Path is always set to "/" using unicode wsgi protocol under Python 3

[ghost]

Originally reported by: Lars Gustäbel (Bitbucket: gustaebel, GitHub: gustaebel)

---

I am using a cherrypy server with wsgi_version = ("u", 0) under Python 3, and found a bug in _cherrypy.cptree.Tree.call().

The problematic code is the following:

#!python

env1x = environ
if environ.get(ntou('wsgi.version')) == (ntou('u'), 0):
    env1x = _cpwsgi.downgrade_wsgi_ux_to_1x(environ)
path = httputil.urljoin(env1x.get('SCRIPT_NAME', ''),
                        env1x.get('PATH_INFO', ''))
sn = self.script_name(path or "/")

The unicode environment is converted to bytes (in Python 3) in order to assemble the path from SCRIPT_NAME and PATH_INFO. The problem is that after this conversion all keys are bytes objects, so both env1x.get() calls fail and return the empty string. This in turn leads to httputil.urljoin() returning the empty string which is then replaced by "/" in the script_name() call.

So, whatever url you pass to Tree.call() they all get converted to "/". My solution to this problem is to test for py3k before downgrade_wsgi_ux_to_1x(). A patch is attached.

#!python

if py3k:
    if environ.get(ntou('wsgi.version')) == (ntou('u'), 0):
        env1x = _cpwsgi.downgrade_wsgi_ux_to_1x(environ)

---

• Bitbucket: https://bitbucket.org/cherrypy/cherrypy/issue/1350

[ghost]

Original comment by David Eriksson (Bitbucket: softhousedae, GitHub: Unknown):

---

Not fixed in over a year even when there is a patch? :-(

Runtime workaround:

#!python

cherrypy._cpwsgi.downgrade_wsgi_ux_to_1x = lambda env: env

[webknjaz]

@gustaebel @twogood could you please confirm that the issue still exists for the latest version of CherryPy?

[twogood]

@webknjaz For various reasons I'm still stuck on an old version of CherryPy, thanks for reminding me of upgrading. I did try a newer version at some point, but my application broke down, so there will be Work involved... :-)

[gustaebel]

I tried the version 8.1.0 of cherrypy but got the following traceback:

Traceback (most recent call last):
  File "/usr/lib/python3.5/site-packages/cherrypy/wsgiserver/__init__.py", line 1406, in communicate
    req.respond()
  File "/usr/lib/python3.5/site-packages/cherrypy/wsgiserver/__init__.py", line 860, in respond
    self.server.gateway(self).respond()
  File "/usr/lib/python3.5/site-packages/cherrypy/wsgiserver/__init__.py", line 2310, in __init__
    self.env = self.get_environ()
  File "/usr/lib/python3.5/site-packages/cherrypy/wsgiserver/__init__.py", line 2498, in get_environ
    env = dict(map(self._decode_key, self.items()))
AttributeError: 'WSGIGateway_u0' object has no attribute 'items'

With the following patch applied, all seems to work well :-)

diff --git a/cherrypy/wsgiserver/__init__.py b/cherrypy/wsgiserver/__init__.py
index 77fb361..98607e4 100644
--- a/cherrypy/wsgiserver/__init__.py
+++ b/cherrypy/wsgiserver/__init__.py
@@ -2497,7 +2497,7 @@ class WSGIGateway_u0(WSGIGateway_10):
         """Return a new environ dict targeting the given wsgi.version"""
         req = self.req
         env_10 = WSGIGateway_10.get_environ(self)
-        env = dict(map(self._decode_key, self.items()))
+        env = dict(self._decode_key(k, v) for k, v in env_10.items())
         env[six.u('wsgi.version')] = ('u', 0)

         # Request-URI
@@ -2511,7 +2511,7 @@ class WSGIGateway_u0(WSGIGateway_10):
             env['PATH_INFO'] = env_10['PATH_INFO']
             env['QUERY_STRING'] = env_10['QUERY_STRING']

-        env.update(map(self._decode_value, self.items()))
+        env.update(dict(self._decode_value(k, v) for k, v in env_10.items()))

         return env

[webknjaz]

@gustaebel feel free to send a PR along with tests

[jaraco]

It seems like the use of self.items is a bug going back to f2ec506.

[jaraco]

Meant to mention this ticket in 481b6f9, but mistakenly mentioned another.