This is a regression caused by revision 901e4e52b20a. Before (e.g. Python 3.4):

>>> urljoin('a', 'b')
'b'

After:

>>> urljoin('a', 'b')
'b/b'

This was identified in <https://bugs.python.org/issue18828#msg238363> but is actually caused by the revision committed for Issue 22278.

Nice work tracking down the commit, thanks for that. I've attached a fix.

Patch looks good enough.

BTW I was thinking of something like this, which is a bit simpler, but I never tested it:

segments[1:-1] = filter(None, segments[1:-1])

Yep, that's a lot cleaner and tests pass. I've updated the patch with that line. Thanks.

Any chance this regression can be fixed before the next release?

New changeset fc0e79387a3a by Berker Peksag in branch 'default':
Issue #23703: Fix a regression in urljoin() introduced in 901e4e52b20a.
https://hg.python.org/cpython/rev/fc0e79387a3a

Thanks!