Skip to content
Projeler
Gruplar
Parçacıklar
Yardım
Yükleniyor...
Oturum aç / Kaydol
Gezinmeyi değiştir
C
cpython
Proje
Proje
Ayrıntılar
Etkinlik
Cycle Analytics
Depo (repository)
Depo (repository)
Dosyalar
Kayıtlar (commit)
Dallar (branch)
Etiketler
Katkıda bulunanlar
Grafik
Karşılaştır
Grafikler
Konular (issue)
0
Konular (issue)
0
Liste
Pano
Etiketler
Kilometre Taşları
Birleştirme (merge) Talepleri
0
Birleştirme (merge) Talepleri
0
CI / CD
CI / CD
İş akışları (pipeline)
İşler
Zamanlamalar
Grafikler
Paketler
Paketler
Wiki
Wiki
Parçacıklar
Parçacıklar
Üyeler
Üyeler
Collapse sidebar
Close sidebar
Etkinlik
Grafik
Grafikler
Yeni bir konu (issue) oluştur
İşler
Kayıtlar (commit)
Konu (issue) Panoları
Kenar çubuğunu aç
Batuhan Osman TASKAYA
cpython
Commits
2b67c7aa
Kaydet (Commit)
2b67c7aa
authored
May 21, 2017
tarafından
Serhiy Storchaka
Kaydeden (comit)
GitHub
May 21, 2017
Dosyalara gözat
Seçenekler
Dosyalara Gözat
İndir
Eposta Yamaları
Sade Fark
[2.7] bpo-30415: Add new tests for the fnmatch module. (GH-1684). (#1696)
(cherry picked from commit
81755471
)
üst
cef8b174
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
104 additions
and
11 deletions
+104
-11
test_fnmatch.py
Lib/test/test_fnmatch.py
+104
-11
No files found.
Lib/test/test_fnmatch.py
Dosyayı görüntüle @
2b67c7aa
...
...
@@ -2,9 +2,10 @@
from
test
import
test_support
import
unittest
import
os
from
fnmatch
import
fnmatch
,
fnmatchcase
,
_MAXCACHE
,
_cache
from
fnmatch
import
fnmatch
,
fnmatchcase
,
_MAXCACHE
,
_cache
,
_purge
from
fnmatch
import
(
fnmatch
,
fnmatchcase
,
translate
,
filter
,
_MAXCACHE
,
_cache
,
_purge
)
class
FnmatchTestCase
(
unittest
.
TestCase
):
...
...
@@ -12,13 +13,13 @@ class FnmatchTestCase(unittest.TestCase):
def
tearDown
(
self
):
_purge
()
def
check_match
(
self
,
filename
,
pattern
,
should_match
=
1
,
fn
=
fnmatch
):
def
check_match
(
self
,
filename
,
pattern
,
should_match
=
True
,
fn
=
fnmatch
):
if
should_match
:
self
.
assertTrue
(
fn
(
filename
,
pattern
),
"expected
%
r to match pattern
%
r"
%
(
filename
,
pattern
))
else
:
self
.
assert
True
(
not
fn
(
filename
,
pattern
),
self
.
assert
False
(
fn
(
filename
,
pattern
),
"expected
%
r not to match pattern
%
r"
%
(
filename
,
pattern
))
...
...
@@ -32,15 +33,15 @@ class FnmatchTestCase(unittest.TestCase):
check
(
'abc'
,
'*'
)
check
(
'abc'
,
'ab[cd]'
)
check
(
'abc'
,
'ab[!de]'
)
check
(
'abc'
,
'ab[de]'
,
0
)
check
(
'a'
,
'??'
,
0
)
check
(
'a'
,
'b'
,
0
)
check
(
'abc'
,
'ab[de]'
,
False
)
check
(
'a'
,
'??'
,
False
)
check
(
'a'
,
'b'
,
False
)
# these test that '\' is handled correctly in character sets;
# see SF bug #409651
check
(
'
\\
'
,
r'[\]'
)
check
(
'a'
,
r'[!\]'
)
check
(
'
\\
'
,
r'[!\]'
,
0
)
check
(
'
\\
'
,
r'[!\]'
,
False
)
# test that filenames with newlines in them are handled correctly.
# http://bugs.python.org/issue6665
...
...
@@ -49,10 +50,29 @@ class FnmatchTestCase(unittest.TestCase):
check
(
'
\n
foo'
,
'foo*'
,
False
)
check
(
'
\n
'
,
'*'
)
def
test_mix_unicode_str
(
self
):
check
=
self
.
check_match
check
(
'test'
,
u'*'
)
check
(
u'test'
,
'*'
)
check
(
'test'
,
u'*'
,
fn
=
fnmatchcase
)
check
(
u'test'
,
'*'
,
fn
=
fnmatchcase
)
with
test_support
.
check_warnings
((
""
,
UnicodeWarning
),
quiet
=
True
):
check
(
'test
\xff
'
,
u'*
\xff
'
)
check
(
u'test
\xff
'
,
'*
\xff
'
)
check
(
'test
\xff
'
,
u'*
\xff
'
,
fn
=
fnmatchcase
)
check
(
u'test
\xff
'
,
'*
\xff
'
,
fn
=
fnmatchcase
)
def
test_fnmatchcase
(
self
):
check
=
self
.
check_match
check
(
'AbC'
,
'abc'
,
0
,
fnmatchcase
)
check
(
'abc'
,
'AbC'
,
0
,
fnmatchcase
)
check
(
'abc'
,
'abc'
,
True
,
fnmatchcase
)
check
(
'AbC'
,
'abc'
,
False
,
fnmatchcase
)
check
(
'abc'
,
'AbC'
,
False
,
fnmatchcase
)
check
(
'AbC'
,
'AbC'
,
True
,
fnmatchcase
)
check
(
'usr/bin'
,
'usr/bin'
,
True
,
fnmatchcase
)
check
(
'usr
\\
bin'
,
'usr/bin'
,
False
,
fnmatchcase
)
check
(
'usr/bin'
,
'usr
\\
bin'
,
False
,
fnmatchcase
)
check
(
'usr
\\
bin'
,
'usr
\\
bin'
,
True
,
fnmatchcase
)
def
test_cache_clearing
(
self
):
# check that caches do not grow too large
...
...
@@ -64,8 +84,81 @@ class FnmatchTestCase(unittest.TestCase):
self
.
assertLessEqual
(
len
(
_cache
),
_MAXCACHE
)
@test_support.requires_unicode
def
test_unicode
(
self
):
with
test_support
.
check_warnings
((
""
,
UnicodeWarning
),
quiet
=
True
):
self
.
check_match
(
u'test'
,
u'te*'
)
self
.
check_match
(
u'test
\xff
'
,
u'te*
\xff
'
)
self
.
check_match
(
u'test'
+
unichr
(
0x20ac
),
u'te*'
+
unichr
(
0x20ac
))
self
.
check_match
(
u'foo
\n
bar'
,
u'foo*'
)
def
test_case
(
self
):
ignorecase
=
os
.
path
.
normcase
(
'ABC'
)
==
os
.
path
.
normcase
(
'abc'
)
check
=
self
.
check_match
check
(
'abc'
,
'abc'
)
check
(
'AbC'
,
'abc'
,
ignorecase
)
check
(
'abc'
,
'AbC'
,
ignorecase
)
check
(
'AbC'
,
'AbC'
)
def
test_sep
(
self
):
normsep
=
os
.
path
.
normcase
(
'
\\
'
)
==
os
.
path
.
normcase
(
'/'
)
check
=
self
.
check_match
check
(
'usr/bin'
,
'usr/bin'
)
check
(
'usr
\\
bin'
,
'usr/bin'
,
normsep
)
check
(
'usr/bin'
,
'usr
\\
bin'
,
normsep
)
check
(
'usr
\\
bin'
,
'usr
\\
bin'
)
class
TranslateTestCase
(
unittest
.
TestCase
):
def
test_translate
(
self
):
self
.
assertEqual
(
translate
(
'*'
),
r'.*\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'?'
),
r'.\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'a?b*'
),
r'a.b.*\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'[abc]'
),
r'[abc]\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'[]]'
),
r'[]]\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'[!x]'
),
r'[^x]\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'[^x]'
),
r'[\^x]\Z(?ms)'
)
self
.
assertEqual
(
translate
(
'[x'
),
r'\[x\Z(?ms)'
)
class
FilterTestCase
(
unittest
.
TestCase
):
def
test_filter
(
self
):
self
.
assertEqual
(
filter
([
'Python'
,
'Ruby'
,
'Perl'
,
'Tcl'
],
'P*'
),
[
'Python'
,
'Perl'
])
self
.
assertEqual
(
filter
([
u'Python'
,
u'Ruby'
,
u'Perl'
,
u'Tcl'
],
u'P*'
),
[
u'Python'
,
u'Perl'
])
with
test_support
.
check_warnings
((
""
,
UnicodeWarning
),
quiet
=
True
):
self
.
assertEqual
(
filter
([
u'test
\xff
'
],
u'*
\xff
'
),
[
u'test
\xff
'
])
@test_support.requires_unicode
def
test_mix_bytes_str
(
self
):
with
test_support
.
check_warnings
((
""
,
UnicodeWarning
),
quiet
=
True
):
self
.
assertEqual
(
filter
([
'test'
],
u'*'
),
[
'test'
])
self
.
assertEqual
(
filter
([
u'test'
],
'*'
),
[
u'test'
])
self
.
assertEqual
(
filter
([
'test
\xff
'
],
u'*'
),
[
'test
\xff
'
])
self
.
assertEqual
(
filter
([
u'test
\xff
'
],
'*'
),
[
u'test
\xff
'
])
self
.
assertEqual
(
filter
([
'test
\xff
'
],
u'*
\xff
'
),
[
'test
\xff
'
])
self
.
assertEqual
(
filter
([
u'test
\xff
'
],
'*
\xff
'
),
[
u'test
\xff
'
])
def
test_case
(
self
):
ignorecase
=
os
.
path
.
normcase
(
'P'
)
==
os
.
path
.
normcase
(
'p'
)
self
.
assertEqual
(
filter
([
'Test.py'
,
'Test.rb'
,
'Test.PL'
],
'*.p*'
),
[
'Test.py'
,
'Test.PL'
]
if
ignorecase
else
[
'Test.py'
])
self
.
assertEqual
(
filter
([
'Test.py'
,
'Test.rb'
,
'Test.PL'
],
'*.P*'
),
[
'Test.py'
,
'Test.PL'
]
if
ignorecase
else
[
'Test.PL'
])
def
test_sep
(
self
):
normsep
=
os
.
path
.
normcase
(
'
\\
'
)
==
os
.
path
.
normcase
(
'/'
)
self
.
assertEqual
(
filter
([
'usr/bin'
,
'usr'
,
'usr
\\
lib'
],
'usr/*'
),
[
'usr/bin'
,
'usr
\\
lib'
]
if
normsep
else
[
'usr/bin'
])
self
.
assertEqual
(
filter
([
'usr/bin'
,
'usr'
,
'usr
\\
lib'
],
'usr
\\
*'
),
[
'usr/bin'
,
'usr
\\
lib'
]
if
normsep
else
[
'usr
\\
lib'
])
def
test_main
():
test_support
.
run_unittest
(
FnmatchTestCase
)
test_support
.
run_unittest
(
FnmatchTestCase
,
TranslateTestCase
,
FilterTestCase
)
if
__name__
==
"__main__"
:
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment