mirror of
https://github.com/arsenetar/dupeguru.git
synced 2024-11-16 12:19:03 +00:00
6d5ae99509
--HG-- extra : convert_revision : svn%3Ac306627e-7827-47d3-bdf0-9a457c9553a1/trunk%40116
837 lines
30 KiB
Python
837 lines
30 KiB
Python
# Created By: Virgil Dupras
|
|
# Created On: 2006/01/29
|
|
# $Id$
|
|
# 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 sys
|
|
|
|
from nose.tools import eq_
|
|
|
|
from hsutil import job
|
|
from hsutil.decorators import log_calls
|
|
from hsutil.testcase import TestCase
|
|
|
|
from .. import engine
|
|
from ..engine import *
|
|
|
|
class NamedObject(object):
|
|
def __init__(self, name="foobar", with_words=False):
|
|
self.name = name
|
|
if with_words:
|
|
self.words = getwords(name)
|
|
|
|
|
|
def get_match_triangle():
|
|
o1 = NamedObject(with_words=True)
|
|
o2 = NamedObject(with_words=True)
|
|
o3 = NamedObject(with_words=True)
|
|
m1 = get_match(o1,o2)
|
|
m2 = get_match(o1,o3)
|
|
m3 = get_match(o2,o3)
|
|
return [m1, m2, m3]
|
|
|
|
def get_test_group():
|
|
m1, m2, m3 = get_match_triangle()
|
|
result = Group()
|
|
result.add_match(m1)
|
|
result.add_match(m2)
|
|
result.add_match(m3)
|
|
return result
|
|
|
|
class TCgetwords(TestCase):
|
|
def test_spaces(self):
|
|
self.assertEqual(['a', 'b', 'c', 'd'], getwords("a b c d"))
|
|
self.assertEqual(['a', 'b', 'c', 'd'], getwords(" a b c d "))
|
|
|
|
def test_splitter_chars(self):
|
|
self.assertEqual(
|
|
[chr(i) for i in xrange(ord('a'),ord('z')+1)],
|
|
getwords("a-b_c&d+e(f)g;h\\i[j]k{l}m:n.o,p<q>r/s?t~u!v@w#x$y*z")
|
|
)
|
|
|
|
def test_joiner_chars(self):
|
|
self.assertEqual(["aec"], getwords(u"a'e\u0301c"))
|
|
|
|
def test_empty(self):
|
|
self.assertEqual([], getwords(''))
|
|
|
|
def test_returns_lowercase(self):
|
|
self.assertEqual(['foo', 'bar'], getwords('FOO BAR'))
|
|
|
|
def test_decompose_unicode(self):
|
|
self.assertEqual(getwords(u'foo\xe9bar'), ['fooebar'])
|
|
|
|
|
|
class TCgetfields(TestCase):
|
|
def test_simple(self):
|
|
self.assertEqual([['a', 'b'], ['c', 'd', 'e']], getfields('a b - c d e'))
|
|
|
|
def test_empty(self):
|
|
self.assertEqual([], getfields(''))
|
|
|
|
def test_cleans_empty_fields(self):
|
|
expected = [['a', 'bc', 'def']]
|
|
actual = getfields(' - a bc def')
|
|
self.assertEqual(expected, actual)
|
|
expected = [['bc', 'def']]
|
|
|
|
|
|
class TCunpack_fields(TestCase):
|
|
def test_with_fields(self):
|
|
expected = ['a', 'b', 'c', 'd', 'e', 'f']
|
|
actual = unpack_fields([['a'], ['b', 'c'], ['d', 'e', 'f']])
|
|
self.assertEqual(expected, actual)
|
|
|
|
def test_without_fields(self):
|
|
expected = ['a', 'b', 'c', 'd', 'e', 'f']
|
|
actual = unpack_fields(['a', 'b', 'c', 'd', 'e', 'f'])
|
|
self.assertEqual(expected, actual)
|
|
|
|
def test_empty(self):
|
|
self.assertEqual([], unpack_fields([]))
|
|
|
|
|
|
class TCWordCompare(TestCase):
|
|
def test_list(self):
|
|
self.assertEqual(100, compare(['a', 'b', 'c', 'd'],['a', 'b', 'c', 'd']))
|
|
self.assertEqual(86, compare(['a', 'b', 'c', 'd'],['a', 'b', 'c']))
|
|
|
|
def test_unordered(self):
|
|
#Sometimes, users don't want fuzzy matching too much When they set the slider
|
|
#to 100, they don't expect a filename with the same words, but not the same order, to match.
|
|
#Thus, we want to return 99 in that case.
|
|
self.assertEqual(99, compare(['a', 'b', 'c', 'd'], ['d', 'b', 'c', 'a']))
|
|
|
|
def test_word_occurs_twice(self):
|
|
#if a word occurs twice in first, but once in second, we want the word to be only counted once
|
|
self.assertEqual(89, compare(['a', 'b', 'c', 'd', 'a'], ['d', 'b', 'c', 'a']))
|
|
|
|
def test_uses_copy_of_lists(self):
|
|
first = ['foo', 'bar']
|
|
second = ['bar', 'bleh']
|
|
compare(first, second)
|
|
self.assertEqual(['foo', 'bar'], first)
|
|
self.assertEqual(['bar', 'bleh'], second)
|
|
|
|
def test_word_weight(self):
|
|
self.assertEqual(int((6.0 / 13.0) * 100), compare(['foo', 'bar'], ['bar', 'bleh'], (WEIGHT_WORDS, )))
|
|
|
|
def test_similar_words(self):
|
|
self.assertEqual(100, compare(['the', 'white', 'stripes'],['the', 'whites', 'stripe'], (MATCH_SIMILAR_WORDS, )))
|
|
|
|
def test_empty(self):
|
|
self.assertEqual(0, compare([], []))
|
|
|
|
def test_with_fields(self):
|
|
self.assertEqual(67, compare([['a', 'b'], ['c', 'd', 'e']], [['a', 'b'], ['c', 'd', 'f']]))
|
|
|
|
def test_propagate_flags_with_fields(self):
|
|
def mock_compare(first, second, flags):
|
|
self.assertEqual((0, 1, 2, 3, 5), flags)
|
|
|
|
self.mock(engine, 'compare_fields', mock_compare)
|
|
compare([['a']], [['a']], (0, 1, 2, 3, 5))
|
|
|
|
|
|
class TCWordCompareWithFields(TestCase):
|
|
def test_simple(self):
|
|
self.assertEqual(67, compare_fields([['a', 'b'], ['c', 'd', 'e']], [['a', 'b'], ['c', 'd', 'f']]))
|
|
|
|
def test_empty(self):
|
|
self.assertEqual(0, compare_fields([], []))
|
|
|
|
def test_different_length(self):
|
|
self.assertEqual(0, compare_fields([['a'], ['b']], [['a'], ['b'], ['c']]))
|
|
|
|
def test_propagates_flags(self):
|
|
def mock_compare(first, second, flags):
|
|
self.assertEqual((0, 1, 2, 3, 5), flags)
|
|
|
|
self.mock(engine, 'compare_fields', mock_compare)
|
|
compare_fields([['a']], [['a']],(0, 1, 2, 3, 5))
|
|
|
|
def test_order(self):
|
|
first = [['a', 'b'], ['c', 'd', 'e']]
|
|
second = [['c', 'd', 'f'], ['a', 'b']]
|
|
self.assertEqual(0, compare_fields(first, second))
|
|
|
|
def test_no_order(self):
|
|
first = [['a','b'],['c','d','e']]
|
|
second = [['c','d','f'],['a','b']]
|
|
self.assertEqual(67, compare_fields(first, second, (NO_FIELD_ORDER, )))
|
|
first = [['a','b'],['a','b']] #a field can only be matched once.
|
|
second = [['c','d','f'],['a','b']]
|
|
self.assertEqual(0, compare_fields(first, second, (NO_FIELD_ORDER, )))
|
|
first = [['a','b'],['a','b','c']]
|
|
second = [['c','d','f'],['a','b']]
|
|
self.assertEqual(33, compare_fields(first, second, (NO_FIELD_ORDER, )))
|
|
|
|
def test_compare_fields_without_order_doesnt_alter_fields(self):
|
|
#The NO_ORDER comp type altered the fields!
|
|
first = [['a','b'],['c','d','e']]
|
|
second = [['c','d','f'],['a','b']]
|
|
self.assertEqual(67, compare_fields(first, second, (NO_FIELD_ORDER, )))
|
|
self.assertEqual([['a','b'],['c','d','e']],first)
|
|
self.assertEqual([['c','d','f'],['a','b']],second)
|
|
|
|
|
|
class TCbuild_word_dict(TestCase):
|
|
def test_with_standard_words(self):
|
|
l = [NamedObject('foo bar',True)]
|
|
l.append(NamedObject('bar baz',True))
|
|
l.append(NamedObject('baz bleh foo',True))
|
|
d = build_word_dict(l)
|
|
self.assertEqual(4,len(d))
|
|
self.assertEqual(2,len(d['foo']))
|
|
self.assert_(l[0] in d['foo'])
|
|
self.assert_(l[2] in d['foo'])
|
|
self.assertEqual(2,len(d['bar']))
|
|
self.assert_(l[0] in d['bar'])
|
|
self.assert_(l[1] in d['bar'])
|
|
self.assertEqual(2,len(d['baz']))
|
|
self.assert_(l[1] in d['baz'])
|
|
self.assert_(l[2] in d['baz'])
|
|
self.assertEqual(1,len(d['bleh']))
|
|
self.assert_(l[2] in d['bleh'])
|
|
|
|
def test_unpack_fields(self):
|
|
o = NamedObject('')
|
|
o.words = [['foo','bar'],['baz']]
|
|
d = build_word_dict([o])
|
|
self.assertEqual(3,len(d))
|
|
self.assertEqual(1,len(d['foo']))
|
|
|
|
def test_words_are_unaltered(self):
|
|
o = NamedObject('')
|
|
o.words = [['foo','bar'],['baz']]
|
|
d = build_word_dict([o])
|
|
self.assertEqual([['foo','bar'],['baz']],o.words)
|
|
|
|
def test_object_instances_can_only_be_once_in_words_object_list(self):
|
|
o = NamedObject('foo foo',True)
|
|
d = build_word_dict([o])
|
|
self.assertEqual(1,len(d['foo']))
|
|
|
|
def test_job(self):
|
|
def do_progress(p,d=''):
|
|
self.log.append(p)
|
|
return True
|
|
|
|
j = job.Job(1,do_progress)
|
|
self.log = []
|
|
s = "foo bar"
|
|
build_word_dict([NamedObject(s, True), NamedObject(s, True), NamedObject(s, True)], j)
|
|
self.assertEqual(0,self.log[0])
|
|
self.assertEqual(33,self.log[1])
|
|
self.assertEqual(66,self.log[2])
|
|
self.assertEqual(100,self.log[3])
|
|
|
|
|
|
class TCmerge_similar_words(TestCase):
|
|
def test_some_similar_words(self):
|
|
d = {
|
|
'foobar':set([1]),
|
|
'foobar1':set([2]),
|
|
'foobar2':set([3]),
|
|
}
|
|
merge_similar_words(d)
|
|
self.assertEqual(1,len(d))
|
|
self.assertEqual(3,len(d['foobar']))
|
|
|
|
|
|
|
|
class TCreduce_common_words(TestCase):
|
|
def test_typical(self):
|
|
d = {
|
|
'foo': set([NamedObject('foo bar',True) for i in range(50)]),
|
|
'bar': set([NamedObject('foo bar',True) for i in range(49)])
|
|
}
|
|
reduce_common_words(d, 50)
|
|
self.assert_('foo' not in d)
|
|
self.assertEqual(49,len(d['bar']))
|
|
|
|
def test_dont_remove_objects_with_only_common_words(self):
|
|
d = {
|
|
'common': set([NamedObject("common uncommon",True) for i in range(50)] + [NamedObject("common",True)]),
|
|
'uncommon': set([NamedObject("common uncommon",True)])
|
|
}
|
|
reduce_common_words(d, 50)
|
|
self.assertEqual(1,len(d['common']))
|
|
self.assertEqual(1,len(d['uncommon']))
|
|
|
|
def test_values_still_are_set_instances(self):
|
|
d = {
|
|
'common': set([NamedObject("common uncommon",True) for i in range(50)] + [NamedObject("common",True)]),
|
|
'uncommon': set([NamedObject("common uncommon",True)])
|
|
}
|
|
reduce_common_words(d, 50)
|
|
self.assert_(isinstance(d['common'],set))
|
|
self.assert_(isinstance(d['uncommon'],set))
|
|
|
|
def test_dont_raise_KeyError_when_a_word_has_been_removed(self):
|
|
#If a word has been removed by the reduce, an object in a subsequent common word that
|
|
#contains the word that has been removed would cause a KeyError.
|
|
d = {
|
|
'foo': set([NamedObject('foo bar baz',True) for i in range(50)]),
|
|
'bar': set([NamedObject('foo bar baz',True) for i in range(50)]),
|
|
'baz': set([NamedObject('foo bar baz',True) for i in range(49)])
|
|
}
|
|
try:
|
|
reduce_common_words(d, 50)
|
|
except KeyError:
|
|
self.fail()
|
|
|
|
def test_unpack_fields(self):
|
|
#object.words may be fields.
|
|
def create_it():
|
|
o = NamedObject('')
|
|
o.words = [['foo','bar'],['baz']]
|
|
return o
|
|
|
|
d = {
|
|
'foo': set([create_it() for i in range(50)])
|
|
}
|
|
try:
|
|
reduce_common_words(d, 50)
|
|
except TypeError:
|
|
self.fail("must support fields.")
|
|
|
|
def test_consider_a_reduced_common_word_common_even_after_reduction(self):
|
|
#There was a bug in the code that causeda word that has already been reduced not to
|
|
#be counted as a common word for subsequent words. For example, if 'foo' is processed
|
|
#as a common word, keeping a "foo bar" file in it, and the 'bar' is processed, "foo bar"
|
|
#would not stay in 'bar' because 'foo' is not a common word anymore.
|
|
only_common = NamedObject('foo bar',True)
|
|
d = {
|
|
'foo': set([NamedObject('foo bar baz',True) for i in range(49)] + [only_common]),
|
|
'bar': set([NamedObject('foo bar baz',True) for i in range(49)] + [only_common]),
|
|
'baz': set([NamedObject('foo bar baz',True) for i in range(49)])
|
|
}
|
|
reduce_common_words(d, 50)
|
|
self.assertEqual(1,len(d['foo']))
|
|
self.assertEqual(1,len(d['bar']))
|
|
self.assertEqual(49,len(d['baz']))
|
|
|
|
|
|
class TCget_match(TestCase):
|
|
def test_simple(self):
|
|
o1 = NamedObject("foo bar",True)
|
|
o2 = NamedObject("bar bleh",True)
|
|
m = get_match(o1,o2)
|
|
self.assertEqual(50,m.percentage)
|
|
self.assertEqual(['foo','bar'],m.first.words)
|
|
self.assertEqual(['bar','bleh'],m.second.words)
|
|
self.assert_(m.first is o1)
|
|
self.assert_(m.second is o2)
|
|
|
|
def test_in(self):
|
|
o1 = NamedObject("foo",True)
|
|
o2 = NamedObject("bar",True)
|
|
m = get_match(o1,o2)
|
|
self.assert_(o1 in m)
|
|
self.assert_(o2 in m)
|
|
self.assert_(object() not in m)
|
|
|
|
def test_word_weight(self):
|
|
self.assertEqual(int((6.0 / 13.0) * 100),get_match(NamedObject("foo bar",True),NamedObject("bar bleh",True),(WEIGHT_WORDS,)).percentage)
|
|
|
|
|
|
class TCMatchFactory(TestCase):
|
|
def test_empty(self):
|
|
self.assertEqual([],MatchFactory().getmatches([]))
|
|
|
|
def test_defaults(self):
|
|
mf = MatchFactory()
|
|
self.assertEqual(50,mf.common_word_threshold)
|
|
self.assertEqual(False,mf.weight_words)
|
|
self.assertEqual(False,mf.match_similar_words)
|
|
self.assertEqual(False,mf.no_field_order)
|
|
self.assertEqual(0,mf.min_match_percentage)
|
|
|
|
def test_simple(self):
|
|
l = [NamedObject("foo bar"),NamedObject("bar bleh"),NamedObject("a b c foo")]
|
|
r = MatchFactory().getmatches(l)
|
|
self.assertEqual(2,len(r))
|
|
seek = [m for m in r if m.percentage == 50] #"foo bar" and "bar bleh"
|
|
m = seek[0]
|
|
self.assertEqual(['foo','bar'],m.first.words)
|
|
self.assertEqual(['bar','bleh'],m.second.words)
|
|
seek = [m for m in r if m.percentage == 33] #"foo bar" and "a b c foo"
|
|
m = seek[0]
|
|
self.assertEqual(['foo','bar'],m.first.words)
|
|
self.assertEqual(['a','b','c','foo'],m.second.words)
|
|
|
|
def test_null_and_unrelated_objects(self):
|
|
l = [NamedObject("foo bar"),NamedObject("bar bleh"),NamedObject(""),NamedObject("unrelated object")]
|
|
r = MatchFactory().getmatches(l)
|
|
self.assertEqual(1,len(r))
|
|
m = r[0]
|
|
self.assertEqual(50,m.percentage)
|
|
self.assertEqual(['foo','bar'],m.first.words)
|
|
self.assertEqual(['bar','bleh'],m.second.words)
|
|
|
|
def test_twice_the_same_word(self):
|
|
l = [NamedObject("foo foo bar"),NamedObject("bar bleh")]
|
|
r = MatchFactory().getmatches(l)
|
|
self.assertEqual(1,len(r))
|
|
|
|
def test_twice_the_same_word_when_preworded(self):
|
|
l = [NamedObject("foo foo bar",True),NamedObject("bar bleh",True)]
|
|
r = MatchFactory().getmatches(l)
|
|
self.assertEqual(1,len(r))
|
|
|
|
def test_two_words_match(self):
|
|
l = [NamedObject("foo bar"),NamedObject("foo bar bleh")]
|
|
r = MatchFactory().getmatches(l)
|
|
self.assertEqual(1,len(r))
|
|
|
|
def test_match_files_with_only_common_words(self):
|
|
#If a word occurs more than 50 times, it is excluded from the matching process
|
|
#The problem with the common_word_threshold is that the files containing only common
|
|
#words will never be matched together. We *should* match them.
|
|
mf = MatchFactory()
|
|
mf.common_word_threshold = 50
|
|
l = [NamedObject("foo") for i in range(50)]
|
|
r = mf.getmatches(l)
|
|
self.assertEqual(1225,len(r))
|
|
|
|
def test_use_words_already_there_if_there(self):
|
|
o1 = NamedObject('foo')
|
|
o2 = NamedObject('bar')
|
|
o2.words = ['foo']
|
|
self.assertEqual(1,len(MatchFactory().getmatches([o1,o2])))
|
|
|
|
def test_job(self):
|
|
def do_progress(p,d=''):
|
|
self.log.append(p)
|
|
return True
|
|
|
|
j = job.Job(1,do_progress)
|
|
self.log = []
|
|
s = "foo bar"
|
|
MatchFactory().getmatches([NamedObject(s),NamedObject(s),NamedObject(s)],j)
|
|
self.assert_(len(self.log) > 2)
|
|
self.assertEqual(0,self.log[0])
|
|
self.assertEqual(100,self.log[-1])
|
|
|
|
def test_weight_words(self):
|
|
mf = MatchFactory()
|
|
mf.weight_words = True
|
|
l = [NamedObject("foo bar"),NamedObject("bar bleh")]
|
|
m = mf.getmatches(l)[0]
|
|
self.assertEqual(int((6.0 / 13.0) * 100),m.percentage)
|
|
|
|
def test_similar_word(self):
|
|
mf = MatchFactory()
|
|
mf.match_similar_words = True
|
|
l = [NamedObject("foobar"),NamedObject("foobars")]
|
|
self.assertEqual(1,len(mf.getmatches(l)))
|
|
self.assertEqual(100,mf.getmatches(l)[0].percentage)
|
|
l = [NamedObject("foobar"),NamedObject("foo")]
|
|
self.assertEqual(0,len(mf.getmatches(l))) #too far
|
|
l = [NamedObject("bizkit"),NamedObject("bizket")]
|
|
self.assertEqual(1,len(mf.getmatches(l)))
|
|
l = [NamedObject("foobar"),NamedObject("foosbar")]
|
|
self.assertEqual(1,len(mf.getmatches(l)))
|
|
|
|
def test_single_object_with_similar_words(self):
|
|
mf = MatchFactory()
|
|
mf.match_similar_words = True
|
|
l = [NamedObject("foo foos")]
|
|
self.assertEqual(0,len(mf.getmatches(l)))
|
|
|
|
def test_double_words_get_counted_only_once(self):
|
|
mf = MatchFactory()
|
|
l = [NamedObject("foo bar foo bleh"),NamedObject("foo bar bleh bar")]
|
|
m = mf.getmatches(l)[0]
|
|
self.assertEqual(75,m.percentage)
|
|
|
|
def test_with_fields(self):
|
|
mf = MatchFactory()
|
|
o1 = NamedObject("foo bar - foo bleh")
|
|
o2 = NamedObject("foo bar - bleh bar")
|
|
o1.words = getfields(o1.name)
|
|
o2.words = getfields(o2.name)
|
|
m = mf.getmatches([o1, o2])[0]
|
|
self.assertEqual(50, m.percentage)
|
|
|
|
def test_with_fields_no_order(self):
|
|
mf = MatchFactory()
|
|
mf.no_field_order = True
|
|
o1 = NamedObject("foo bar - foo bleh")
|
|
o2 = NamedObject("bleh bang - foo bar")
|
|
o1.words = getfields(o1.name)
|
|
o2.words = getfields(o2.name)
|
|
m = mf.getmatches([o1, o2])[0]
|
|
self.assertEqual(50 ,m.percentage)
|
|
|
|
def test_only_match_similar_when_the_option_is_set(self):
|
|
mf = MatchFactory()
|
|
mf.match_similar_words = False
|
|
l = [NamedObject("foobar"),NamedObject("foobars")]
|
|
self.assertEqual(0,len(mf.getmatches(l)))
|
|
|
|
def test_dont_recurse_do_match(self):
|
|
# with nosetests, the stack is increased. The number has to be high enough not to be failing falsely
|
|
sys.setrecursionlimit(100)
|
|
mf = MatchFactory()
|
|
files = [NamedObject('foo bar') for i in range(101)]
|
|
try:
|
|
mf.getmatches(files)
|
|
except RuntimeError:
|
|
self.fail()
|
|
finally:
|
|
sys.setrecursionlimit(1000)
|
|
|
|
def test_min_match_percentage(self):
|
|
l = [NamedObject("foo bar"),NamedObject("bar bleh"),NamedObject("a b c foo")]
|
|
mf = MatchFactory()
|
|
mf.min_match_percentage = 50
|
|
r = mf.getmatches(l)
|
|
self.assertEqual(1,len(r)) #Only "foo bar" / "bar bleh" should match
|
|
|
|
def test_limit(self):
|
|
l = [NamedObject(),NamedObject(),NamedObject()]
|
|
mf = MatchFactory()
|
|
mf.limit = 2
|
|
r = mf.getmatches(l)
|
|
self.assertEqual(2,len(r))
|
|
|
|
def test_MemoryError(self):
|
|
@log_calls
|
|
def mocked_match(first, second, flags):
|
|
if len(mocked_match.calls) > 42:
|
|
raise MemoryError()
|
|
return Match(first, second, 0)
|
|
|
|
objects = [NamedObject() for i in range(10)] # results in 45 matches
|
|
self.mock(engine, 'get_match', mocked_match)
|
|
mf = MatchFactory()
|
|
try:
|
|
r = mf.getmatches(objects)
|
|
except MemoryError:
|
|
self.fail('MemorryError must be handled')
|
|
self.assertEqual(42, len(r))
|
|
|
|
|
|
class TCGroup(TestCase):
|
|
def test_empy(self):
|
|
g = Group()
|
|
self.assertEqual(None,g.ref)
|
|
self.assertEqual([],g.dupes)
|
|
self.assertEqual(0,len(g.matches))
|
|
|
|
def test_add_match(self):
|
|
g = Group()
|
|
m = get_match(NamedObject("foo",True),NamedObject("bar",True))
|
|
g.add_match(m)
|
|
self.assert_(g.ref is m.first)
|
|
self.assertEqual([m.second],g.dupes)
|
|
self.assertEqual(1,len(g.matches))
|
|
self.assert_(m in g.matches)
|
|
|
|
def test_multiple_add_match(self):
|
|
g = Group()
|
|
o1 = NamedObject("a",True)
|
|
o2 = NamedObject("b",True)
|
|
o3 = NamedObject("c",True)
|
|
o4 = NamedObject("d",True)
|
|
g.add_match(get_match(o1,o2))
|
|
self.assert_(g.ref is o1)
|
|
self.assertEqual([o2],g.dupes)
|
|
self.assertEqual(1,len(g.matches))
|
|
g.add_match(get_match(o1,o3))
|
|
self.assertEqual([o2],g.dupes)
|
|
self.assertEqual(2,len(g.matches))
|
|
g.add_match(get_match(o2,o3))
|
|
self.assertEqual([o2,o3],g.dupes)
|
|
self.assertEqual(3,len(g.matches))
|
|
g.add_match(get_match(o1,o4))
|
|
self.assertEqual([o2,o3],g.dupes)
|
|
self.assertEqual(4,len(g.matches))
|
|
g.add_match(get_match(o2,o4))
|
|
self.assertEqual([o2,o3],g.dupes)
|
|
self.assertEqual(5,len(g.matches))
|
|
g.add_match(get_match(o3,o4))
|
|
self.assertEqual([o2,o3,o4],g.dupes)
|
|
self.assertEqual(6,len(g.matches))
|
|
|
|
def test_len(self):
|
|
g = Group()
|
|
self.assertEqual(0,len(g))
|
|
g.add_match(get_match(NamedObject("foo",True),NamedObject("bar",True)))
|
|
self.assertEqual(2,len(g))
|
|
|
|
def test_add_same_match_twice(self):
|
|
g = Group()
|
|
m = get_match(NamedObject("foo",True),NamedObject("foo",True))
|
|
g.add_match(m)
|
|
self.assertEqual(2,len(g))
|
|
self.assertEqual(1,len(g.matches))
|
|
g.add_match(m)
|
|
self.assertEqual(2,len(g))
|
|
self.assertEqual(1,len(g.matches))
|
|
|
|
def test_in(self):
|
|
g = Group()
|
|
o1 = NamedObject("foo",True)
|
|
o2 = NamedObject("bar",True)
|
|
self.assert_(o1 not in g)
|
|
g.add_match(get_match(o1,o2))
|
|
self.assert_(o1 in g)
|
|
self.assert_(o2 in g)
|
|
|
|
def test_remove(self):
|
|
g = Group()
|
|
o1 = NamedObject("foo",True)
|
|
o2 = NamedObject("bar",True)
|
|
o3 = NamedObject("bleh",True)
|
|
g.add_match(get_match(o1,o2))
|
|
g.add_match(get_match(o1,o3))
|
|
g.add_match(get_match(o2,o3))
|
|
self.assertEqual(3,len(g.matches))
|
|
self.assertEqual(3,len(g))
|
|
g.remove_dupe(o3)
|
|
self.assertEqual(1,len(g.matches))
|
|
self.assertEqual(2,len(g))
|
|
g.remove_dupe(o1)
|
|
self.assertEqual(0,len(g.matches))
|
|
self.assertEqual(0,len(g))
|
|
|
|
def test_remove_with_ref_dupes(self):
|
|
g = Group()
|
|
o1 = NamedObject("foo",True)
|
|
o2 = NamedObject("bar",True)
|
|
o3 = NamedObject("bleh",True)
|
|
g.add_match(get_match(o1,o2))
|
|
g.add_match(get_match(o1,o3))
|
|
g.add_match(get_match(o2,o3))
|
|
o1.is_ref = True
|
|
o2.is_ref = True
|
|
g.remove_dupe(o3)
|
|
self.assertEqual(0,len(g))
|
|
|
|
def test_switch_ref(self):
|
|
o1 = NamedObject(with_words=True)
|
|
o2 = NamedObject(with_words=True)
|
|
g = Group()
|
|
g.add_match(get_match(o1,o2))
|
|
self.assert_(o1 is g.ref)
|
|
g.switch_ref(o2)
|
|
self.assert_(o2 is g.ref)
|
|
self.assertEqual([o1],g.dupes)
|
|
g.switch_ref(o2)
|
|
self.assert_(o2 is g.ref)
|
|
g.switch_ref(NamedObject('',True))
|
|
self.assert_(o2 is g.ref)
|
|
|
|
def test_get_match_of(self):
|
|
g = Group()
|
|
for m in get_match_triangle():
|
|
g.add_match(m)
|
|
o = g.dupes[0]
|
|
m = g.get_match_of(o)
|
|
self.assert_(g.ref in m)
|
|
self.assert_(o in m)
|
|
self.assert_(g.get_match_of(NamedObject('',True)) is None)
|
|
self.assert_(g.get_match_of(g.ref) is None)
|
|
|
|
def test_percentage(self):
|
|
#percentage should return the avg percentage in relation to the ref
|
|
m1,m2,m3 = get_match_triangle()
|
|
m1 = Match(m1[0], m1[1], 100)
|
|
m2 = Match(m2[0], m2[1], 50)
|
|
m3 = Match(m3[0], m3[1], 33)
|
|
g = Group()
|
|
g.add_match(m1)
|
|
g.add_match(m2)
|
|
g.add_match(m3)
|
|
self.assertEqual(75,g.percentage)
|
|
g.switch_ref(g.dupes[0])
|
|
self.assertEqual(66,g.percentage)
|
|
g.remove_dupe(g.dupes[0])
|
|
self.assertEqual(33,g.percentage)
|
|
g.add_match(m1)
|
|
g.add_match(m2)
|
|
self.assertEqual(66,g.percentage)
|
|
|
|
def test_percentage_on_empty_group(self):
|
|
g = Group()
|
|
self.assertEqual(0,g.percentage)
|
|
|
|
def test_prioritize(self):
|
|
m1,m2,m3 = get_match_triangle()
|
|
o1 = m1.first
|
|
o2 = m1.second
|
|
o3 = m2.second
|
|
o1.name = 'c'
|
|
o2.name = 'b'
|
|
o3.name = 'a'
|
|
g = Group()
|
|
g.add_match(m1)
|
|
g.add_match(m2)
|
|
g.add_match(m3)
|
|
self.assert_(o1 is g.ref)
|
|
g.prioritize(lambda x:x.name)
|
|
self.assert_(o3 is g.ref)
|
|
|
|
def test_prioritize_with_tie_breaker(self):
|
|
# if the ref has the same key as one or more of the dupe, run the tie_breaker func among them
|
|
g = get_test_group()
|
|
o1, o2, o3 = g.ordered
|
|
tie_breaker = lambda ref, dupe: dupe is o3
|
|
g.prioritize(lambda x:0, tie_breaker)
|
|
self.assertTrue(g.ref is o3)
|
|
|
|
def test_prioritize_with_tie_breaker_runs_on_all_dupes(self):
|
|
# Even if a dupe is chosen to switch with ref with a tie breaker, we still run the tie breaker
|
|
# with other dupes and the newly chosen ref
|
|
g = get_test_group()
|
|
o1, o2, o3 = g.ordered
|
|
o1.foo = 1
|
|
o2.foo = 2
|
|
o3.foo = 3
|
|
tie_breaker = lambda ref, dupe: dupe.foo > ref.foo
|
|
g.prioritize(lambda x:0, tie_breaker)
|
|
self.assertTrue(g.ref is o3)
|
|
|
|
def test_prioritize_with_tie_breaker_runs_only_on_tie_dupes(self):
|
|
# The tie breaker only runs on dupes that had the same value for the key_func
|
|
g = get_test_group()
|
|
o1, o2, o3 = g.ordered
|
|
o1.foo = 2
|
|
o2.foo = 2
|
|
o3.foo = 1
|
|
o1.bar = 1
|
|
o2.bar = 2
|
|
o3.bar = 3
|
|
key_func = lambda x: -x.foo
|
|
tie_breaker = lambda ref, dupe: dupe.bar > ref.bar
|
|
g.prioritize(key_func, tie_breaker)
|
|
self.assertTrue(g.ref is o2)
|
|
|
|
def test_list_like(self):
|
|
g = Group()
|
|
o1,o2 = (NamedObject("foo",True),NamedObject("bar",True))
|
|
g.add_match(get_match(o1,o2))
|
|
self.assert_(g[0] is o1)
|
|
self.assert_(g[1] is o2)
|
|
|
|
def test_discard_matches(self):
|
|
g = Group()
|
|
o1,o2,o3 = (NamedObject("foo",True),NamedObject("bar",True),NamedObject("baz",True))
|
|
g.add_match(get_match(o1,o2))
|
|
g.add_match(get_match(o1,o3))
|
|
g.discard_matches()
|
|
self.assertEqual(1,len(g.matches))
|
|
self.assertEqual(0,len(g.candidates))
|
|
|
|
|
|
class TCget_groups(TestCase):
|
|
def test_empty(self):
|
|
r = get_groups([])
|
|
self.assertEqual([],r)
|
|
|
|
def test_simple(self):
|
|
l = [NamedObject("foo bar"),NamedObject("bar bleh")]
|
|
matches = MatchFactory().getmatches(l)
|
|
m = matches[0]
|
|
r = get_groups(matches)
|
|
self.assertEqual(1,len(r))
|
|
g = r[0]
|
|
self.assert_(g.ref is m.first)
|
|
self.assertEqual([m.second],g.dupes)
|
|
|
|
def test_group_with_multiple_matches(self):
|
|
#This results in 3 matches
|
|
l = [NamedObject("foo"),NamedObject("foo"),NamedObject("foo")]
|
|
matches = MatchFactory().getmatches(l)
|
|
r = get_groups(matches)
|
|
self.assertEqual(1,len(r))
|
|
g = r[0]
|
|
self.assertEqual(3,len(g))
|
|
|
|
def test_must_choose_a_group(self):
|
|
l = [NamedObject("a b"),NamedObject("a b"),NamedObject("b c"),NamedObject("c d"),NamedObject("c d")]
|
|
#There will be 2 groups here: group "a b" and group "c d"
|
|
#"b c" can go either of them, but not both.
|
|
matches = MatchFactory().getmatches(l)
|
|
r = get_groups(matches)
|
|
self.assertEqual(2,len(r))
|
|
self.assertEqual(5,len(r[0])+len(r[1]))
|
|
|
|
def test_should_all_go_in_the_same_group(self):
|
|
l = [NamedObject("a b"),NamedObject("a b"),NamedObject("a b"),NamedObject("a b")]
|
|
#There will be 2 groups here: group "a b" and group "c d"
|
|
#"b c" can fit in both, but it must be in only one of them
|
|
matches = MatchFactory().getmatches(l)
|
|
r = get_groups(matches)
|
|
self.assertEqual(1,len(r))
|
|
|
|
def test_give_priority_to_matches_with_higher_percentage(self):
|
|
o1 = NamedObject(with_words=True)
|
|
o2 = NamedObject(with_words=True)
|
|
o3 = NamedObject(with_words=True)
|
|
m1 = Match(o1, o2, 1)
|
|
m2 = Match(o2, o3, 2)
|
|
r = get_groups([m1,m2])
|
|
self.assertEqual(1,len(r))
|
|
g = r[0]
|
|
self.assertEqual(2,len(g))
|
|
self.assert_(o1 not in g)
|
|
self.assert_(o2 in g)
|
|
self.assert_(o3 in g)
|
|
|
|
def test_four_sized_group(self):
|
|
l = [NamedObject("foobar") for i in xrange(4)]
|
|
m = MatchFactory().getmatches(l)
|
|
r = get_groups(m)
|
|
self.assertEqual(1,len(r))
|
|
self.assertEqual(4,len(r[0]))
|
|
|
|
def test_referenced_by_ref2(self):
|
|
o1 = NamedObject(with_words=True)
|
|
o2 = NamedObject(with_words=True)
|
|
o3 = NamedObject(with_words=True)
|
|
m1 = get_match(o1,o2)
|
|
m2 = get_match(o3,o1)
|
|
m3 = get_match(o3,o2)
|
|
r = get_groups([m1,m2,m3])
|
|
self.assertEqual(3,len(r[0]))
|
|
|
|
def test_job(self):
|
|
def do_progress(p,d=''):
|
|
self.log.append(p)
|
|
return True
|
|
|
|
self.log = []
|
|
j = job.Job(1,do_progress)
|
|
m1,m2,m3 = get_match_triangle()
|
|
#101%: To make sure it is processed first so the job test works correctly
|
|
m4 = Match(NamedObject('a',True), NamedObject('a',True), 101)
|
|
get_groups([m1,m2,m3,m4],j)
|
|
self.assertEqual(0,self.log[0])
|
|
self.assertEqual(100,self.log[-1])
|
|
|
|
def test_group_admissible_discarded_dupes(self):
|
|
# If, with a (A, B, C, D) set, all match with A, but C and D don't match with B and that the
|
|
# (A, B) match is the highest (thus resulting in an (A, B) group), still match C and D
|
|
# in a separate group instead of discarding them.
|
|
A, B, C, D = [NamedObject() for _ in range(4)]
|
|
m1 = Match(A, B, 90) # This is the strongest "A" match
|
|
m2 = Match(A, C, 80) # Because C doesn't match with B, it won't be in the group
|
|
m3 = Match(A, D, 80) # Same thing for D
|
|
m4 = Match(C, D, 70) # However, because C and D match, they should have their own group.
|
|
groups = get_groups([m1, m2, m3, m4])
|
|
eq_(len(groups), 2)
|
|
g1, g2 = groups
|
|
assert A in g1
|
|
assert B in g1
|
|
assert C in g2
|
|
assert D in g2
|
|
|