urllib.parse.urlsplit raises an exception for an url including a non-ascii hostname in NFKD form and a port number.

example:
>>> urlsplit('http://\u30d5\u309a:80')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/ito/.maltybrew/deen/lib/python3.7/urllib/parse.py", line 437, in urlsplit
    _checknetloc(netloc)
  File "/Users/ito/.maltybrew/deen/lib/python3.7/urllib/parse.py", line 407, in _checknetloc
    "characters under NFKC normalization")
ValueError: netloc 'プ:80' contains invalid characters under NFKC normalization
>>> urlsplit('http://\u30d5\u309a')
SplitResult(scheme='http', netloc='プ', path='', query='', fragment='')
>>> urlsplit(unicodedata.normalize('NFKC', 'http://\u30d5\u309a:80'))
SplitResult(scheme='http', netloc='プ:80', path='', query='', fragment='')

I believe this behavior was introduced at Python 3.7.3. Python 3.7.2 doesn't raise any exception for these lines.

This could be due to issue36216.

Yes, it's due to that. I guess we need to do netloc.rpartition(':') like we currently do for '@' in _checknetloc.

Promoting to release blocker and security issue to match the original issue. I can't get to this today, but I should be able to at the PyCon sprints next week if nobody else gets it sooner.

I found the time to get the first patch. Hopefully backports to 3.6 and 3.7 are easy, but I think 2.7 will take manual steps.

Chihiro Ito - if you have other test scenarios, it would be great if you could try them out with the fix in PR 13017. It should be easy enough to copy into your installed Python.

New changeset d537ab0ff9767ef024f26246899728f0116b1ec3 by Steve Dower in branch 'master':
bpo-36742: Fixes handling of pre-normalization characters in urlsplit() (GH-13017)
https://github.com/python/cpython/commit/d537ab0ff9767ef024f26246899728f0116b1ec3

New changeset 4d723e76e1ad17e9e7d5e828e59bb47e76f2174b by Miss Islington (bot) in branch '3.7':
bpo-36742: Fixes handling of pre-normalization characters in urlsplit() (GH-13017)
https://github.com/python/cpython/commit/4d723e76e1ad17e9e7d5e828e59bb47e76f2174b

I have confirmed that all of my app's test cases have passed.

What I've done:
1. Installed Python 3.7.3.
2. Replaced urllib/parse.py with the one from 781ffb1.
3. Ran my app's test cases.

Thank you for the quick fix!

New changeset 98a4dcefbbc3bce5ab07e7c0830a183157250259 by Steve Dower in branch '2.7':
bpo-36742: Fixes handling of pre-normalization characters in urlsplit() (GH-13017)
https://github.com/python/cpython/commit/98a4dcefbbc3bce5ab07e7c0830a183157250259

I'll leave the 3.6 backport in Ned's hands and close this issue.

> I'll leave the 3.6 backport in Ned's hands and close this issue.

3.5 was added as an affected version and seems the original fix was merged to 3.5 too. 3.4 is EoL so is it worthy of backporting to 3.5? I guess the backport would not have merge conflicts and is straightforward.

Yes, you're right. I'll do that port as well.

New changeset e5f9f4adb95233c66578e6f7ea176687af2f78ca by Ned Deily (Miss Islington (bot)) in branch '3.6':
bpo-36742: Fixes handling of pre-normalization characters in urlsplit() (GH-13017) (GH-13024)
https://github.com/python/cpython/commit/e5f9f4adb95233c66578e6f7ea176687af2f78ca

New changeset 8d0ef0b5edeae52960c7ed05ae8a12388324f87e by Łukasz Langa (Steve Dower) in branch 'master':
bpo-36742: Corrects fix to handle decomposition in usernames (#13812)
https://github.com/python/cpython/commit/8d0ef0b5edeae52960c7ed05ae8a12388324f87e

Thanks for this engagement and pull requests, Steve.
Thanks for reviews Karthikeyan.

New changeset 250b62acc59921d399f0db47db3b462cd6037e09 by Miss Islington (bot) in branch '3.7':
bpo-36742: Corrects fix to handle decomposition in usernames (GH-13812)
https://github.com/python/cpython/commit/250b62acc59921d399f0db47db3b462cd6037e09

New changeset f61599b050c621386a3fc6bc480359e2d3bb93de by Steve Dower in branch '2.7':
bpo-36742: Corrects fix to handle decomposition in usernames (GH-13812)
https://github.com/python/cpython/commit/f61599b050c621386a3fc6bc480359e2d3bb93de

New changeset fd1771dbdd28709716bd531580c40ae5ed814468 by Ned Deily (Miss Islington (bot)) in branch '3.6':
bpo-36742: Corrects fix to handle decomposition in usernames (GH-13812) (GH-13814)
https://github.com/python/cpython/commit/fd1771dbdd28709716bd531580c40ae5ed814468

CVE-2019-10160 has been assigned by Red Hat to this flaw.

The fix for python-2.7 (https://github.com/python/cpython/pull/13815/files#diff-b577545d73dd0cdb2c337a4c5f89e1d7R183) causes errors when netloc contains characters that can't be encoded by 'ascii' codec.

You can see it by doing:
>>> netloc = u'example.com\uFF03@bing.com'
>>> raise ValueError(u"netloc '" + netloc + u"' contains invalid characters under NFKC normalization")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: <exception str() failed>
>>> str(netloc)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
UnicodeEncodeError: 'ascii' codec can't encode character u'\uff03' in position 11: ordinal not in range(128)

I suggest we use `repr(netloc)` instead of `netloc` in the ValueError message.

> CVE-2019-10160 has been assigned by Red Hat to this flaw.

For clarity, CVE-2019-10160 has been assigned to the bug introduced with the fix for the functional regression mentioned in this bug, and not to the bug itself explained in the first comment. See https://bugzilla.redhat.com/show_bug.cgi?id=1718388 for more details about it.

New changeset 2b578479b96aa3deeeb8bac313a02b5cf3cb1aff by Victor Stinner in branch '2.7':
[2.7] bpo-36742: Fix urlparse.urlsplit() error message for Unicode URL (GH-13937)
https://github.com/python/cpython/commit/2b578479b96aa3deeeb8bac313a02b5cf3cb1aff

New changeset 4655d576141ee56a69d2052431c636858fcb916a by larryhastings (Steve Dower) in branch '3.5':
bpo-36742: Fixes handling of pre-normalization characters in urlsplit() (GH-13017) (#13042)
https://github.com/python/cpython/commit/4655d576141ee56a69d2052431c636858fcb916a

New changeset 095373c32d16df575ba5fcb5f44bf44119b26193 by larryhastings (Victor Stinner) in branch '3.5':
bpo-36742: Corrects fix to handle decomposition in usernames (GH-13812) (GH-13814) (#14772)
https://github.com/python/cpython/commit/095373c32d16df575ba5fcb5f44bf44119b26193