1
0
mirror of https://github.com/arsenetar/dupeguru.git synced 2024-11-16 04:09:02 +00:00
dupeguru/base/py/scanner.py

133 lines
5.7 KiB
Python
Raw Normal View History

# Created By: Virgil Dupras
# Created On: 2006/03/03
# $Id$
# Copyright 2009 Hardcoded Software (http://www.hardcoded.net)
#
# This software is licensed under the "HS" License as described in the "LICENSE" file,
# which should be included with this package. The terms are also available at
# http://www.hardcoded.net/licenses/hs_license
import logging
from ignore import IgnoreList
from hsutil import job
from hsutil.misc import dedupe
from hsutil.str import get_file_ext, rem_file_ext
from . import engine
(SCAN_TYPE_FILENAME,
SCAN_TYPE_FIELDS,
SCAN_TYPE_FIELDS_NO_ORDER,
SCAN_TYPE_TAG,
SCAN_TYPE_TAG_WITH_ALBUM, # Obsolete
SCAN_TYPE_CONTENT,
SCAN_TYPE_CONTENT_AUDIO) = range(7)
SCANNABLE_TAGS = ['track', 'artist', 'album', 'title', 'genre', 'year']
class Scanner(object):
def __init__(self):
self.ignore_list = IgnoreList()
self.discarded_file_count = 0
def _getmatches(self, files, j):
j = j.start_subjob(2)
mf = engine.MatchFactory()
if self.scan_type != SCAN_TYPE_CONTENT:
mf.match_similar_words = self.match_similar_words
mf.weight_words = self.word_weighting
mf.min_match_percentage = self.min_match_percentage
if self.scan_type == SCAN_TYPE_FIELDS_NO_ORDER:
self.scan_type = SCAN_TYPE_FIELDS
mf.no_field_order = True
if self.scan_type == SCAN_TYPE_TAG_WITH_ALBUM:
self.scan_type = SCAN_TYPE_TAG
self.scanned_tags = set(['artist', 'album', 'title'])
func = {
SCAN_TYPE_FILENAME: lambda f: engine.getwords(rem_file_ext(f.name)),
SCAN_TYPE_FIELDS: lambda f: engine.getfields(rem_file_ext(f.name)),
SCAN_TYPE_TAG: lambda f: [engine.getwords(unicode(getattr(f, attrname))) for attrname in SCANNABLE_TAGS if attrname in self.scanned_tags],
SCAN_TYPE_CONTENT: lambda f: [str(f.size)],
SCAN_TYPE_CONTENT_AUDIO: lambda f: [str(f.audiosize)]
}[self.scan_type]
for f in j.iter_with_progress(files, 'Read metadata of %d/%d files'):
if self.size_threshold:
f.size # pre-read, makes a smoother progress if read here (especially for bundles)
f.words = func(f)
if self.size_threshold:
files = [f for f in files if f.size >= self.size_threshold]
return mf.getmatches(files, j)
@staticmethod
def _key_func(dupe):
return (not dupe.is_ref, -dupe.size)
@staticmethod
def _tie_breaker(ref, dupe):
refname = rem_file_ext(ref.name).lower()
dupename = rem_file_ext(dupe.name).lower()
if 'copy' in refname and 'copy' not in dupename:
return True
if refname.startswith(dupename) and (refname[len(dupename):].strip().isdigit()):
return True
return len(dupe.path) > len(ref.path)
def GetDupeGroups(self, files, j=job.nulljob):
j = j.start_subjob([8, 2])
for f in [f for f in files if not hasattr(f, 'is_ref')]:
f.is_ref = False
logging.info('Getting matches')
if self.match_factory is None:
matches = self._getmatches(files, j)
else:
matches = self.match_factory.getmatches(files, j)
logging.info('Found %d matches' % len(matches))
if not self.mix_file_kind:
j.set_progress(100, 'Removing false matches')
matches = [m for m in matches if get_file_ext(m.first.name) == get_file_ext(m.second.name)]
if self.ignore_list:
j = j.start_subjob(2)
iter_matches = j.iter_with_progress(matches, 'Processed %d/%d matches against the ignore list')
matches = [m for m in iter_matches
if not self.ignore_list.AreIgnored(unicode(m.first.path), unicode(m.second.path))]
matched_files = dedupe([m.first for m in matches] + [m.second for m in matches])
if self.scan_type in (SCAN_TYPE_CONTENT, SCAN_TYPE_CONTENT_AUDIO):
md5attrname = 'md5partial' if self.scan_type == SCAN_TYPE_CONTENT_AUDIO else 'md5'
md5 = lambda f: getattr(f, md5attrname)
j = j.start_subjob(2)
for matched_file in j.iter_with_progress(matched_files, 'Analyzed %d/%d matching files'):
md5(matched_file)
j.set_progress(100, 'Removing false matches')
matches = [m for m in matches if md5(m.first) == md5(m.second)]
words_for_content = ['--'] # We compared md5. No words were involved.
for m in matches:
m.first.words = words_for_content
m.second.words = words_for_content
logging.info('Grouping matches')
groups = engine.get_groups(matches, j)
groups = [g for g in groups if any(not f.is_ref for f in g)]
logging.info('Created %d groups' % len(groups))
j.set_progress(100, 'Doing group prioritization')
for g in groups:
g.prioritize(self._key_func, self._tie_breaker)
matched_files = dedupe([m.first for m in matches] + [m.second for m in matches])
self.discarded_file_count = len(matched_files) - sum(len(g) for g in groups)
return groups
match_factory = None
match_similar_words = False
min_match_percentage = 80
mix_file_kind = True
scan_type = SCAN_TYPE_FILENAME
scanned_tags = set(['artist', 'title'])
size_threshold = 0
word_weighting = False
class ScannerME(Scanner): # Scanner for Music Edition
@staticmethod
def _key_func(dupe):
return (not dupe.is_ref, -dupe.bitrate, -dupe.size)