1
0
mirror of https://github.com/arsenetar/dupeguru.git synced 2024-11-16 12:19:03 +00:00
dupeguru/core_se/data.py

72 lines
2.2 KiB
Python

# Created By: Virgil Dupras
# Created On: 2006/03/15
# Copyright 2010 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
from hsutil.str import format_size
from core.data import (format_path, format_timestamp, format_words, format_perc,
format_dupe_count, cmp_value)
COLUMNS = [
{'attr':'name','display':'Filename'},
{'attr':'path','display':'Directory'},
{'attr':'size','display':'Size (KB)'},
{'attr':'extension','display':'Kind'},
{'attr':'ctime','display':'Creation'},
{'attr':'mtime','display':'Modification'},
{'attr':'percentage','display':'Match %'},
{'attr':'words','display':'Words Used'},
{'attr':'dupe_count','display':'Dupe Count'},
]
METADATA_TO_READ = ['size', 'ctime', 'mtime']
def GetDisplayInfo(dupe, group, delta):
size = dupe.size
ctime = dupe.ctime
mtime = dupe.mtime
m = group.get_match_of(dupe)
if m:
percentage = m.percentage
dupe_count = 0
if delta:
r = group.ref
size -= r.size
ctime -= r.ctime
mtime -= r.mtime
else:
percentage = group.percentage
dupe_count = len(group.dupes)
return [
dupe.name,
format_path(dupe.path),
format_size(size, 0, 1, False),
dupe.extension,
format_timestamp(ctime, delta and m),
format_timestamp(mtime, delta and m),
format_perc(percentage),
format_words(dupe.words) if hasattr(dupe, 'words') else '',
format_dupe_count(dupe_count)
]
def GetDupeSortKey(dupe, get_group, key, delta):
if key == 6:
m = get_group().get_match_of(dupe)
return m.percentage
if key == 8:
return 0
r = cmp_value(getattr(dupe, COLUMNS[key]['attr'], ''))
if delta and (key in (2, 4, 5)):
r -= cmp_value(getattr(get_group().ref, COLUMNS[key]['attr'], ''))
return r
def GetGroupSortKey(group, key):
if key == 6:
return group.percentage
if key == 8:
return len(group)
return cmp_value(getattr(group.ref, COLUMNS[key]['attr'], ''))