)
|
|
elif accessSystem in ('local', 'files'):
|
|
# The local access system was removed since IMDbPY 4.2.
|
|
raise IMDbError('the local access system was removed since IMDbPY 4.2')
|
|
elif accessSystem in ('sql', 'db', 'database'):
|
|
try:
|
|
from parser.sql import IMDbSqlAccessSystem
|
|
except ImportError:
|
|
raise IMDbError('the sql access system is not installed')
|
|
return IMDbSqlAccessSystem(*arguments, **keywords)
|
|
else:
|
|
raise IMDbError('unknown kind of data access system: "%s"' \
|
|
% accessSystem)
|
|
|
|
|
|
def available_access_systems():
|
|
"""Return the list of available data access systems."""
|
|
asList = []
|
|
# XXX: trying to import modules is a good thing?
|
|
try:
|
|
from parser.http import IMDbHTTPAccessSystem
|
|
asList.append('http')
|
|
except ImportError:
|
|
pass
|
|
try:
|
|
from parser.mobile import IMDbMobileAccessSystem
|
|
asList.append('mobile')
|
|
except ImportError:
|
|
pass
|
|
try:
|
|
from parser.sql import IMDbSqlAccessSystem
|
|
asList.append('sql')
|
|
except ImportError:
|
|
pass
|
|
return asList
|
|
|
|
|
|
# XXX: I'm not sure this is a good guess.
|
|
# I suppose that an argument of the IMDb function can be used to
|
|
# set a default encoding for the output, and then Movie, Person and
|
|
# Character objects can use this default encoding, returning strings.
|
|
# Anyway, passing unicode strings to search_movie(), search_person()
|
|
# and search_character() methods is always safer.
|
|
encoding = getattr(sys.stdin, 'encoding', '') or sys.getdefaultencoding()
|
|
|
|
class IMDbBase:
|
|
"""The base class used to search for a movie/person/character and
|
|
to get a Movie/Person/Character object.
|
|
|
|
This class cannot directly fetch data of any kind and so you
|
|
have to search the "real" code into a subclass."""
|
|
|
|
# The name of the preferred access system (MUST be overridden
|
|
# in the subclasses).
|
|
accessSystem = 'UNKNOWN'
|
|
|
|
# Top-level logger for IMDbPY.
|
|
_imdb_logger = logging.getLogger('imdbpy')
|
|
|
|
# Whether to re-raise caught exceptions or not.
|
|
_reraise_exceptions = False
|
|
|
|
def __init__(self, defaultModFunct=None, results=20, keywordsResults=100,
|
|
*arguments, **keywords):
|
|
"""Initialize the access system.
|
|
If specified, defaultModFunct is the function used by
|
|
default by the Person, Movie and Character objects, when
|
|
accessing their text fields.
|
|
"""
|
|
# The function used to output the strings that need modification (the
|
|
# ones containing references to movie titles and person names).
|
|
self._defModFunct = defaultModFunct
|
|
# Number of results to get.
|
|
try:
|
|
results = int(results)
|
|
except (TypeError, ValueError):
|
|
results = 20
|
|
if results < 1:
|
|
results = 20
|
|
self._results = results
|
|
try:
|
|
keywordsResults = int(keywordsResults)
|
|
except (TypeError, ValueError):
|
|
keywordsResults = 100
|
|
if keywordsResults < 1:
|
|
keywordsResults = 100
|
|
self._keywordsResults = keywordsResults
|
|
self._reraise_exceptions = keywords.get('reraiseExceptions') or False
|
|
self.set_imdb_urls(keywords.get('imdbURL_base') or imdbURL_base)
|
|
|
|
def set_imdb_urls(self, imdbURL_base):
|
|
"""Set the urls used accessing the IMDb site."""
|
|
imdbURL_base = imdbURL_base.strip().strip('"\'')
|
|
if not imdbURL_base.startswith('http://'):
|
|
imdbURL_base = 'http://%s' % imdbURL_base
|
|
if not imdbURL_base.endswith('/'):
|
|
imdbURL_base = '%s/' % imdbURL_base
|
|
# http://akas.imdb.com/title/
|
|
imdbURL_movie_base='%stitle/' % imdbURL_base
|
|
# http://akas.imdb.com/title/tt%s/
|
|
imdbURL_movie_main=imdbURL_movie_base + 'tt%s/'
|
|
# http://akas.imdb.com/name/
|
|
imdbURL_person_base='%sname/' % imdbURL_base
|
|
# http://akas.imdb.com/name/nm%s/
|
|
imdbURL_person_main=imdbURL_person_base + 'nm%s/'
|
|
# http://akas.imdb.com/character/
|
|
imdbURL_character_base='%scharacter/' % imdbURL_base
|
|
# http://akas.imdb.com/character/ch%s/
|
|
imdbURL_character_main=imdbURL_character_base + 'ch%s/'
|
|
# http://akas.imdb.com/company/
|
|
imdbURL_company_base='%scompany/' % imdbURL_base
|
|
# http://akas.imdb.com/company/co%s/
|
|
imdbURL_company_main=imdbURL_company_base + 'co%s/'
|
|
# http://akas.imdb.com/keyword/%s/
|
|
imdbURL_keyword_main=imdbURL_base + 'keyword/%s/'
|
|
# http://akas.imdb.com/chart/top
|
|
imdbURL_top250=imdbURL_base + 'chart/top'
|
|
# http://akas.imdb.com/chart/bottom
|
|
imdbURL_bottom100=imdbURL_base + 'chart/bottom'
|
|
# http://akas.imdb.com/find?%s
|
|
imdbURL_find=imdbURL_base + 'find?%s'
|
|
self.urls = dict(
|
|
movie_base=imdbURL_movie_base,
|
|
movie_main=imdbURL_movie_main,
|
|
person_base=imdbURL_person_base,
|
|
person_main=imdbURL_person_main,
|
|
character_base=imdbURL_character_base,
|
|
character_main=imdbURL_character_main,
|
|
company_base=imdbURL_company_base,
|
|
company_main=imdbURL_company_main,
|
|
keyword_main=imdbURL_keyword_main,
|
|
top250=imdbURL_top250,
|
|
bottom100=imdbURL_bottom100,
|
|
find=imdbURL_find)
|
|
|
|
def _normalize_movieID(self, movieID):
|
|
"""Normalize the given movieID."""
|
|
# By default, do nothing.
|
|
return movieID
|
|
|
|
def _normalize_personID(self, personID):
|
|
"""Normalize the given personID."""
|
|
# By default, do nothing.
|
|
return personID
|
|
|
|
def _normalize_characterID(self, characterID):
|
|
"""Normalize the given characterID."""
|
|
# By default, do nothing.
|
|
return characterID
|
|
|
|
def _normalize_companyID(self, companyID):
|
|
"""Normalize the given companyID."""
|
|
# By default, do nothing.
|
|
return companyID
|
|
|
|
def _get_real_movieID(self, movieID):
|
|
"""Handle title aliases."""
|
|
# By default, do nothing.
|
|
return movieID
|
|
|
|
def _get_real_personID(self, personID):
|
|
"""Handle name aliases."""
|
|
# By default, do nothing.
|
|
return personID
|
|
|
|
def _get_real_characterID(self, characterID):
|
|
"""Handle character name aliases."""
|
|
# By default, do nothing.
|
|
return characterID
|
|
|
|
def _get_real_companyID(self, companyID):
|
|
"""Handle company name aliases."""
|
|
# By default, do nothing.
|
|
return companyID
|
|
|
|
def _get_infoset(self, prefname):
|
|
"""Return methods with the name starting with prefname."""
|
|
infoset = []
|
|
excludes = ('%sinfoset' % prefname,)
|
|
preflen = len(prefname)
|
|
for name in dir(self.__class__):
|
|
if name.startswith(prefname) and name not in excludes:
|
|
member = getattr(self.__class__, name)
|
|
if isinstance(member, MethodType):
|
|
infoset.append(name[preflen:].replace('_', ' '))
|
|
return infoset
|
|
|
|
def get_movie_infoset(self):
|
|
"""Return the list of info set available for movies."""
|
|
return self._get_infoset('get_movie_')
|
|
|
|
def get_person_infoset(self):
|
|
"""Return the list of info set available for persons."""
|
|
return self._get_infoset('get_person_')
|
|
|
|
def get_character_infoset(self):
|
|
"""Return the list of info set available for characters."""
|
|
return self._get_infoset('get_character_')
|
|
|
|
def get_company_infoset(self):
|
|
"""Return the list of info set available for companies."""
|
|
return self._get_infoset('get_company_')
|
|
|
|
def get_movie(self, movieID, info=Movie.Movie.default_info, modFunct=None):
|
|
"""Return a Movie object for the given movieID.
|
|
|
|
The movieID is something used to univocally identify a movie;
|
|
it can be the imdbID used by the IMDb web server, a file
|
|
pointer, a line number in a file, an ID in a database, etc.
|
|
|
|
info is the list of sets of information to retrieve.
|
|
|
|
If specified, modFunct will be the function used by the Movie
|
|
object when accessing its text fields (like 'plot')."""
|
|
movieID = self._normalize_movieID(movieID)
|
|
movieID = self._get_real_movieID(movieID)
|
|
movie = Movie.Movie(movieID=movieID, accessSystem=self.accessSystem)
|
|
modFunct = modFunct or self._defModFunct
|
|
if modFunct is not None:
|
|
movie.set_mod_funct(modFunct)
|
|
self.update(movie, info)
|
|
return movie
|
|
|
|
get_episode = get_movie
|
|
|
|
def _search_movie(self, title, results):
|
|
"""Return a list of tuples (movieID, {movieData})"""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def search_movie(self, title, results=None, _episodes=False):
|
|
"""Return a list of Movie objects for a query for the given title.
|
|
The results argument is the maximum number of results to return."""
|
|
if results is None:
|
|
results = self._results
|
|
try:
|
|
results = int(results)
|
|
except (ValueError, OverflowError):
|
|
results = 20
|
|
# XXX: I suppose it will be much safer if the user provides
|
|
# an unicode string... this is just a guess.
|
|
if not isinstance(title, unicode):
|
|
title = unicode(title, encoding, 'replace')
|
|
if not _episodes:
|
|
res = self._search_movie(title, results)
|
|
else:
|
|
res = self._search_episode(title, results)
|
|
return [Movie.Movie(movieID=self._get_real_movieID(mi),
|
|
data=md, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for mi, md in res][:results]
|
|
|
|
def _search_episode(self, title, results):
|
|
"""Return a list of tuples (movieID, {movieData})"""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def search_episode(self, title, results=None):
|
|
"""Return a list of Movie objects for a query for the given title.
|
|
The results argument is the maximum number of results to return;
|
|
this method searches only for titles of tv (mini) series' episodes."""
|
|
return self.search_movie(title, results=results, _episodes=True)
|
|
|
|
def get_person(self, personID, info=Person.Person.default_info,
|
|
modFunct=None):
|
|
"""Return a Person object for the given personID.
|
|
|
|
The personID is something used to univocally identify a person;
|
|
it can be the imdbID used by the IMDb web server, a file
|
|
pointer, a line number in a file, an ID in a database, etc.
|
|
|
|
info is the list of sets of information to retrieve.
|
|
|
|
If specified, modFunct will be the function used by the Person
|
|
object when accessing its text fields (like 'mini biography')."""
|
|
personID = self._normalize_personID(personID)
|
|
personID = self._get_real_personID(personID)
|
|
person = Person.Person(personID=personID,
|
|
accessSystem=self.accessSystem)
|
|
modFunct = modFunct or self._defModFunct
|
|
if modFunct is not None:
|
|
person.set_mod_funct(modFunct)
|
|
self.update(person, info)
|
|
return person
|
|
|
|
def _search_person(self, name, results):
|
|
"""Return a list of tuples (personID, {personData})"""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def search_person(self, name, results=None):
|
|
"""Return a list of Person objects for a query for the given name.
|
|
|
|
The results argument is the maximum number of results to return."""
|
|
if results is None:
|
|
results = self._results
|
|
try:
|
|
results = int(results)
|
|
except (ValueError, OverflowError):
|
|
results = 20
|
|
if not isinstance(name, unicode):
|
|
name = unicode(name, encoding, 'replace')
|
|
res = self._search_person(name, results)
|
|
return [Person.Person(personID=self._get_real_personID(pi),
|
|
data=pd, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for pi, pd in res][:results]
|
|
|
|
def get_character(self, characterID, info=Character.Character.default_info,
|
|
modFunct=None):
|
|
"""Return a Character object for the given characterID.
|
|
|
|
The characterID is something used to univocally identify a character;
|
|
it can be the imdbID used by the IMDb web server, a file
|
|
pointer, a line number in a file, an ID in a database, etc.
|
|
|
|
info is the list of sets of information to retrieve.
|
|
|
|
If specified, modFunct will be the function used by the Character
|
|
object when accessing its text fields (like 'biography')."""
|
|
characterID = self._normalize_characterID(characterID)
|
|
characterID = self._get_real_characterID(characterID)
|
|
character = Character.Character(characterID=characterID,
|
|
accessSystem=self.accessSystem)
|
|
modFunct = modFunct or self._defModFunct
|
|
if modFunct is not None:
|
|
character.set_mod_funct(modFunct)
|
|
self.update(character, info)
|
|
return character
|
|
|
|
def _search_character(self, name, results):
|
|
"""Return a list of tuples (characterID, {characterData})"""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def search_character(self, name, results=None):
|
|
"""Return a list of Character objects for a query for the given name.
|
|
|
|
The results argument is the maximum number of results to return."""
|
|
if results is None:
|
|
results = self._results
|
|
try:
|
|
results = int(results)
|
|
except (ValueError, OverflowError):
|
|
results = 20
|
|
if not isinstance(name, unicode):
|
|
name = unicode(name, encoding, 'replace')
|
|
res = self._search_character(name, results)
|
|
return [Character.Character(characterID=self._get_real_characterID(pi),
|
|
data=pd, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for pi, pd in res][:results]
|
|
|
|
def get_company(self, companyID, info=Company.Company.default_info,
|
|
modFunct=None):
|
|
"""Return a Company object for the given companyID.
|
|
|
|
The companyID is something used to univocally identify a company;
|
|
it can be the imdbID used by the IMDb web server, a file
|
|
pointer, a line number in a file, an ID in a database, etc.
|
|
|
|
info is the list of sets of information to retrieve.
|
|
|
|
If specified, modFunct will be the function used by the Company
|
|
object when accessing its text fields (none, so far)."""
|
|
companyID = self._normalize_companyID(companyID)
|
|
companyID = self._get_real_companyID(companyID)
|
|
company = Company.Company(companyID=companyID,
|
|
accessSystem=self.accessSystem)
|
|
modFunct = modFunct or self._defModFunct
|
|
if modFunct is not None:
|
|
company.set_mod_funct(modFunct)
|
|
self.update(company, info)
|
|
return company
|
|
|
|
def _search_company(self, name, results):
|
|
"""Return a list of tuples (companyID, {companyData})"""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def search_company(self, name, results=None):
|
|
"""Return a list of Company objects for a query for the given name.
|
|
|
|
The results argument is the maximum number of results to return."""
|
|
if results is None:
|
|
results = self._results
|
|
try:
|
|
results = int(results)
|
|
except (ValueError, OverflowError):
|
|
results = 20
|
|
if not isinstance(name, unicode):
|
|
name = unicode(name, encoding, 'replace')
|
|
res = self._search_company(name, results)
|
|
return [Company.Company(companyID=self._get_real_companyID(pi),
|
|
data=pd, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for pi, pd in res][:results]
|
|
|
|
def _search_keyword(self, keyword, results):
|
|
"""Return a list of 'keyword' strings."""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def search_keyword(self, keyword, results=None):
|
|
"""Search for existing keywords, similar to the given one."""
|
|
if results is None:
|
|
results = self._keywordsResults
|
|
try:
|
|
results = int(results)
|
|
except (ValueError, OverflowError):
|
|
results = 100
|
|
if not isinstance(keyword, unicode):
|
|
keyword = unicode(keyword, encoding, 'replace')
|
|
return self._search_keyword(keyword, results)
|
|
|
|
def _get_keyword(self, keyword, results):
|
|
"""Return a list of tuples (movieID, {movieData})"""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def get_keyword(self, keyword, results=None):
|
|
"""Return a list of movies for the given keyword."""
|
|
if results is None:
|
|
results = self._keywordsResults
|
|
try:
|
|
results = int(results)
|
|
except (ValueError, OverflowError):
|
|
results = 100
|
|
# XXX: I suppose it will be much safer if the user provides
|
|
# an unicode string... this is just a guess.
|
|
if not isinstance(keyword, unicode):
|
|
keyword = unicode(keyword, encoding, 'replace')
|
|
res = self._get_keyword(keyword, results)
|
|
return [Movie.Movie(movieID=self._get_real_movieID(mi),
|
|
data=md, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for mi, md in res][:results]
|
|
|
|
def _get_top_bottom_movies(self, kind):
|
|
"""Return the list of the top 250 or bottom 100 movies."""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
# This method must return a list of (movieID, {movieDict})
|
|
# tuples. The kind parameter can be 'top' or 'bottom'.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def get_top250_movies(self):
|
|
"""Return the list of the top 250 movies."""
|
|
res = self._get_top_bottom_movies('top')
|
|
return [Movie.Movie(movieID=self._get_real_movieID(mi),
|
|
data=md, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for mi, md in res]
|
|
|
|
def get_bottom100_movies(self):
|
|
"""Return the list of the bottom 100 movies."""
|
|
res = self._get_top_bottom_movies('bottom')
|
|
return [Movie.Movie(movieID=self._get_real_movieID(mi),
|
|
data=md, modFunct=self._defModFunct,
|
|
accessSystem=self.accessSystem) for mi, md in res]
|
|
|
|
def new_movie(self, *arguments, **keywords):
|
|
"""Return a Movie object."""
|
|
# XXX: not really useful...
|
|
if 'title' in keywords:
|
|
if not isinstance(keywords['title'], unicode):
|
|
keywords['title'] = unicode(keywords['title'],
|
|
encoding, 'replace')
|
|
elif len(arguments) > 1:
|
|
if not isinstance(arguments[1], unicode):
|
|
arguments[1] = unicode(arguments[1], encoding, 'replace')
|
|
return Movie.Movie(accessSystem=self.accessSystem,
|
|
*arguments, **keywords)
|
|
|
|
def new_person(self, *arguments, **keywords):
|
|
"""Return a Person object."""
|
|
# XXX: not really useful...
|
|
if 'name' in keywords:
|
|
if not isinstance(keywords['name'], unicode):
|
|
keywords['name'] = unicode(keywords['name'],
|
|
encoding, 'replace')
|
|
elif len(arguments) > 1:
|
|
if not isinstance(arguments[1], unicode):
|
|
arguments[1] = unicode(arguments[1], encoding, 'replace')
|
|
return Person.Person(accessSystem=self.accessSystem,
|
|
*arguments, **keywords)
|
|
|
|
def new_character(self, *arguments, **keywords):
|
|
"""Return a Character object."""
|
|
# XXX: not really useful...
|
|
if 'name' in keywords:
|
|
if not isinstance(keywords['name'], unicode):
|
|
keywords['name'] = unicode(keywords['name'],
|
|
encoding, 'replace')
|
|
elif len(arguments) > 1:
|
|
if not isinstance(arguments[1], unicode):
|
|
arguments[1] = unicode(arguments[1], encoding, 'replace')
|
|
return Character.Character(accessSystem=self.accessSystem,
|
|
*arguments, **keywords)
|
|
|
|
def new_company(self, *arguments, **keywords):
|
|
"""Return a Company object."""
|
|
# XXX: not really useful...
|
|
if 'name' in keywords:
|
|
if not isinstance(keywords['name'], unicode):
|
|
keywords['name'] = unicode(keywords['name'],
|
|
encoding, 'replace')
|
|
elif len(arguments) > 1:
|
|
if not isinstance(arguments[1], unicode):
|
|
arguments[1] = unicode(arguments[1], encoding, 'replace')
|
|
return Company.Company(accessSystem=self.accessSystem,
|
|
*arguments, **keywords)
|
|
|
|
def update(self, mop, info=None, override=0):
|
|
"""Given a Movie, Person, Character or Company object with only
|
|
partial information, retrieve the required set of information.
|
|
|
|
info is the list of sets of information to retrieve.
|
|
|
|
If override is set, the information are retrieved and updated
|
|
even if they're already in the object."""
|
|
# XXX: should this be a method of the Movie/Person/Character/Company
|
|
# classes? NO! What for instances created by external functions?
|
|
mopID = None
|
|
prefix = ''
|
|
if isinstance(mop, Movie.Movie):
|
|
mopID = mop.movieID
|
|
prefix = 'movie'
|
|
elif isinstance(mop, Person.Person):
|
|
mopID = mop.personID
|
|
prefix = 'person'
|
|
elif isinstance(mop, Character.Character):
|
|
mopID = mop.characterID
|
|
prefix = 'character'
|
|
elif isinstance(mop, Company.Company):
|
|
mopID = mop.companyID
|
|
prefix = 'company'
|
|
else:
|
|
raise IMDbError('object ' + repr(mop) + \
|
|
' is not a Movie, Person, Character or Company instance')
|
|
if mopID is None:
|
|
# XXX: enough? It's obvious that there are Characters
|
|
# objects without characterID, so I think they should
|
|
# just do nothing, when an i.update(character) is tried.
|
|
if prefix == 'character':
|
|
return
|
|
raise IMDbDataAccessError( \
|
|
'the supplied object has null movieID, personID or companyID')
|
|
if mop.accessSystem == self.accessSystem:
|
|
aSystem = self
|
|
else:
|
|
aSystem = IMDb(mop.accessSystem)
|
|
if info is None:
|
|
info = mop.default_info
|
|
elif info == 'all':
|
|
if isinstance(mop, Movie.Movie):
|
|
info = self.get_movie_infoset()
|
|
elif isinstance(mop, Person.Person):
|
|
info = self.get_person_infoset()
|
|
elif isinstance(mop, Character.Character):
|
|
info = self.get_character_infoset()
|
|
else:
|
|
info = self.get_company_infoset()
|
|
if not isinstance(info, (tuple, list)):
|
|
info = (info,)
|
|
res = {}
|
|
for i in info:
|
|
if i in mop.current_info and not override:
|
|
continue
|
|
if not i:
|
|
continue
|
|
self._imdb_logger.debug('retrieving "%s" info set', i)
|
|
try:
|
|
method = getattr(aSystem, 'get_%s_%s' %
|
|
(prefix, i.replace(' ', '_')))
|
|
except AttributeError:
|
|
self._imdb_logger.error('unknown information set "%s"', i)
|
|
# Keeps going.
|
|
method = lambda *x: {}
|
|
try:
|
|
ret = method(mopID)
|
|
except Exception, e:
|
|
self._imdb_logger.critical('caught an exception retrieving ' \
|
|
'or parsing "%s" info set for mopID ' \
|
|
'"%s" (accessSystem: %s)',
|
|
i, mopID, mop.accessSystem, exc_info=True)
|
|
ret = {}
|
|
# If requested by the user, reraise the exception.
|
|
if self._reraise_exceptions:
|
|
raise
|
|
keys = None
|
|
if 'data' in ret:
|
|
res.update(ret['data'])
|
|
if isinstance(ret['data'], dict):
|
|
keys = ret['data'].keys()
|
|
if 'info sets' in ret:
|
|
for ri in ret['info sets']:
|
|
mop.add_to_current_info(ri, keys, mainInfoset=i)
|
|
else:
|
|
mop.add_to_current_info(i, keys)
|
|
if 'titlesRefs' in ret:
|
|
mop.update_titlesRefs(ret['titlesRefs'])
|
|
if 'namesRefs' in ret:
|
|
mop.update_namesRefs(ret['namesRefs'])
|
|
if 'charactersRefs' in ret:
|
|
mop.update_charactersRefs(ret['charactersRefs'])
|
|
mop.set_data(res, override=0)
|
|
|
|
def get_imdbMovieID(self, movieID):
|
|
"""Translate a movieID in an imdbID (the ID used by the IMDb
|
|
web server); must be overridden by the subclass."""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def get_imdbPersonID(self, personID):
|
|
"""Translate a personID in a imdbID (the ID used by the IMDb
|
|
web server); must be overridden by the subclass."""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def get_imdbCharacterID(self, characterID):
|
|
"""Translate a characterID in a imdbID (the ID used by the IMDb
|
|
web server); must be overridden by the subclass."""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def get_imdbCompanyID(self, companyID):
|
|
"""Translate a companyID in a imdbID (the ID used by the IMDb
|
|
web server); must be overridden by the subclass."""
|
|
# XXX: for the real implementation, see the method of the
|
|
# subclass, somewhere under the imdb.parser package.
|
|
raise NotImplementedError('override this method')
|
|
|
|
def _searchIMDb(self, kind, ton, title_kind=None):
|
|
"""Search the IMDb akas server for the given title or name."""
|
|
# The Exact Primary search system has gone AWOL, so we resort
|
|
# to the mobile search. :-/
|
|
if not ton:
|
|
return None
|
|
ton = ton.strip('"')
|
|
aSystem = IMDb('mobile')
|
|
if kind == 'tt':
|
|
searchFunct = aSystem.search_movie
|
|
check = 'long imdb title'
|
|
elif kind == 'nm':
|
|
searchFunct = aSystem.search_person
|
|
check = 'long imdb name'
|
|
elif kind == 'char':
|
|
searchFunct = aSystem.search_character
|
|
check = 'long imdb name'
|
|
elif kind == 'co':
|
|
# XXX: are [COUNTRY] codes included in the results?
|
|
searchFunct = aSystem.search_company
|
|
check = 'long imdb name'
|
|
try:
|
|
searchRes = searchFunct(ton)
|
|
except IMDbError:
|
|
return None
|
|
# When only one result is returned, assume it was from an
|
|
# exact match.
|
|
if len(searchRes) == 1:
|
|
return searchRes[0].getID()
|
|
title_only_matches = []
|
|
for item in searchRes:
|
|
# Return the first perfect match.
|
|
if item[check].strip('"') == ton:
|
|
# For titles do additional check for kind
|
|
if kind != 'tt' or title_kind == item['kind']:
|
|
return item.getID()
|
|
elif kind == 'tt':
|
|
title_only_matches.append(item.getID())
|
|
# imdbpy2sql.py could detected wrong type, so if no title and kind
|
|
# matches found - collect all results with title only match
|
|
# Return list of IDs if multiple matches (can happen when searching
|
|
# titles with no title_kind specified)
|
|
# Example: DB: Band of Brothers "tv series" vs "tv mini-series"
|
|
if title_only_matches:
|
|
if len(title_only_matches) == 1:
|
|
return title_only_matches[0]
|
|
else:
|
|
return title_only_matches
|
|
return None
|
|
|
|
def title2imdbID(self, title, kind=None):
|
|
"""Translate a movie title (in the plain text data files format)
|
|
to an imdbID.
|
|
Try an Exact Primary Title search on IMDb;
|
|
return None if it's unable to get the imdbID;
|
|
Always specify kind: movie, tv series, video game etc. or search can
|
|
return list of IDs if multiple matches found
|
|
"""
|
|
return self._searchIMDb('tt', title, kind)
|
|
|
|
def name2imdbID(self, name):
|
|
"""Translate a person name in an imdbID.
|
|
Try an Exact Primary Name search on IMDb;
|
|
return None if it's unable to get the imdbID."""
|
|
return self._searchIMDb('nm', name)
|
|
|
|
def character2imdbID(self, name):
|
|
"""Translate a character name in an imdbID.
|
|
Try an Exact Primary Name search on IMDb;
|
|
return None if it's unable to get the imdbID."""
|
|
return self._searchIMDb('char', name)
|
|
|
|
def company2imdbID(self, name):
|
|
"""Translate a company name in an imdbID.
|
|
Try an Exact Primary Name search on IMDb;
|
|
return None if it's unable to get the imdbID."""
|
|
return self._searchIMDb('co', name)
|
|
|
|
def get_imdbID(self, mop):
|
|
"""Return the imdbID for the given Movie, Person, Character or Company
|
|
object."""
|
|
imdbID = None
|
|
if mop.accessSystem == self.accessSystem:
|
|
aSystem = self
|
|
else:
|
|
aSystem = IMDb(mop.accessSystem)
|
|
if isinstance(mop, Movie.Movie):
|
|
if mop.movieID is not None:
|
|
imdbID = aSystem.get_imdbMovieID(mop.movieID)
|
|
else:
|
|
imdbID = aSystem.title2imdbID(build_title(mop, canonical=0,
|
|
ptdf=0, appendKind=False),
|
|
mop['kind'])
|
|
elif isinstance(mop, Person.Person):
|
|
if mop.personID is not None:
|
|
imdbID = aSystem.get_imdbPersonID(mop.personID)
|
|
else:
|
|
imdbID = aSystem.name2imdbID(build_name(mop, canonical=1))
|
|
elif isinstance(mop, Character.Character):
|
|
if mop.characterID is not None:
|
|
imdbID = aSystem.get_imdbCharacterID(mop.characterID)
|
|
else:
|
|
# canonical=0 ?
|
|
imdbID = aSystem.character2imdbID(build_name(mop, canonical=1))
|
|
elif isinstance(mop, Company.Company):
|
|
if mop.companyID is not None:
|
|
imdbID = aSystem.get_imdbCompanyID(mop.companyID)
|
|
else:
|
|
imdbID = aSystem.company2imdbID(build_company_name(mop))
|
|
else:
|
|
raise IMDbError('object ' + repr(mop) + \
|
|
' is not a Movie, Person or Character instance')
|
|
return imdbID
|
|
|
|
def get_imdbURL(self, mop):
|
|
"""Return the main IMDb URL for the given Movie, Person,
|
|
Character or Company object, or None if unable to get it."""
|
|
imdbID = self.get_imdbID(mop)
|
|
if imdbID is None:
|
|
return None
|
|
if isinstance(mop, Movie.Movie):
|
|
url_firstPart = imdbURL_movie_main
|
|
elif isinstance(mop, Person.Person):
|
|
url_firstPart = imdbURL_person_main
|
|
elif isinstance(mop, Character.Character):
|
|
url_firstPart = imdbURL_character_main
|
|
elif isinstance(mop, Company.Company):
|
|
url_firstPart = imdbURL_company_main
|
|
else:
|
|
raise IMDbError('object ' + repr(mop) + \
|
|
' is not a Movie, Person, Character or Company instance')
|
|
return url_firstPart % imdbID
|
|
|
|
def get_special_methods(self):
|
|
"""Return the special methods defined by the subclass."""
|
|
sm_dict = {}
|
|
base_methods = []
|
|
for name in dir(IMDbBase):
|
|
member = getattr(IMDbBase, name)
|
|
if isinstance(member, MethodType):
|
|
base_methods.append(name)
|
|
for name in dir(self.__class__):
|
|
if name.startswith('_') or name in base_methods or \
|
|
name.startswith('get_movie_') or \
|
|
name.startswith('get_person_') or \
|
|
name.startswith('get_company_') or \
|
|
name.startswith('get_character_'):
|
|
continue
|
|
member = getattr(self.__class__, name)
|
|
if isinstance(member, MethodType):
|
|
sm_dict.update({name: member.__doc__})
|
|
return sm_dict
|
|
|