projects
/
youtube-dl
/ blobdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
|
commitdiff
|
tree
raw
|
inline
| side by side
Fix MD5 calculation
[youtube-dl]
/
test
/
test_download.py
diff --git
a/test/test_download.py
b/test/test_download.py
index 00b9a487bc6acee80216d1328bdf70ef5a1e9dc4..571fbe9ad5ef253b9464b66bbe8eabb4334ae6b9 100644
(file)
--- a/
test/test_download.py
+++ b/
test/test_download.py
@@
-20,19
+20,6
@@
import youtube_dl.InfoExtractors
def _file_md5(fn):
with open(fn, 'rb') as f:
return hashlib.md5(f.read()).hexdigest()
def _file_md5(fn):
with open(fn, 'rb') as f:
return hashlib.md5(f.read()).hexdigest()
-
-def md5_for_file(filename, block_size=2**20):
- with open(filename) as f:
- md5 = hashlib.md5()
- while True:
- data = f.read(block_size)
- if not data:
- break
- md5.update(data)
- return md5.hexdigest()
-_file_md5 = md5_for_file
-
-
try:
_skip_unless = unittest.skipUnless
except AttributeError: # Python 2.6
try:
_skip_unless = unittest.skipUnless
except AttributeError: # Python 2.6
@@
-96,7
+83,7
@@
class DownloadTest(unittest.TestCase):
fd.download(['http://blip.tv/cbr/cbr-exclusive-gotham-city-imposters-bats-vs-jokerz-short-3-5796352'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
fd.download(['http://blip.tv/cbr/cbr-exclusive-gotham-city-imposters-bats-vs-jokerz-short-3-5796352'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
- self.assertEqual(md5_for_file, '
93c24d2f4e0782af13b8a7606ea97ba7
')
+ self.assertEqual(md5_for_file, '
4962f94441605832eb1008eb820ef47a
')
@_skip_unless(youtube_dl.InfoExtractors.XVideosIE._WORKING, "IE marked as not _WORKING")
def test_XVideos(self):
@_skip_unless(youtube_dl.InfoExtractors.XVideosIE._WORKING, "IE marked as not _WORKING")
def test_XVideos(self):
@@
-106,7
+93,7
@@
class DownloadTest(unittest.TestCase):
fd.download(['http://www.xvideos.com/video939581/funny_porns_by_s_-1'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
fd.download(['http://www.xvideos.com/video939581/funny_porns_by_s_-1'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
- self.assertEqual(md5_for_file, '
1ab4dedc01f771cb2a65e91caa801aaf
')
+ self.assertEqual(md5_for_file, '
aecab2ea59b7996110a7e409f0c55da3
')
@_skip_unless(youtube_dl.InfoExtractors.VimeoIE._WORKING, "IE marked as not _WORKING")
@_skip("No output file specified")
@_skip_unless(youtube_dl.InfoExtractors.VimeoIE._WORKING, "IE marked as not _WORKING")
@_skip("No output file specified")
@@
-127,7
+114,7
@@
class DownloadTest(unittest.TestCase):
fd.download(['http://soundcloud.com/ethmusic/lostin-powers-she-so-heavy'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
fd.download(['http://soundcloud.com/ethmusic/lostin-powers-she-so-heavy'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
- self.assertEqual(md5_for_file, 'c
e3775768ebb6432fa8495d446a078ed
')
+ self.assertEqual(md5_for_file, 'c
1b9b9ea8bfd620b96b2628664576e1c
')
@_skip_unless(youtube_dl.InfoExtractors.StanfordOpenClassroomIE._WORKING, "IE marked as not _WORKING")
def test_StanfordOpenClassroom(self):
@_skip_unless(youtube_dl.InfoExtractors.StanfordOpenClassroomIE._WORKING, "IE marked as not _WORKING")
def test_StanfordOpenClassroom(self):
@@
-137,7
+124,7
@@
class DownloadTest(unittest.TestCase):
fd.download(['http://openclassroom.stanford.edu/MainFolder/VideoPage.php?course=PracticalUnix&video=intro-environment&speed=100'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
fd.download(['http://openclassroom.stanford.edu/MainFolder/VideoPage.php?course=PracticalUnix&video=intro-environment&speed=100'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
- self.assertEqual(md5_for_file, '
22c8206291368c4e2c9c1a307f0ea0f
4')
+ self.assertEqual(md5_for_file, '
8aac7873a07dcfaed66b1559ab12851
4')
@_skip_unless(youtube_dl.InfoExtractors.CollegeHumorIE._WORKING, "IE marked as not _WORKING")
@_skip("No output file specified")
@_skip_unless(youtube_dl.InfoExtractors.CollegeHumorIE._WORKING, "IE marked as not _WORKING")
@_skip("No output file specified")
@@
-158,7
+145,7
@@
class DownloadTest(unittest.TestCase):
fd.download(['http://video.xnxx.com/video1135332/lida_naked_funny_actress_5_'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
fd.download(['http://video.xnxx.com/video1135332/lida_naked_funny_actress_5_'])
self.assertTrue(os.path.exists(filename))
md5_for_file = _file_md5(filename)
- self.assertEqual(md5_for_file, '
5f0469c8d1dfd1bc38c8e6deb5e0a21d
')
+ self.assertEqual(md5_for_file, '
c5c67df477eb0d9b058200351448ba4c
')
def tearDown(self):
def tearDown(self):