# -*- coding: utf-8 -*-
# vi:si:et:sw=4:sts=4:ts=4
from datetime import datetime
import os.path
import random
import re
from django.db import models
from django.db.models import Q
from django.contrib.auth.models import User
from django.core.files.base import ContentFile
from django.utils import simplejson as json
from django.conf import settings
from oxdjango import fields
import oxlib
from oxlib import stripTags
from oxlib.normalize import canonicalTitle, canonicalName
from firefogg import Firefogg
import managers
import load
import utils
import extract
class MovieImdb(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
imdbId = models.CharField(max_length=7, unique=True)
title = models.CharField(max_length=1000)
year = models.CharField(max_length=4)
runtime = models.IntegerField(null=True, blank=True)
release_date = models.DateField(null=True, blank=True)
tagline = models.TextField(blank=True)
plot = models.TextField(blank=True)
plot_outline = models.TextField(blank=True)
rating = models.FloatField(null=True, blank=True)
votes = models.IntegerField(null=True, blank=True)
budget = models.IntegerField(null=True, blank=True)
gross = models.IntegerField(null=True, blank=True)
profit = models.IntegerField(null=True, blank=True)
series_imdb = models.CharField(max_length=7, default='')
series_title = models.CharField(max_length=1000, blank=True, default='')
episode_title = models.CharField(max_length=1000, blank=True, default='')
season = models.IntegerField(default=-1)
episode = models.IntegerField(default=-1)
def __unicode__(self):
return u"%s (%s)" % (self.title, self.imdbId)
class MovieOxdb(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
title = models.CharField(max_length=1000)
year = models.CharField(max_length=4)
runtime = models.IntegerField(null=True, blank=True)
release_date = models.DateField(null=True, blank=True)
tagline = models.TextField(blank=True)
plot = models.TextField(blank=True)
plot_outline = models.TextField(blank=True)
rating = models.FloatField(null=True, blank=True)
votes = models.IntegerField(null=True, blank=True)
budget = models.IntegerField(null=True, blank=True)
gross = models.IntegerField(null=True, blank=True)
profit = models.IntegerField(null=True, blank=True)
series_imdb = models.CharField(max_length=7, default='')
series_title = models.TextField(blank=True, default='')
episode_title = models.TextField(blank=True, default='')
season = models.IntegerField(default=-1)
episode = models.IntegerField(default=-1)
def __unicode__(self):
return u"%s (%s)" % (self.title, self.year)
class MovieExtra(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
title = models.CharField(max_length=1000)
description = models.TextField(blank=True)
contributor = models.CharField(max_length=1000)
rights_level = models.IntegerField(default=-1)
def getMovie(info):
'''
info dict with:
imdbId, title, director, episode_title, season, series
'''
if 'imdbId' in info and info['imdbId']:
try:
movie = Movie.byImdbId(info['imdbId'])
except Movie.DoesNotExist:
movie = load.loadIMDb(info['imdbId'])
else:
q = Movie.objects.filter(oxdb__title=info['title'])
if q.count() > 1:
print "FIXME: check more than title here!!"
movie = q[0]
else:
print info
movie = newMovie(info['title'], info['director'], '')
updated = False
for key in ('episode_title', 'season', 'year'):
if key in info:
setattr(movie.oxdb, key, info[key])
updated = True
if updated:
movie.save()
return movie
def newMovie(title, director, year):
movie = Movie()
oxdb = MovieOxdb()
oxdb.save()
movie.oxdb = oxdb
movie.oxdb.title = title
movie.oxdb.year = str(year)
movie.oxdb.save()
movie.oxdbId = "__init__%s" % random.randint(0, 100000)
movie.save()
movie.oxdbId = movie.oxid()
print title, director, year
print movie.oxdbId
print movie.movieId
movie.save()
return movie
def movie_path(f, size):
name = "%s.%s" % (size, 'ogv')
url_hash = f.movieId
return os.path.join('movie', url_hash[:2], url_hash[2:4], url_hash[4:6], name)
class Movie(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
published = models.DateTimeField(default=datetime.now, editable=False)
#only movies that have metadata from files are available,
#this is indicated by setting available to True
available = models.BooleanField(default=False, db_index=True)
movieId = models.CharField(max_length=128, unique=True, blank=True)
oxdbId = models.CharField(max_length=42, unique=True, blank=True)
imdb = models.OneToOneField('MovieImdb', null=True, related_name='movie')
oxdb = models.OneToOneField('MovieOxdb', null=True, related_name='movie')
extra = models.OneToOneField('MovieExtra', null=True, related_name='movie')
objects = managers.MovieManager()
def get(self, key, default=None):
if self.extra and getattr(self.extra, key):
return getattr(self.extra, key)
if self.oxdb and getattr(self.oxdb, key):
return getattr(self.oxdb, key)
if self.imdb:
return getattr(self.imdb, key)
return default
def editable(self, user):
#FIXME: make permissions work
return False
def edit(self, data):
#FIXME: how to map the keys to the right place to write them to?
for key in data:
if key != 'id':
setattr(self.oxdb, key, data[key])
self.oxdb.save()
self.save()
def _manual(self, qs, f='manual'):
if qs.filter(**{f:True}).count() > 0:
return qs.exclude(**{f:False})
return qs.exclude(**{f:True})
def directors(self):
qs = self.people.filter(cast__role='directors').order_by('cast__position')
return self._manual(qs, 'cast__manual')
def writers(self):
qs = self.people.filter(cast__role='writers').order_by('cast__position')
return self._manual(qs, 'cast__manual')
def editors(self):
qs = self.people.filter(cast__role='editors').order_by('cast__position')
return self._manual(qs, 'cast__manual')
def producers(self):
qs = self.people.filter(cast__role='producers').order_by('cast__position')
return self._manual(qs, 'cast__manual')
def cinematographers(self):
qs = self.people.filter(cast__role='cinematographers').order_by('cast__position')
return self._manual(qs, 'cast__manual')
def cast(self):
cast = []
qs = Cast.objects.filter(movie=self, role='cast').order_by('position')
qs = self._manual(qs)
for c in qs:
cast.append((c.person.name, c.character))
return tuple(cast)
def alternative_titles(self):
return self._manual(self.alternative_titles_all)
def genres(self):
return self._manual(self.genres_all)
def keywords(self):
return self._manual(self.keywords_all)
def countries(self):
return self._manual(self.countries_all, 'moviecountry__manual').order_by('moviecountry__position')
def languages(self):
return self._manual(self.languages_all, 'movielanguage__manual').order_by('movielanguage__position')
def trivia(self):
return self._manual(self.trivia_all)
def locations(self):
return self._manual(self.locations_all)
def connections(self):
return self._manual(self.connections_all)
def connections_json(self):
connections = {}
for connection in self.connections():
if connection.relation not in connections:
connections[connection.relation] = []
connections[connection.relation].append(connection.object.movieId)
return connections
def reviews(self):
q = self.reviews_all.filter(manual=True)
if q.count() > 0:
return q
whitelist = ReviewWhitelist.objects.all()
q = Q(id=-1)
for w in whitelist:
q = q | Q(url__contains=w.url)
return self.reviews_all.filter(q).filter(manual=False)
'''
#these values get populated with imdb/oxdb values on save()
#edits will be overwritten
title = models.CharField(max_length=1000)
year = models.CharField(max_length=4)
runtime = models.IntegerField(null=True, blank=True)
release_date = models.DateField(null=True, blank=True)
tagline = models.TextField(blank=True)
plot = models.TextField(blank=True)
plot_outline = models.TextField(blank=True)
rating = models.FloatField(null=True, blank=True)
votes = models.IntegerField(null=True, blank=True)
budget = models.IntegerField(null=True, blank=True)
gross = models.IntegerField(null=True, blank=True)
profit = models.IntegerField(null=True, blank=True)
series_imdb = models.CharField(max_length=7, default='')
series_title = models.TextField(blank=True, default='')
episode_title = models.TextField(blank=True, default='')
season = models.IntegerField(default=-1)
episode = models.IntegerField(default=-1)
'''
json = fields.DictField(default={}, editable=False)
'''
directors = fields.TupleField(default=())
writers = fields.TupleField(default=())
editors = fields.TupleField(default=())
producers = fields.TupleField(default=())
cinematographers = fields.TupleField(default=())
cast = fields.TupleField(default=())
alternative_titles = fields.TupleField(default=())
genres = fields.TupleField(default=())
keywords = fields.TupleField(default=())
countries = fields.TupleField(default=())
languages = fields.TupleField(default=())
trivia = fields.TupleField(default=())
locations = fields.TupleField(default=())
connections = fields.DictField(default={})
reviews = fields.TupleField(default=())
'''
#FIXME: use data.0xdb.org
'''
tpb_id = models.CharField(max_length=128, blank=True)
kg_id = models.CharField(max_length=128, blank=True)
open_subtitle_id = models.IntegerField(null=True, blank=True)
wikipedia_url = models.TextField(blank=True)
#FIXME: use data.0xdb.org/posters for that
#what of this is still required?
still_pos = models.IntegerField(null=True, blank=True)
poster = models.TextField(blank=True)
posters_disabled = models.TextField(blank=True)
posters_available = models.TextField(blank=True)
'''
poster_height = models.IntegerField(default=0)
poster_width = models.IntegerField(default=0)
#stream related fields
'''
'''
stream_low = models.FileField(default=None, blank=True, upload_to=lambda f, x: movie_path(f, 'low'))
stream_mid = models.FileField(default=None, blank=True, upload_to=lambda f, x: movie_path(f, 'mid'))
stream_high = models.FileField(default=None, blank=True, upload_to=lambda f, x: movie_path(f, 'high'))
#FIXME: is this still required? should this not be aspect ratio? depends on stream???
scene_height = models.IntegerField(null=True, blank=True)
def __unicode__(self):
return u'%s (%s)' % (self.get('title'), self.get('year'))
def save(self, *args, **kwargs):
if not self.oxdb:
oxdb = MovieOxdb()
oxdb.save()
self.oxdb = oxdb
if self.imdb:
mid = self.imdb.imdbId
else:
mid = self.oxdbId
self.movieId = mid
if self.id:
self.json = self.get_json()
'''
#populate auto values for faster results
#FIXME: why is it not possible to use setattr to set List instead of db value?
common_fields = [f.name for f in MovieImdb._meta.fields]
only_relevant = lambda f: f not in ('id', 'created', 'modified', 'imdbId')
common_fields = filter(only_relevant, common_fields)
for f in common_fields:
setattr(self, f, self.get(f))
for f in ('directors', 'writers', 'editors', 'producers', 'cinematographers',
'reviews', 'countries', 'languages',
'keywords', 'genres', 'trivia', 'alternative_titles'):
value = getattr(self, 'get_' + f)
setattr(self, f, tuple([v.json() for v in value()]))
for f in ('cast', ):
value = getattr(self, 'get_' + f)
setattr(self, f, value())
self.connections = self.connections_json()
'''
super(Movie, self).save(*args, **kwargs)
self.updateFind()
self.updateSort()
_public_fields = {
'movieId': 'id',
'title': 'title',
'year': 'year',
'runtime': 'runtime',
'release_date': 'release_date',
'countries': 'country',
'directors': 'director',
'languages': 'language',
'genres': 'genre',
'keywords': 'keyword',
'cast': 'cast',
'series_title': 'series_title',
'episode_title': 'episode_title',
'season': 'season',
'episode': 'episode',
'reviews': 'reviews',
'trivia': 'trivia',
'rating': 'rating',
'votes': 'votes',
'alternative_titles': 'alternative_titles',
'connections_json': 'connections',
'poster_width': 'posterWidth',
'poster_height': 'posterHeight'
}
def get_json(self, fields=None):
movie = {}
for key in self._public_fields:
pub_key = self._public_fields.get(key, key)
if not fields or pub_key in fields:
if hasattr(self, key):
value = getattr(self, key)
else:
value = self.get(key)
if key in ('directors', 'writers', 'editors', 'cinematographers', 'producers',
'reviews', 'countries', 'languages',
'keywords', 'genres', 'trivia', 'alternative_titles'):
movie[pub_key] = tuple([v.json() for v in value()])
elif callable(value):
movie[pub_key] = value()
else:
movie[pub_key] = value
if fields:
for f in fields:
if f.endswith('.length') and f[:-7] in ('cast', 'genre', 'trivia'):
movie[f] = getattr(self.sort.all()[0], f[:-7])
return movie
def fields(self):
fields = {}
for f in self._meta.fields:
if f.name in self._public_fields:
fields[f.name] = {}
fields[f.name]['order'] = 'desc'
fields[f.name]['type'] = type(f)
return fields
fields = classmethod(fields)
#Class functions to get Movies by ID, right now movieId, imdbId and oxdbId
#FIXME: this should go into a manager
def byMovieId(self, movieId):
if len(movieId) == 7:
return self.byImdbId(movieId)
return self.byOxdbId(movieId)
byMovieId = classmethod(byMovieId)
def byImdbId(self, imdbId):
return self.objects.get(imdb__imdbId=imdbId)
byImdbId = classmethod(byImdbId)
def byOxdbId(self, oxdbId):
return self.objects.get(oxdbId=oxdbId)
byOxdbId = classmethod(byOxdbId)
def oxid(self):
directors = ','.join([d.name for d in self.directors()])
return utils.oxid(self.get('title', ''), directors, self.get('year', ''),
self.get('series_title', ''), self.get('episode_title', ''),
self.get('season', ''), self.get('episode', ''))
def frame(self, position, width=128):
#FIXME: compute offset and so on
f = self.files.all()[0]
return f.frame(position, width)
def updateFind(self):
try:
f = self.find
except MovieFind.DoesNotExist:
f = MovieFind(movie=self)
f.title = self.get('title')
#FIXME: filter us/int title
#f.title += ' '.join([t.title for t in self.alternative_titles()])
f.director = '|%s|'%'|'.join([i.name for i in self.directors()])
f.country = '|%s|'%'|'.join([i.name for i in self.countries()])
f.year = self.get('year', '')
f.language = '|%s|'%'|'.join([i.name for i in self.languages()])
f.writer = '|%s|'%'|'.join([i.name for i in self.writers()])
f.producer = '|%s|'%'|'.join([i.name for i in self.producers()])
f.editor = '|%s|'%'|'.join([i.name for i in self.editors()])
f.cinematographer = '|%s|'%'|'.join([i.name for i in self.cinematographers()])
f.cast = ' '.join(['%s %s' % i for i in self.cast()])
f.genre = '|%s|'%'|'.join([i.name for i in self.genres()])
f.keywords = '|%s|'%'|'.join([i.name for i in self.keywords()])
f.summary = self.get('plot', '') + self.get('plot_outline', '')
f.trivia = ' '.join([i.trivia for i in self.trivia()])
f.location = '|%s|'%'|'.join([i.name for i in self.locations()])
#FIXME: collate filenames
#f.filename = self.filename
f.all = ' '.join(filter(None, [f.title, f.director, f.country, f.year, f.language,
f.writer, f.producer, f.editor, f.cinematographer,
f.cast, f.genre, f.keywords, f.summary, f.trivia,
f.location, f.filename]))
f.save()
def updateSort(self):
try:
s = self.sort
except MovieSort.DoesNotExist:
s = MovieSort(movie=self)
def sortName(value):
sort_value = '~'
if value:
sort_value = stripTags(value).split(',')
sort_value = '; '.join([canonicalName(name.strip()) for name in sort_value])
sort_value = sort_value.replace(u'\xc5k', 'A')
if not sort_value:
sort_value = '~'
return sort_value
#title
title = canonicalTitle(self.get('title'))
title = re.sub(u'[\'!¿¡,\.;\-"\:\*\[\]]', '', title)
title = title.replace(u'Æ', 'Ae')
#pad numbered titles
numbers = re.compile('^(\d+)').findall(title)
if numbers:
padded = '%010d' % int(numbers[0])
title = re.sub('^(\d+)', padded, title)
s.title = title.strip()
directors = ','.join([i.name for i in self.directors()])
s.director = sortName(directors)
s.country = ','.join([i.name for i in self.countries()])
s.year = self.get('year', '')
names = ','.join([i.name for i in self.producers()])
s.producer = sortName(names)
names = ','.join([i.name for i in self.writers()])
s.writer = sortName(names)
names = ','.join([i.name for i in self.editors()])
s.editor = sortName(names)
names = ','.join([i.name for i in self.cinematographers()])
s.cinematographer = sortName(names)
s.language = ','.join([i.name for i in self.languages()])
s.country = ','.join([i.name for i in self.countries()])
s.runtime = self.get('runtime', 0)
s.keywords = self.keywords().count()
s.genre = self.genres().count()
s.cast = len(self.cast())
s.summary = len(self.get('plot', '').split())
s.trivia = self.trivia().count()
s.connections = self.connections().count()
s.movieId = self.movieId
s.rating = self.get('rating', -1)
s.votes = self.get('votes', -1)
# data from related subtitles
s.scenes = 0 #FIXME
s.words = 0 #FIXME
s.wpm = 0 #FIXME
s.risk = 0 #FIXME
# data from related files
s.duration = 0 #FIXME
s.resolution = 0 #FIXME
s.aspectratio = 0 #FIXME
s.bitrate = 0 #FIXME
s.pixels = 0 #FIXME
s.filename = 0 #FIXME
s.files = 0 #FIXME
s.size = 0 #FIXME
s.save()
class MovieFind(models.Model):
"""
used to search movies, all search values are in here
"""
movie = models.OneToOneField('Movie', related_name='find', primary_key=True)
all = models.TextField(blank=True)
title = models.TextField(blank=True)
director = models.TextField(blank=True, default='')
country = models.TextField(blank=True, default='')
year = models.CharField(max_length=4)
language = models.TextField(blank=True, default='')
writer = models.TextField(blank=True, default='')
producer = models.TextField(blank=True, default='')
editor = models.TextField(blank=True, default='')
cinematographer = models.TextField(blank=True, default='')
cast = models.TextField(blank=True, default='')
#person
genre = models.TextField(blank=True)
keywords = models.TextField(blank=True)
summary = models.TextField(blank=True)
trivia = models.TextField(blank=True)
locations = models.TextField(blank=True, default='')
#only for own files or as admin?
filename = models.TextField(blank=True, default='')
_private_fields = ('id', 'movie')
#return available find fields
#FIXME: should return mapping name -> verbose_name
def fields(self):
fields = []
for f in self._meta.fields:
if f.name not in self._private_fields:
name = f.verbose_name
name = name[0].capitalize() + name[1:]
fields.append(name)
return tuple(fields)
fields = classmethod(fields)
class MovieSort(models.Model):
"""
used to sort movies, all sort values are in here
"""
movie = models.OneToOneField('Movie', related_name='sort', primary_key=True)
title = models.CharField(max_length=1000, db_index=True)
director = models.TextField(blank=True, db_index=True)
country = models.TextField(blank=True, db_index=True)
year = models.CharField(max_length=4, db_index=True)
producer = models.TextField(blank=True, db_index=True)
writer = models.TextField(blank=True, db_index=True)
editor = models.TextField(blank=True, db_index=True)
cinematographer = models.TextField(blank=True, db_index=True)
language = models.TextField(blank=True, db_index=True)
runtime = models.IntegerField(blank=True, null=True, db_index=True)
keywords = models.IntegerField(blank=True, db_index=True)
genre = models.TextField(blank=True, db_index=True)
cast = models.IntegerField(blank=True, db_index=True)
summary = models.IntegerField(blank=True, db_index=True)
trivia = models.IntegerField(blank=True, db_index=True)
connections = models.IntegerField(blank=True, db_index=True)
rating = models.FloatField(blank=True, db_index=True)
votes = models.IntegerField(blank=True, db_index=True)
scenes = models.IntegerField(blank=True, db_index=True)
words = models.IntegerField(null=True, blank=True, db_index=True)
wpm = models.IntegerField('Words per Minute', null=True, blank=True, db_index=True)
risk = models.IntegerField(null=True, blank=True, db_index=True)
movieId = models.CharField('ID', max_length=128, blank=True, db_index=True)
duration = models.FloatField(default=-1, db_index=True)
resolution = models.IntegerField(blank=True, db_index=True)
aspectratio = models.IntegerField('Aspect Ratio', blank=True, db_index=True)
bitrate = models.IntegerField(blank=True, db_index=True)
pixels = models.BigIntegerField(blank=True, db_index=True)
filename = models.IntegerField(blank=True, db_index=True)
files = models.IntegerField(blank=True, db_index=True)
size = models.BigIntegerField(blank=True, db_index=True)
_private_fields = ('id', 'movie')
#return available sort fields
#FIXME: should return mapping name -> verbose_name
def fields(self):
fields = []
for f in self._meta.fields:
if f.name not in self._private_fields:
name = f.verbose_name
name = name[0].capitalize() + name[1:]
fields.append(name)
return tuple(fields)
fields = classmethod(fields)
class AlternativeTitle(models.Model):
movie = models.ForeignKey(Movie, related_name='alternative_titles_all')
title = models.TextField()
type = models.CharField(max_length=1000)
manual = models.BooleanField(default=False)
class Meta:
ordering = ('title', )
def __unicode__(self):
return self.title
def json(self):
return (self.title, self.type)
class Person(models.Model):
name = models.CharField(max_length=200)
imdbId = models.CharField(max_length=7, blank=True)
name_sort = models.CharField(max_length=200)
movies = models.ManyToManyField(Movie, related_name='people', through='Cast')
class Meta:
ordering = ('name_sort', )
def __unicode__(self):
return self.name
def save(self, *args, **kwargs):
if not self.name_sort:
self.name_sort = oxlib.normalize.canonicalName(self.name)
super(Person, self).save(*args, **kwargs)
def get_or_create(model, name, imdbId=None):
if imdbId:
q = model.objects.filter(name=name, imdbId=imdbId)
else:
q = model.objects.all().filter(name=name)
if q.count() > 0:
o = q[0]
else:
o = model.objects.create(name=name)
if imdbId:
o.imdbId = imdbId
o.save()
return o
get_or_create = classmethod(get_or_create)
def json(self):
return self.name
class Cast(models.Model):
movie = models.ForeignKey(Movie, related_name='cast_relation')
person = models.ForeignKey(Person)
role = models.CharField(max_length=200)
character = models.CharField(max_length=1000, blank=True)
position = models.IntegerField()
manual = models.BooleanField(default=False)
class Meta:
ordering = ('position', 'person__name_sort')
def __unicode__(self):
return "%s <> %s" % (self.person, self.movie)
def link(self, movie, person, role, character, position, manual=False):
q = self.objects.filter(movie=movie, person=person, role=role, character=character)
if q.count() > 0:
link = q[0]
link.position = position
link.manual = manual
link.save()
else:
link = self()
link.movie=movie
link.person=person
link.role=role
link.character=character
link.position = position
link.manual = manual
link.save()
return link
link = classmethod(link)
def json(self):
return (self.person.json(), self.character)
class Country(models.Model):
name = models.CharField(max_length=200, unique=True)
movies = models.ManyToManyField(Movie, related_name='countries_all', through='MovieCountry')
class Meta:
#!! adding this to ordering, breaks:
# models.Country.objects.values("name").annotate(movies=Count('movies'))
#'moviecountry__position',
ordering = ('name', )
def __unicode__(self):
return self.name
def json(self):
return self.name
class MovieCountry(models.Model):
movie = models.ForeignKey(Movie)
country = models.ForeignKey(Country)
position = models.IntegerField()
manual = models.BooleanField(default=False)
class Meta:
ordering = ('position', 'country')
def __unicode__(self):
return "%s <> %s" % (self.country, self.movie)
def link(self, movie, country, position):
q = self.objects.filter(movie=movie, country=country)
if q.count() > 0:
link = q[0]
link.position = position
link.save()
else:
link = self()
link.movie=movie
link.country=country
link.position=position
link.save()
return link
link = classmethod(link)
class Language(models.Model):
name = models.CharField(max_length=200, unique=True)
movies = models.ManyToManyField(Movie, related_name='languages_all', through="MovieLanguage")
class Meta:
ordering = ('name', )
def __unicode__(self):
return self.name
def json(self):
return self.name
class MovieLanguage(models.Model):
movie = models.ForeignKey(Movie)
language = models.ForeignKey(Language)
position = models.IntegerField()
manual = models.BooleanField(default=False)
class Meta:
ordering = ('position', 'language')
def __unicode__(self):
return self.language.name
def link(self, movie, language, position):
q = self.objects.filter(movie=movie, language=language)
if q.count() > 0:
link = q[0]
link.position = position
link.save()
else:
link = self()
link.movie=movie
link.language=language
link.position=position
link.save()
return link
link = classmethod(link)
class Keyword(models.Model):
name = models.CharField(max_length=200, unique=True)
manual = models.BooleanField(default=False)
movies = models.ManyToManyField(Movie, related_name='keywords_all')
class Meta:
ordering = ('name', )
def __unicode__(self):
return self.name
def json(self):
return self.name
class Genre(models.Model):
name = models.CharField(max_length=200, unique=True)
manual = models.BooleanField(default=False)
movies = models.ManyToManyField(Movie, related_name='genres_all')
class Meta:
ordering = ('name', )
def __unicode__(self):
return self.name
def json(self):
return self.name
class Location(models.Model):
name = models.CharField(max_length=200, unique=True)
manual = models.BooleanField(default=False)
movies = models.ManyToManyField(Movie, related_name='locations_all')
#fixme: geo data
lat_sw = models.FloatField(default=0)
lng_sw = models.FloatField(default=0)
lat_ne = models.FloatField(default=0)
lng_ne = models.FloatField(default=0)
lat_center = models.FloatField(default=0)
lng_center = models.FloatField(default=0)
area = models.FloatField(default=-1)
class Meta:
ordering = ('name', )
def __unicode__(self):
return self.name
def json(self):
return self.name
class Trivia(models.Model):
trivia = models.TextField()
manual = models.BooleanField(default=False)
position = models.IntegerField()
movie = models.ForeignKey(Movie, related_name='trivia_all')
class Meta:
ordering = ('position', )
def __unicode__(self):
return self.trivia
def json(self):
trivia = self.trivia
trivia = oxlib.fixAmpersands(trivia)
trivia = re.sub(' ', '', trivia)
trivia = re.sub('(.*?)', '\\2', trivia)
trivia = re.sub('(.*?)', '\\2', trivia)
return trivia
class Connection(models.Model):
subject = models.ForeignKey(Movie, related_name='connections_all')
relation = models.CharField(max_length=512)
object = models.ForeignKey(Movie)
manual = models.BooleanField(default=False)
def get_or_create(model, subject, relation, object, reverse=True, manual=False):
q = model.objects.filter(subject=subject, relation=relation, object=object)
if q.count() > 0:
o = q[0]
else:
o = model.objects.create(subject=subject, relation=relation, object=object, manual=manual)
o.save()
if reverse:
_map = {
'Edited into': 'Edited from',
'Features': 'Featured in',
'Follows': 'Followed by',
'References': 'Referenced in',
'Remake of': 'Remade as',
'Spin off from': 'Spin off',
'Spoofs': 'Spoofed in',
'Version of': 'Version of',
'Alternate language version of': 'Alternate language version of',
}
if relation in _map.values():
for k in _map:
if _map[k] == relation:
reverse_relation = k
else:
reverse_relation = _map[relation]
o2 = model.get_or_create(object, reverse_relation, subject, reverse=False)
return o
get_or_create = classmethod(get_or_create)
def __unicode__(self):
return '%s %s %s' % (self.subject, self.relation, self.object)
class Review(models.Model):
movie = models.ForeignKey('Movie', related_name='reviews_all')
title = models.CharField(blank=True, max_length=2048)
url = models.CharField(blank=True, max_length=2048)
manual = models.BooleanField(default=False)
def __unicode__(self):
return self.title
def get_or_create(self, movie, url):
q = self.objects.filter(movie=movie, url=url)
if q.count() > 0:
o = q[0]
else:
o = self.objects.create(movie=movie, url=url)
o.save()
return o
get_or_create = classmethod(get_or_create)
def name(self):
for w in ReviewWhitelist.objects.all():
if w.url in self.url:
return w.name
return self.title
def json(self):
return (self.name(), self.url)
class ReviewWhitelist(models.Model):
name = models.CharField(max_length=255, unique=True)
url = models.CharField(max_length=255, unique=True)
class List(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
user = models.ForeignKey(User)
name = models.CharField(max_length=255, unique=True)
public = models.BooleanField(default=False)
movies = models.ManyToManyField(Movie, related_name='lists', through='ListItem')
def add(self, movie):
q = self.movies.filter(id=movie.id)
if q.count() == 0:
l = ListItem()
l.list = self
l.movie = movie
l.save()
def remove(self, movie):
self.ListItem.objects.all().filter(movie=movie, list=self).delete()
def __unicode__(self):
return u'%s (%s)' % (self.title, unicode(self.user))
def editable(self, user):
#FIXME: make permissions work
if self.user == user or user.has_perm('Ox.admin'):
return True
return False
class ListItem(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
list = models.ForeignKey(List)
movie = models.ForeignKey(Movie)
def __unicode__(self):
return u'%s in %s' % (unicode(self.movie), unicode(self.list))
def stream_path(f, size):
name = "%s.%s" % (size, 'ogv')
url_hash = f.oshash
return os.path.join(url_hash[:2], url_hash[2:4], url_hash[4:6], url_hash, name)
def timeline_path(f):
name = "timeline.png"
url_hash = f.oshash
return os.path.join(url_hash[:2], url_hash[2:4], url_hash[4:6], url_hash, name)
def frame_path(f):
position = oxlib.formatDuration(f.position*1000).replace(':', '.')
name = "%s.%s" % (position, 'png')
url_hash = f.file.oshash
return os.path.join(url_hash[:2], url_hash[2:4], url_hash[4:6], url_hash, 'frames', name)
FILE_TYPES = (
(0, 'unknown'),
(1, 'video'),
(2, 'audio'),
(3, 'subtitle'),
)
class File(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
published = models.DateTimeField(default=datetime.now, editable=False)
verified = models.BooleanField(default=False)
oshash = models.CharField(blank=True, unique=True, max_length=16)
sha1 = models.CharField(blank=True, null=True, unique=True, max_length=40)
md5 = models.CharField(blank=True, null=True, unique=True, max_length=32)
movie = models.ForeignKey(Movie, related_name="files", default=None, null=True)
type = models.IntegerField(default=0, choices=FILE_TYPES)
info = fields.DictField(default={})
#FIXME: why do i need those in the db? could just have them in info
path = models.CharField(blank=True, max_length=2048)
size = models.BigIntegerField(default=-1)
duration = models.FloatField(default=-1)
video_codec = models.CharField(blank=True, max_length=256)
pixel_format = models.CharField(blank=True, max_length=256)
width = models.IntegerField(default=-1)
height = models.IntegerField(default=-1)
pixel_aspect_ratio = models.CharField(blank=True, max_length=256)
display_aspect_ratio = models.CharField(blank=True, max_length=256)
framerate = models.CharField(blank=True, max_length=256)
audio_codec = models.CharField(blank=True, max_length=256)
samplerate = models.IntegerField(default=-1)
channels = models.IntegerField(default=-1)
#computed values
bpp = models.FloatField(default=-1)
pixels = models.BigIntegerField(default=0)
part = models.IntegerField(default=0)
needs_data = models.BooleanField(default=True)
#stream related fields
available = models.BooleanField(default=False)
stream_low = models.FileField(default=None, upload_to=lambda f, x: stream_path(f, 'low'))
stream_mid = models.FileField(default=None, upload_to=lambda f, x: stream_path(f, 'mid'))
stream_high = models.FileField(default=None, upload_to=lambda f, x: stream_path(f, 'high'))
def timeline_base_url(self):
return '%s/timeline' % os.path.dirname(self.stream_low.url)
def save_chunk(self, chunk, name='video.ogv'):
if not self.available:
#FIXME: this should use stream_low or stream_high depending on configuration
video = getattr(self, 'stream_%s'%settings.VIDEO_PROFILE)
if not video:
video.save(name, chunk)
self.save()
else:
f = open(video.path, 'a')
f.write(chunk.read())
f.close()
return True
print "somehing failed, not sure what?", self.available
return False
objects = managers.FileManager()
def __unicode__(self):
return "%s (%s)" % (self.path, self.oshash)
def update(self, data=None):
"""
only add, do not overwrite keys in file
"""
if data and not self.info:
self.info = data
_video_map = {
'codec': 'video_codec',
}
if 'video' in self.info and self.info['video']:
for key in ('codec', 'pixel_format', 'width', 'height',
'pixel_aspect_ratio', 'display_aspect_ratio', 'framerate'):
if key in self.info['video'][0]:
setattr(self, _video_map.get(key, key), self.info['video'][0][key])
_audio_map = {
'codec': 'audio_codec',
}
if 'audio' in self.info and self.info['audio']:
for key in ('codec', 'samplerate', 'channels'):
if key in self.info['audio'][0]:
setattr(self, _audio_map.get(key, key), self.info['audio'][0][key])
for key in ('duration', 'size', 'sha1', 'md5'):
if key in self.info:
setattr(self, key, self.info[key])
#detect and assign type based on extension or detected video track
if os.path.splitext(self.info['path'])[-1] in ('.rar', '.sub', '.idx'):
self.type = 0
elif 'video' in self.info:
self.type = 1
elif os.path.splitext(self.info['path'])[-1] in ('.mp3', '.oga'):
self.type = 2
elif os.path.splitext(self.info['path'])[-1] in ('.srt', ):
self.type = 3
#FIXME: this should be computed and not submitted path
self.path = self.info['path']
self.save()
def findMovie(self):
info = utils.parsePath(self.path)
self.movie = getMovie(info)
self.save()
def extract_timeline(self):
if self.stream_high:
video = self.stream_high.path
elif self.stream_mid:
video = self.stream_mid.path
elif self.stream_low:
video = self.stream_low.path
else:
return False
prefix = os.path.join(os.path.dirname(video), 'timeline')
cmd = ['oxtimeline', '-i', video, '-o', prefix]
p = subprocess.Popen(cmd)
p.wait()
return p.returncode == 0
def extract_video(self):
ogg = Firefogg()
source = None
profiles = []
if self.stream_high:
source = self.stream_high
profiles = ['low', 'mid']
elif self.stream_mid:
source = self.stream_mid
profiles = ['low', ]
for profile in profiles:
output = getattr(self, 'stream_%s'%profile)
output.name = stream_path(self, profile)
output.save()
ogg.encode(source.path, output.path, settings.VIDEO_ENCODING[profile])
def extract(self):
#FIXME: do stuff, like create timeline or create smaller videos etc
self.extract_video()
self.extract_timeline()
return
def frame(self, position, width=128):
videoFile = getattr(self, 'stream_%s'%settings.VIDEO_PROFILE).path
frameFolder = os.path.join(os.path.dirname(videoFile), 'frames')
if position<= self.duration:
return extract.frame(videoFile, position, frameFolder, width)
return None
def editable(self, user):
'''
#FIXME: this should use a queryset!!!
archives = []
for a in self.archive_files.all():
archives.append(a.archive)
users = []
for a in archives:
users += a.users.all()
return user in users
'''
return self.archive_files.filter(archive__users__id=user.id).count() > 0
class Frame(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
file = models.ForeignKey(File, related_name="frames")
position = models.FloatField()
frame = models.ImageField(default=None, null=True, upload_to=lambda f, x: frame_path(f))
#FIXME: frame path should be renamed on save to match current position
def __unicode__(self):
return '%s at %s' % (self.file, self.position)
class Layer(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
user = models.ForeignKey(User)
movie = models.ForeignKey(Movie)
#seconds
time_in = models.FloatField(default=-1)
time_out = models.FloatField(default=-1)
type = models.CharField(blank=True, max_length=255)
value = models.TextField()
#FIXME: relational layers, Locations, clips etc
#location = models.ForeignKey('Location', default=None)
def editable(self, user):
if user.is_authenticated():
if obj.user == user.id or user.has_perm('0x.admin'):
return True
if user.groups.filter(id__in=obj.groups.all()).count() > 0:
return True
return False
class Archive(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
name = models.CharField(max_length=255, unique=True)
public = models.BooleanField(default=False)
users = models.ManyToManyField(User, related_name='archives')
def __unicode__(self):
return '%s' % (self.name)
def editable(self, user):
return self.users.filter(id=user.id).count() > 0
class ArchiveFile(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
archive = models.ForeignKey(Archive, related_name='files')
file = models.ForeignKey(File, related_name='archive_files')
path = models.CharField(blank=True, max_length=2048)
objects = managers.ArchiveFileManager()
def update(self, data):
"""
only add, do not overwrite keys in file
"""
self.file.update(data)
self.path = data.get('path', '')
self.save()
def get_or_create(model, archive, oshash):
try:
f = model.objects.by_oshash(oshash=oshash)
except model.DoesNotExist:
file, created = File.objects.get_or_create(oshash)
if created:
file.save()
f = model.objects.create(archive=archive, file=file)
f.save()
return f
get_or_create = classmethod(get_or_create)
def __unicode__(self):
return '%s (%s)' % (self.path, unicode(self.archive))
def editable(self, user):
return self.archive.editable(user)
class Collection(models.Model):
created = models.DateTimeField(auto_now_add=True)
modified = models.DateTimeField(auto_now=True)
users = models.ManyToManyField(User, related_name='collections')
name = models.CharField(blank=True, max_length=2048)
subdomain = models.CharField(unique=True, max_length=2048)
movies = models.ForeignKey(Movie)
def editable(self, user):
return self.users.filter(id=user.id).count() > 0