I can confirm that the patch applies cleanly to the release26-maint
branch, and that the updated test fails without the updated
implementation.

However, the entire approach seems wrong to me:  the patched method
has just called 'request_path', which already cracked the URL using
urlparse.urlparse and put it back together again with
urlparse.urlunparse.

A better fix would be either to call urlparse.urlparse directly, or
else to add an argument to 'request_path' which kept it from doing the
unwanted reassembly.

Attaching a patch which implements the latter strategy.  Greg's new
test passes with this patch applied in place of the part of his patch
which touches Lib/cookielib.py.