CherryPy's default dispatcher is a very powerful and intuitive system for most applications. This is especially true for "RESTful" apps (where by "RESTful" people mean "cool" URI's like /thing/12/bit/84) because the hierarchy of containers in the URI is mirrored in the tree of page handlers. That is, it's natural to map the URI /thing/12/bit/84 to an object reference like Thing.12.bit.84 or even Thing.bit.
Now, Python doesn't naturally allow attribute names that are digits, like 12. It does, however, allow them via the "magic method" called __getattr__. We can use this to our advantage when constructing an object tree for these kinds of URI's. Here's an example application which handles the URI /thing/12/bit/84. It also handles the intermediate URI's /thing, /thing/12, /thing/new, and /thing/12/bit, by the way:
import cherrypy from myapp import bits class Root: @cherrypy.expose def index(self): return "Things galore!" @cherrypy.expose def new(self): if cherrypy.request.method != 'POST': cherrypy.response.headers['Allow'] = 'POST' raise cherrypy.HTTPError(405) thing = Thing() raise cherrypy.HTTPRedirect('/%d' % thing.id) @cherrypy.expose def default(self, thingid): if not thingid.isdigit(): raise cherrypy.NotFound() thing = Thing(id=int(thingid)) if thing is None: raise cherrypy.NotFound() return thing.html() default.bit = bits.Bits() def __getattr__(self, name): if name.isdigit(): cherrypy.request.params['thingid'] = name return self.default raise AttributeError("%r object has no attribute %r" % (self.__class__.__name__, name)) root = Root() if __name__ == '__main__': cherrypy.quickstart(root, '/thing')
The only portion worth discussing in the above is the __getattr__ method; it is called while the dispatcher is traversing the URI from left to right. At the point when the dispatcher tries to find a candidate for /thing/12, it calls getattr(root, '12', None). Our __getattr__ method above answers the call, and returns self.default. It must return this function so that a request for /thing/12 is handled by the default method. If we didn't include the __getattr__ method, CherryPy would pass the path segment 12 as a "virtual path" positional argument. Because we're returning default ourselves manually, we have to capture that argument and pass it as a "param" keyword arg instead.
But we've done one more bit of trickery here: we've set default.bit = bits.Bits(). By doing this, a request for /thing/12/bit/... will continue traversal into the Bits class.
Here is the 'Bits' module which the above imports:
import cherrypy class Bits: @cherrypy.expose def index(self, thingid=None): return "Bits galore!" @cherrypy.expose def default(self, bitid, thingid=None): thing = Thing.get(id=int(thingid)) if thing is None: raise cherrypy.NotFound() bit = Thing.bits.get(id=int(bitid)) if bit is None: raise cherrypy.NotFound() return bit.html()
There's nothing special about the Bits class; it's a normal CherryPy handler class. The only thing we have to be careful about is the order of arguments to the default method. Remember that we captured the thingid argument and stuck it in request.params, so it's not going to be passed as a positional argument as you might expect. The bitid argument will be passed positionally, however, because the Bits class has no such __getattr__ method. Therefore, it must come first, and the thingid argument must come last.
In this example, we've limited ourselves to numeric "dynamic path segments", because they're easiest to explain. You can also handle non-numeric bytes, but it's a bit trickier since those might collide with normal attributes on the class. One way to differentiate them would be to perform the database lookup inside the __getattr__ method instead of the subsequent methods. You could even pass a fully-resolved database object to your page handlers this way, instead of just the id.
Regardless of whether you handle non-numeric atoms or not, be aware that the __getattr__ method might be called in some situations other than the traversal performed by the dispatcher. That's why, for example, we took pains to write self.__class__.__name__ instead of just repr(self) in that method: if self has no __repr__ method, you'd enter an infinite loop! Make sure you raise AttributeError if you don't want to handle the name as a valid URI path segment.