1
0
mirror of https://github.com/arsenetar/dupeguru.git synced 2024-11-04 23:39:02 +00:00
dupeguru/core/results.py
2012-03-15 14:28:40 -04:00

365 lines
14 KiB
Python

# Created By: Virgil Dupras
# Created On: 2006/02/23
# Copyright 2012 Hardcoded Software (http://www.hardcoded.net)
#
# This software is licensed under the "BSD" 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/bsd_license
import logging
import re
import os
import os.path as op
from xml.etree import ElementTree as ET
from jobprogress.job import nulljob
from hscommon.conflict import get_conflicted_name
from hscommon.util import flatten, nonone, FileOrPath, format_size
from hscommon.trans import tr
from . import engine
from .markable import Markable
class Results(Markable):
#---Override
def __init__(self, app):
Markable.__init__(self)
self.__groups = []
self.__group_of_duplicate = {}
self.__groups_sort_descriptor = None # This is a tuple (key, asc)
self.__dupes = None
self.__dupes_sort_descriptor = None # This is a tuple (key, asc, delta)
self.__filters = None
self.__filtered_dupes = None
self.__filtered_groups = None
self.__recalculate_stats()
self.__marked_size = 0
self.app = app
self.problems = [] # (dupe, error_msg)
self.is_modified = False
def _did_mark(self, dupe):
self.__marked_size += dupe.size
def _did_unmark(self, dupe):
self.__marked_size -= dupe.size
def _get_markable_count(self):
return self.__total_count
def _is_markable(self, dupe):
if dupe.is_ref:
return False
g = self.get_group_of_duplicate(dupe)
if not g:
return False
if dupe is g.ref:
return False
if self.__filtered_dupes and dupe not in self.__filtered_dupes:
return False
return True
def mark_all(self):
if self.__filters:
self.mark_multiple(self.__filtered_dupes)
else:
Markable.mark_all(self)
def mark_invert(self):
if self.__filters:
self.mark_toggle_multiple(self.__filtered_dupes)
else:
Markable.mark_invert(self)
def mark_none(self):
if self.__filters:
self.unmark_multiple(self.__filtered_dupes)
else:
Markable.mark_none(self)
#---Private
def __get_dupe_list(self):
if self.__dupes is None:
self.__dupes = flatten(group.dupes for group in self.groups)
if None in self.__dupes:
# This is debug logging to try to figure out #44
logging.warning("There is a None value in the Results' dupe list. dupes: %r groups: %r", self.__dupes, self.groups)
if self.__filtered_dupes:
self.__dupes = [dupe for dupe in self.__dupes if dupe in self.__filtered_dupes]
sd = self.__dupes_sort_descriptor
if sd:
self.sort_dupes(sd[0], sd[1], sd[2])
return self.__dupes
def __get_groups(self):
if self.__filtered_groups is None:
return self.__groups
else:
return self.__filtered_groups
def __get_stat_line(self):
if self.__filtered_dupes is None:
mark_count = self.mark_count
marked_size = self.__marked_size
total_count = self.__total_count
total_size = self.__total_size
else:
mark_count = len([dupe for dupe in self.__filtered_dupes if self.is_marked(dupe)])
marked_size = sum(dupe.size for dupe in self.__filtered_dupes if self.is_marked(dupe))
total_count = len([dupe for dupe in self.__filtered_dupes if self.is_markable(dupe)])
total_size = sum(dupe.size for dupe in self.__filtered_dupes if self.is_markable(dupe))
if self.mark_inverted:
marked_size = self.__total_size - marked_size
result = tr("%d / %d (%s / %s) duplicates marked.") % (
mark_count,
total_count,
format_size(marked_size, 2),
format_size(total_size, 2),
)
if self.__filters:
result += tr(" filter: %s") % ' --> '.join(self.__filters)
return result
def __recalculate_stats(self):
self.__total_size = 0
self.__total_count = 0
for group in self.groups:
markable = [dupe for dupe in group.dupes if self._is_markable(dupe)]
self.__total_count += len(markable)
self.__total_size += sum(dupe.size for dupe in markable)
def __set_groups(self, new_groups):
self.mark_none()
self.__groups = new_groups
self.__group_of_duplicate = {}
for g in self.__groups:
for dupe in g:
self.__group_of_duplicate[dupe] = g
if not hasattr(dupe, 'is_ref'):
dupe.is_ref = False
self.is_modified = bool(self.__groups)
old_filters = nonone(self.__filters, [])
self.apply_filter(None)
for filter_str in old_filters:
self.apply_filter(filter_str)
#---Public
def apply_filter(self, filter_str):
''' Applies a filter 'filter_str' to self.groups
When you apply the filter, only dupes with the filename matching 'filter_str' will be in
in the results. To cancel the filter, just call apply_filter with 'filter_str' to None,
and the results will go back to normal.
If call apply_filter on a filtered results, the filter will be applied
*on the filtered results*.
'filter_str' is a string containing a regexp to filter dupes with.
'''
if not filter_str:
self.__filtered_dupes = None
self.__filtered_groups = None
self.__filters = None
else:
if not self.__filters:
self.__filters = []
try:
filter_re = re.compile(filter_str, re.IGNORECASE)
except re.error:
return # don't apply this filter.
self.__filters.append(filter_str)
if self.__filtered_dupes is None:
self.__filtered_dupes = flatten(g[:] for g in self.groups)
self.__filtered_dupes = set(dupe for dupe in self.__filtered_dupes if filter_re.search(str(dupe.path)))
filtered_groups = set()
for dupe in self.__filtered_dupes:
filtered_groups.add(self.get_group_of_duplicate(dupe))
self.__filtered_groups = list(filtered_groups)
self.__recalculate_stats()
sd = self.__groups_sort_descriptor
if sd:
self.sort_groups(sd[0], sd[1])
self.__dupes = None
def get_group_of_duplicate(self, dupe):
try:
return self.__group_of_duplicate[dupe]
except (TypeError, KeyError):
return None
is_markable = _is_markable
def load_from_xml(self, infile, get_file, j=nulljob):
def do_match(ref_file, other_files, group):
if not other_files:
return
for other_file in other_files:
group.add_match(engine.get_match(ref_file, other_file))
do_match(other_files[0], other_files[1:], group)
self.apply_filter(None)
try:
root = ET.parse(infile).getroot()
except Exception:
return
group_elems = list(root.getiterator('group'))
groups = []
marked = set()
for group_elem in j.iter_with_progress(group_elems, every=100):
group = engine.Group()
dupes = []
for file_elem in group_elem.getiterator('file'):
path = file_elem.get('path')
words = file_elem.get('words', '')
if not path:
continue
file = get_file(path)
if file is None:
continue
file.words = words.split(',')
file.is_ref = file_elem.get('is_ref') == 'y'
dupes.append(file)
if file_elem.get('marked') == 'y':
marked.add(file)
for match_elem in group_elem.getiterator('match'):
try:
attrs = match_elem.attrib
first_file = dupes[int(attrs['first'])]
second_file = dupes[int(attrs['second'])]
percentage = int(attrs['percentage'])
group.add_match(engine.Match(first_file, second_file, percentage))
except (IndexError, KeyError, ValueError): # Covers missing attr, non-int values and indexes out of bounds
pass
if (not group.matches) and (len(dupes) >= 2):
do_match(dupes[0], dupes[1:], group)
group.prioritize(lambda x: dupes.index(x))
if len(group):
groups.append(group)
j.add_progress()
self.groups = groups
for dupe_file in marked:
self.mark(dupe_file)
self.is_modified = False
def make_ref(self, dupe):
g = self.get_group_of_duplicate(dupe)
r = g.ref
self._remove_mark_flag(dupe)
g.switch_ref(dupe);
if not r.is_ref:
self.__total_count += 1
self.__total_size += r.size
if not dupe.is_ref:
self.__total_count -= 1
self.__total_size -= dupe.size
self.__dupes = None
self.is_modified = True
def perform_on_marked(self, func, remove_from_results):
# Performs `func` on all marked dupes. If an EnvironmentError is raised during the call,
# the problematic dupe is added to self.problems.
self.problems = []
to_remove = []
marked = (dupe for dupe in self.dupes if self.is_marked(dupe))
for dupe in marked:
try:
func(dupe)
to_remove.append(dupe)
except EnvironmentError as e:
self.problems.append((dupe, str(e)))
if remove_from_results:
self.remove_duplicates(to_remove)
self.mark_none()
for dupe, _ in self.problems:
self.mark(dupe)
def remove_duplicates(self, dupes):
'''Remove 'dupes' from their respective group, and remove the group is it ends up empty.
'''
affected_groups = set()
for dupe in dupes:
group = self.get_group_of_duplicate(dupe)
if dupe not in group.dupes:
return
ref = group.ref
group.remove_dupe(dupe, False)
del self.__group_of_duplicate[dupe]
self._remove_mark_flag(dupe)
self.__total_count -= 1
self.__total_size -= dupe.size
if not group:
del self.__group_of_duplicate[ref]
self.__groups.remove(group)
if self.__filtered_groups:
self.__filtered_groups.remove(group)
else:
affected_groups.add(group)
for group in affected_groups:
group.discard_matches()
self.__dupes = None
self.is_modified = bool(self.__groups)
def save_to_xml(self, outfile):
self.apply_filter(None)
root = ET.Element('results')
# writer = XMLGenerator(outfile, 'utf-8')
for g in self.groups:
group_elem = ET.SubElement(root, 'group')
dupe2index = {}
for index, d in enumerate(g):
dupe2index[d] = index
try:
words = engine.unpack_fields(d.words)
except AttributeError:
words = ()
file_elem = ET.SubElement(group_elem, 'file')
try:
file_elem.set('path', str(d.path))
file_elem.set('words', ','.join(words))
except ValueError: # If there's an invalid character, just skip the file
file_elem.set('path', '')
file_elem.set('is_ref', ('y' if d.is_ref else 'n'))
file_elem.set('marked', ('y' if self.is_marked(d) else 'n'))
for match in g.matches:
match_elem = ET.SubElement(group_elem, 'match')
match_elem.set('first', str(dupe2index[match.first]))
match_elem.set('second', str(dupe2index[match.second]))
match_elem.set('percentage', str(int(match.percentage)))
tree = ET.ElementTree(root)
def do_write(outfile):
with FileOrPath(outfile, 'wb') as fp:
tree.write(fp, encoding='utf-8')
try:
do_write(outfile)
except IOError as e:
# If our IOError is because dest is already a directory, we want to handle that. 21 is
# the code we get on OS X and Linux, 13 is what we get on Windows.
if e.errno in {21, 13}:
p = str(outfile)
dirname, basename = op.split(p)
otherfiles = os.listdir(dirname)
newname = get_conflicted_name(otherfiles, basename)
do_write(op.join(dirname, newname))
else:
raise
self.is_modified = False
def sort_dupes(self, key, asc=True, delta=False):
if not self.__dupes:
self.__get_dupe_list()
keyfunc = lambda d: self.app._get_dupe_sort_key(d, lambda: self.get_group_of_duplicate(d), key, delta)
self.__dupes.sort(key=keyfunc, reverse=not asc)
self.__dupes_sort_descriptor = (key,asc,delta)
def sort_groups(self,key,asc=True):
keyfunc = lambda g: self.app._get_group_sort_key(g, key)
self.groups.sort(key=keyfunc, reverse=not asc)
self.__groups_sort_descriptor = (key,asc)
#---Properties
dupes = property(__get_dupe_list)
groups = property(__get_groups, __set_groups)
stat_line = property(__get_stat_line)