aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--test/test_download.py30
-rw-r--r--test/test_utils.py10
2 files changed, 23 insertions, 17 deletions
diff --git a/test/test_download.py b/test/test_download.py
index 84a98908a..7de9f2a48 100644
--- a/test/test_download.py
+++ b/test/test_download.py
@@ -74,9 +74,10 @@ class DownloadTest(unittest.TestCase):
md5_down_file = md5_for_file(DownloadTest.DAILYMOTION_FILE)
self.assertEqual(md5_down_file, DownloadTest.DAILYMOTION_MD5)
- @unittest.skip("no suitable ie")
def test_metacafe(self):
- with open("test/json") as f:
+ #this emulate a skip,to be 2.6 compatible
+ return
+ with open(DownloadTest.PARAMETERS_FILE) as f:
fd = FileDownloader(json.load(f))
print fd
fd.add_info_extractor(MetacafeIE())
@@ -85,45 +86,50 @@ class DownloadTest(unittest.TestCase):
md5_down_file = md5_for_file(DownloadTest.METACAFE_FILE)
self.assertEqual(md5_down_file, DownloadTest.METACAFE_MD5)
- @unittest.skip("no suitable url")
def test_photobucket(self):
- fd = FileDownloader({})
+ return
+ with open(DownloadTest.PARAMETERS_FILE) as f:
+ fd = FileDownloader(json.load(f))
fd.add_info_extractor(PhotobucketIE())
fd.download([DownloadTest.PHOTOBUCKET_URL])
self.assertTrue(os.path.exists(DownloadTest.PHOTOBUCKET_FILE))
md5_down_file = md5_for_file(DownloadTest.PHOTOBUCKET_FILE)
self.assertEqual(md5_down_file, DownloadTest.PHOTOBUCKET_MD5)
- @unittest.skip("no suitable url")
def test_facebook(self):
- fd = FileDownloader({})
+ return
+ with open(DownloadTest.PARAMETERS_FILE) as f:
+ fd = FileDownloader(json.load(f))
fd.add_info_extractor(FacebookIE())
fd.download([DownloadTest.FACEBOOK_URL])
self.assertTrue(os.path.exists(DownloadTest.FACEBOOK_FILE))
md5_down_file = md5_for_file(DownloadTest.FACEBOOK_FILE)
self.assertEqual(md5_down_file, DownloadTest.FACEBOOK_MD5)
- @unittest.skip("no suitable url")
def test_blip(self):
- fd = FileDownloader({})
+ return
+ with open(DownloadTest.PARAMETERS_FILE) as f:
+ fd = FileDownloader(json.load(f))
fd.add_info_extractor(BlipTVIE())
fd.download([DownloadTest.BLIP_URL])
self.assertTrue(os.path.exists(DownloadTest.BLIP_FILE))
md5_down_file = md5_for_file(DownloadTest.BLIP_FILE)
self.assertEqual(md5_down_file, DownloadTest.BLIP_MD5)
- @unittest.skip("no suitable url")
def test_vimeo(self):
- fd = FileDownloader({})
+ return
+ with open(DownloadTest.PARAMETERS_FILE) as f:
+ fd = FileDownloader(json.load(f))
fd.add_info_extractor(VimeoIE())
fd.download([DownloadTest.VIMEO_URL])
self.assertTrue(os.path.exists(DownloadTest.VIMEO_FILE))
md5_down_file = md5_for_file(DownloadTest.VIMEO_FILE)
self.assertEqual(md5_down_file, DownloadTest.VIMEO_MD5)
- @unittest.skip("no suitable url")
def test_xvideo(self):
- fd = FileDownloader({})
+ return
+ with open(DownloadTest.PARAMETERS_FILE) as f:
+ fd = FileDownloader(json.load(f))
fd.add_info_extractor(XVideosIE())
fd.download([DownloadTest.XVIDEO_URL])
self.assertTrue(os.path.exists(DownloadTest.XVIDEO_FILE))
diff --git a/test/test_utils.py b/test/test_utils.py
index fd3ac4d46..eb0af703f 100644
--- a/test/test_utils.py
+++ b/test/test_utils.py
@@ -13,8 +13,8 @@ from youtube_dl.utils import orderedSet
class TestUtil(unittest.TestCase):
def test_timeconvert(self):
- self.assertIsNone(timeconvert(''))
- self.assertIsNone(timeconvert('bougrg'))
+ self.assertTrue(timeconvert('') is None)
+ self.assertTrue(timeconvert('bougrg') is None)
def test_sanitize_filename(self):
self.assertEqual(sanitize_filename(u'abc'), u'abc')
@@ -23,12 +23,12 @@ class TestUtil(unittest.TestCase):
self.assertEqual(sanitize_filename(u'123'), u'123')
self.assertEqual(u'abc_de', sanitize_filename(u'abc/de'))
- self.assertIn(u'de', sanitize_filename(u'abc/de'))
- self.assertNotIn(u'/', sanitize_filename(u'abc/de///'))
+ self.assertTrue(u'de' in sanitize_filename(u'abc/de'))
+ self.assertFalse(u'/' in sanitize_filename(u'abc/de///'))
self.assertEqual(u'abc_de', sanitize_filename(u'abc\\de'))
self.assertEqual(u'abc_de', sanitize_filename(u'abc\\de'))
- self.assertIn(u'de', sanitize_filename(u'abc\\de'))
+ self.assertTrue(u'de' in sanitize_filename(u'abc\\de'))
self.assertEqual(sanitize_filename(u'ä'), u'ä')
self.assertEqual(sanitize_filename(u'кириллица'), u'кириллица')