Ticket #613 (defect)

Patch file for this bug

The patch does two things, it renames the is_index attribute of the request object to is_dir, and causes it to be set to true for both index and default handlers. I renamed this attribute because I thought it would be more appropriate than just setting is_index to true for a default handler.

Have you tried this:

oops it did not work

import cherrypy

class Root:
    @cherrypy.expose
    def default(self, path=None):
        if not path:
            raise cherrypy.HTTPRedirect(cherrypy.url('/some'))
        return path
    
if __name__ == '__main__':
    cherrypy.quickstart(Root(), '/test')

True, that works, but this doesn't unfortunately:

import cherrypy

class Root:
    def __init__(self):
        self.test = Test()

class Test:
    @cherrypy.expose
    def default(self, path=None):
        if not path:
            raise cherrypy.HTTPRedirect(cherrypy.url('/some'))
        return path
    
if __name__ == '__main__':
    cherrypy.quickstart(Root())

Your testcase mounted the default handler on the base (the scriptname) of the site, which caused cherrypy.url to return the new url properly. Mine isn't setup that way as I have several controller classes mounted on the root of the site, so I can't put them each as the scriptname.

In that case I'd say CP's behavior is correct and as expected. I think you understand that mounting the application as you do does not tell CP that this is a distinct application. This is precisely where the subtlety resides.

In your example as you say you mount your controllers on the root application and they CherryPy treats them as such. Controllers.

If you want them to behave as applications do, you have to mount them via cherrypy.tree.mount.

Maybe with something like:

import cherrypy

class Root:
    def __init__(self, base=''):
        cherrypy.tree.mount(Test(), "%s/test" % base)

class Test:
    @cherrypy.expose
    def default(self, path=None):
        if not path:
            raise cherrypy.HTTPRedirect(cherrypy.url('/some'))
        return path
    
if __name__ == '__main__':
    cherrypy.quickstart(Root())

Ok, maybe I just don't understand the role of a default controller. The way I understand cherrypy's routing methods (which may be wrong, I am very new at this) is that classes represent directories, and functions represent files. So since the default handler is a function, it should be considered a file under a directory, and be able to redirect to other files in that directory.

For instance, if I redirect from the index handler I would get a file in that directory. So if I have an index handler for /students/, which redirects to 'contact', I would get /students/contact, but not for a default handler. This behavior seems inconsistent to me, but that could be because of a lack of understanding on my part.

Also, in my application the separate classes are not different apps, they are just different parts of the same app.

Thanks for helping me understand this, Ethilien

Oh, and here's the thread on the users' mailing list regarding this bug.

http://groups-beta.google.com/group/cherrypy-users/browse_thread/thread/80eb4dc85e2a87f9

Ok, maybe I just don't understand the role of a
default controller. The way I understand cherrypy's
routing methods (which may be wrong, I am very new
at this) is that classes represent directories,
and functions represent files. So since the default
handler is a function, it should be considered a
file under a directory, and be able to redirect
to other files in that directory.

That's a bit of a non sequitur, since URI's are not files or directories, and HTTP is not a file system. All handlers are exposed functions, and return responses "based on the request content". See http://roy.gbiv.com/pubs/dissertation/evaluation.htm#sec_6_2_3

The upshot is that any handler, and therefore any URI, should be able to respond with a redirect. This includes "normal" handlers, as well as "default" and "index" handlers.

For instance, if I redirect from the index handler
I would get a file in that directory. So if I have
an index handler for /students/, which redirects to
'contact', I would get /students/contact, but not for
a default handler. This behavior seems inconsistent
to me, but that could be because of a lack of under-
standing on my part.

Correct; it is inconsistent. A test case which demonstrates the problem is in [1461], and a fix in [1463]. Note that this differs from the proposed patch; not all default methods should be flagged as index handlers--only those which are looked up using a URI with a trailing slash. Also, for the reasons stated above, the name will remain "is_index", not "is_dir" for now. This is less than ideal, but the full semantic would dictate an attribute named "request.rightmost_path_segment_is_empty_except_possible_parameters" and that's not fun to type nor remember. ;) See http://rfc.net/rfc2396.html#s3.3.