You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

331 lines
12 KiB

import traceback
11 years ago
from couchpotato import get_db
from couchpotato.api import addApiView
14 years ago
from couchpotato.core.event import addEvent
from couchpotato.core.helpers.encoding import toUnicode, ss
from couchpotato.core.helpers.variable import mergeDicts, getExt
14 years ago
from couchpotato.core.logger import CPLog
from couchpotato.core.plugins.base import Plugin
11 years ago
from couchpotato.core.plugins.quality.index import QualityIndex
import re
14 years ago
log = CPLog(__name__)
class QualityPlugin(Plugin):
14 years ago
qualities = [
{'identifier': 'bd50', 'hd': True, 'size': (15000, 60000), 'label': 'BR-Disk', 'alternative': ['bd25'], 'allow': ['1080p'], 'ext':[], 'tags': ['bdmv', 'certificate', ('complete', 'bluray')]},
{'identifier': '1080p', 'hd': True, 'size': (4000, 20000), 'label': '1080p', 'width': 1920, 'height': 1080, 'alternative': [], 'allow': [], 'ext':['mkv', 'm2ts'], 'tags': ['m2ts', 'x264', 'h264']},
{'identifier': '720p', 'hd': True, 'size': (3000, 10000), 'label': '720p', 'width': 1280, 'height': 720, 'alternative': [], 'allow': [], 'ext':['mkv', 'ts'], 'tags': ['x264', 'h264']},
{'identifier': 'brrip', 'hd': True, 'size': (700, 7000), 'label': 'BR-Rip', 'alternative': ['bdrip'], 'allow': ['720p', '1080p'], 'ext':[], 'tags': ['hdtv', 'hdrip', 'webdl', ('web', 'dl')]},
{'identifier': 'dvdr', 'size': (3000, 10000), 'label': 'DVD-R', 'alternative': ['br2dvd'], 'allow': [], 'ext':['iso', 'img', 'vob'], 'tags': ['pal', 'ntsc', 'video_ts', 'audio_ts', ('dvd', 'r')]},
{'identifier': 'dvdrip', 'size': (600, 2400), 'label': 'DVD-Rip', 'width': 720, 'alternative': [], 'allow': [], 'ext':[], 'tags': [('dvd', 'rip'), ('dvd', 'xvid'), ('dvd', 'divx')]},
{'identifier': 'scr', 'size': (600, 1600), 'label': 'Screener', 'alternative': ['screener', 'dvdscr', 'ppvrip', 'dvdscreener', 'hdscr'], 'allow': ['dvdr', 'dvdrip', '720p', '1080p'], 'ext':[], 'tags': ['webrip', ('web', 'rip')]},
{'identifier': 'r5', 'size': (600, 1000), 'label': 'R5', 'alternative': ['r6'], 'allow': ['dvdr'], 'ext':[]},
{'identifier': 'tc', 'size': (600, 1000), 'label': 'TeleCine', 'alternative': ['telecine'], 'allow': [], 'ext':[]},
{'identifier': 'ts', 'size': (600, 1000), 'label': 'TeleSync', 'alternative': ['telesync', 'hdts'], 'allow': [], 'ext':[]},
{'identifier': 'cam', 'size': (600, 1000), 'label': 'Cam', 'alternative': ['camrip', 'hdcam'], 'allow': [], 'ext':[]}
14 years ago
]
pre_releases = ['cam', 'ts', 'tc', 'r5', 'scr']
cached_qualities = None
cached_order = None
14 years ago
def __init__(self):
addEvent('quality.all', self.all)
addEvent('quality.single', self.single)
addEvent('quality.guess', self.guess)
addEvent('quality.pre_releases', self.preReleases)
11 years ago
addEvent('quality.order', self.getOrder)
addApiView('quality.size.save', self.saveSize)
addApiView('quality.list', self.allView, docs = {
'desc': 'List all available qualities',
'return': {'type': 'object', 'example': """{
'success': True,
'list': array, qualities
}"""}
})
addEvent('app.initialize', self.fill, priority = 10)
11 years ago
addEvent('database.setup', self.databaseSetup)
14 years ago
addEvent('app.test', self.doTest)
11 years ago
self.addOrder()
def addOrder(self):
self.order = []
for q in self.qualities:
self.order.append(q.get('identifier'))
def getOrder(self):
return self.order
def databaseSetup(self):
db = get_db()
# Quality index
try:
db.add_index(QualityIndex(db.path, 'quality'))
except:
log.debug('Index already exists')
db.edit_index(QualityIndex(db.path, 'quality'))
def preReleases(self):
return self.pre_releases
def allView(self, **kwargs):
return {
'success': True,
'list': self.all()
}
14 years ago
def all(self):
if self.cached_qualities:
return self.cached_qualities
11 years ago
db = get_db()
14 years ago
11 years ago
qualities = db.all('quality', with_doc = True)
14 years ago
temp = []
for quality in qualities:
11 years ago
quality = quality['doc']
q = mergeDicts(self.getQuality(quality.get('identifier')), quality)
14 years ago
temp.append(q)
self.cached_qualities = temp
11 years ago
14 years ago
return temp
def single(self, identifier = ''):
11 years ago
db = get_db()
quality_dict = {}
11 years ago
quality = db.get('quality', identifier, with_doc = True)['doc']
if quality:
11 years ago
quality_dict = mergeDicts(self.getQuality(quality['identifier']), quality)
return quality_dict
14 years ago
def getQuality(self, identifier):
for q in self.qualities:
if identifier == q.get('identifier'):
return q
def saveSize(self, **kwargs):
try:
11 years ago
db = get_db()
quality = db.get('quality', kwargs.get('identifier'), with_doc = True)
if quality:
11 years ago
quality['doc'][kwargs.get('value_type')] = kwargs.get('value')
db.update(quality['doc'])
self.cached_qualities = None
return {
'success': True
}
except:
log.error('Failed: %s', traceback.format_exc())
return {
'success': False
}
14 years ago
def fill(self):
try:
11 years ago
db = get_db()
order = 0
for q in self.qualities:
11 years ago
db.insert({
'_t': 'quality',
'order': order,
'identifier': q.get('identifier'),
'size_min': q.get('size')[0],
'size_max': q.get('size')[1]
})
log.info('Creating profile: %s', q.get('label'))
db.insert({
'_t': 'profile',
'order': order + 20, # Make sure it goes behind other profiles
'core': True,
'qualities': [q.get('identifier')],
'label': toUnicode(q.get('label')),
'finish': [True],
'wait_for': [0],
})
order += 1
14 years ago
return True
except:
log.error('Failed: %s', traceback.format_exc())
14 years ago
return False
def guess(self, files, extra = None):
if not extra: extra = {}
# Create hash for cache
cache_key = str([f.replace('.' + getExt(f), '') if len(getExt(f)) < 4 else f for f in files])
cached = self.getCache(cache_key)
if cached and len(extra) == 0:
return cached
qualities = self.all()
# Start with 0
score = {}
for quality in qualities:
score[quality.get('identifier')] = 0
for cur_file in files:
words = re.split('\W+', cur_file.lower())
for quality in qualities:
contains_score = self.containsTagScore(quality, words, cur_file)
self.calcScore(score, quality, contains_score)
# Try again with loose testing
for quality in qualities:
loose_score = self.guessLooseScore(quality, files = files, extra = extra)
self.calcScore(score, quality, loose_score)
# Return nothing if all scores are 0
has_non_zero = 0
for s in score:
if score[s] > 0:
has_non_zero += 1
if not has_non_zero:
return None
heighest_quality = max(score, key = score.get)
if heighest_quality:
for quality in qualities:
if quality.get('identifier') == heighest_quality:
return self.setCache(cache_key, quality)
return None
def containsTagScore(self, quality, words, cur_file = ''):
cur_file = ss(cur_file)
score = 0
points = {
'identifier': 10,
'label': 10,
'alternative': 9,
'tags': 9,
'ext': 3,
}
# Check alt and tags
for tag_type in ['identifier', 'alternative', 'tags', 'label']:
qualities = quality.get(tag_type, [])
qualities = [qualities] if isinstance(qualities, (str, unicode)) else qualities
for alt in qualities:
if (isinstance(alt, tuple)):
if len(set(words) & set(alt)) == len(alt):
log.debug('Found %s via %s %s in %s', (quality['identifier'], tag_type, quality.get(tag_type), cur_file))
score += points.get(tag_type)
if (isinstance(alt, (str, unicode)) and ss(alt.lower()) in cur_file.lower()):
log.debug('Found %s via %s %s in %s', (quality['identifier'], tag_type, quality.get(tag_type), cur_file))
score += points.get(tag_type) / 2
if list(set(qualities) & set(words)):
log.debug('Found %s via %s %s in %s', (quality['identifier'], tag_type, quality.get(tag_type), cur_file))
score += points.get(tag_type)
# Check extention
for ext in quality.get('ext', []):
if ext == words[-1]:
log.debug('Found %s extension in %s', (ext, cur_file))
score += points['ext']
return score
def guessLooseScore(self, quality, files = None, extra = None):
score = 0
if extra:
# Check width resolution, range 20
if quality.get('width') and (quality.get('width') - 20) <= extra.get('resolution_width', 0) <= (quality.get('width') + 20):
log.debug('Found %s via resolution_width: %s == %s', (quality['identifier'], quality.get('width'), extra.get('resolution_width', 0)))
score += 5
# Check height resolution, range 20
if quality.get('height') and (quality.get('height') - 20) <= extra.get('resolution_height', 0) <= (quality.get('height') + 20):
log.debug('Found %s via resolution_height: %s == %s', (quality['identifier'], quality.get('height'), extra.get('resolution_height', 0)))
score += 5
if quality.get('identifier') == 'dvdrip' and 480 <= extra.get('resolution_width', 0) <= 720:
log.debug('Add point for correct dvdrip resolutions')
score += 1
return score
def calcScore(self, score, quality, add_score):
score[quality['identifier']] += add_score
# Set order for allow calculation (and cache)
if not self.cached_order:
self.cached_order = {}
for q in self.qualities:
self.cached_order[q.get('identifier')] = self.qualities.index(q)
if add_score != 0:
for allow in quality.get('allow', []):
score[allow] -= 40 if self.cached_order[allow] < self.cached_order[quality['identifier']] else 5
def doTest(self):
tests = {
'Movie Name (1999)-DVD-Rip.avi': 'dvdrip',
'Movie Name 1999 720p Bluray.mkv': '720p',
'Movie Name 1999 BR-Rip 720p.avi': 'brrip',
'Movie Name 1999 720p Web Rip.avi': 'scr',
'Movie Name 1999 Web DL.avi': 'brrip',
'Movie.Name.1999.1080p.WEBRip.H264-Group': 'scr',
'Movie.Name.1999.DVDRip-Group': 'dvdrip',
'Movie.Name.1999.DVD-Rip-Group': 'dvdrip',
'Movie.Name.1999.DVD-R-Group': 'dvdr',
'Movie.Name.Camelie.1999.720p.BluRay.x264-Group': '720p',
}
correct = 0
for name in tests:
success = self.guess([name]).get('identifier') == tests[name]
if not success:
log.error('%s failed check, thinks it\'s %s', (name, self.guess([name]).get('identifier')))
correct += success
if correct == len(tests):
log.info('Quality test successful')
return True
else:
log.error('Quality test failed: %s out of %s succeeded', (correct, len(tests)))