A Do-It-Yourself Framework
++++++++++++++++++++++++++
:author: Ian Bicking <ianb@colorstudy.com>
:revision: $Rev$
:date: $LastChangedDate$
This tutorial has been translated `into Portuguese
<http://montegasppa.blogspot.com/2007/06/um-framework-faa-voc-mesmo.html>`_.
A newer version of this article is available `using WebOb
<http://pythonpaste.org/webob/do-it-yourself.html>`_.
.. contents::
.. comments:
Explain SCRIPT_NAME/PATH_INFO better
Introduction and Audience
=========================
This short tutorial is meant to teach you a little about WSGI, and as
an example a bit about the architecture that Paste has enabled and
encourages.
This isn't an introduction to all the parts of Paste -- in fact, we'll
only use a few, and explain each part. This isn't to encourage
everyone to go off and make their own framework (though honestly I
wouldn't mind). The goal is that when you have finished reading this
you feel more comfortable with some of the frameworks built using this
architecture, and a little more secure that you will understand the
internals if you look under the hood.
What is WSGI?
=============
At its simplest WSGI is an interface between web servers and web
applications. We'll explain the mechanics of WSGI below, but a higher
level view is to say that WSGI lets code pass around web requests in a
fairly formal way. But there's more! WSGI is more than just HTTP.
It might seem like it is just *barely* more than HTTP, but that little
bit is important:
* You pass around a CGI-like environment, which means data like
``REMOTE_USER`` (the logged-in username) can be securely passed
about.
* A CGI-like environment can be passed around with more context --
specifically instead of just one path you two: ``SCRIPT_NAME`` (how
we got here) and ``PATH_INFO`` (what we have left).
* You can -- and often should -- put your own extensions into the WSGI
environment. This allows for callbacks, extra information,
arbitrary Python objects, or whatever you want. These are things
you can't put in custom HTTP headers.
This means that WSGI can be used not just between a web server an an
application, but can be used at all levels for communication. This
allows web applications to become more like libraries -- well
encapsulated and reusable, but still with rich reusable functionality.
Writing a WSGI Application
==========================
The first part is about how to use `WSGI
<http://www.python.org/peps/pep-0333.html>`_ at its most basic. You
can read the spec, but I'll do a very brief summary:
* You will be writing a *WSGI application*. That's an object that
responds to requests. An application is just a callable object
(like a function) that takes two arguments: ``environ`` and
``start_response``.
* The environment looks a lot like a CGI environment, with keys like
``REQUEST_METHOD``, ``HTTP_HOST``, etc.
* The environment also has some special keys like ``wsgi.input`` (the
input stream, like the body of a POST request).
* ``start_response`` is a function that starts the response -- you
give the status and headers here.
* Lastly the application returns an iterator with the body response
(commonly this is just a list of strings, or just a list containing
one string that is the entire body.)
So, here's a simple application::
def app(environ, start_response):
start_response('200 OK', [('content-type', 'text/html')])
return ['Hello world!']
Well... that's unsatisfying. Sure, you can imagine what it does, but
you can't exactly point your web browser at it.
There's other cleaner ways to do this, but this tutorial isn't about
*clean* it's about *easy-to-understand*. So just add this to the
bottom of your file::
if __name__ == '__main__':
from paste import httpserver
httpserver.serve(app, host='127.0.0.1', port='8080')
Now visit http://localhost:8080 and you should see your new app.
If you want to understand how a WSGI server works, I'd recommend
looking at the `CGI WSGI server
<http://www.python.org/peps/pep-0333.html#the-server-gateway-side>`_
in the WSGI spec.
An Interactive App
------------------
That last app wasn't very interesting. Let's at least make it
interactive. To do that we'll give a form, and then parse the form
fields::
from paste.request import parse_formvars
def app(environ, start_response):
fields = parse_formvars(environ)
if environ['REQUEST_METHOD'] == 'POST':
start_response('200 OK', [('content-type', 'text/html')])
return ['Hello, ', fields['name'], '!']
else:
start_response('200 OK', [('content-type', 'text/html')])
return ['<form method="POST">Name: <input type="text" '
'name="name"><input type="submit"></form>']
The ``parse_formvars`` function just takes the WSGI environment and
calls the `cgi <http://python.org/doc/current/lib/module-cgi.html>`_
module (the ``FieldStorage`` class) and turns that into a MultiDict.
Now For a Framework
===================
Now, this probably feels a bit crude. After all, we're testing for
things like REQUEST_METHOD to handle more than one thing, and it's
unclear how you can have more than one page.
We want to build a framework, which is just a kind of generic
application. In this tutorial we'll implement an *object publisher*,
which is something you may have seen in Zope, Quixote, or CherryPy.
Object Publishing
-----------------
In a typical Python object publisher you translate ``/`` to ``.``. So
``/articles/view?id=5`` turns into ``root.articles.view(id=5)``. We
have to start with some root object, of course, which we'll pass in...
::
class ObjectPublisher(object):
def __init__(self, root):
self.root = root
def __call__(self, environ, start_response):
...
app = ObjectPublisher(my_root_object)
We override ``__call__`` to make instances of ``ObjectPublisher``
callable objects, just like a function, and just like WSGI
applications. Now all we have to do is translate that ``environ``
into the thing we are publishing, then call that thing, then turn the
response into what WSGI wants.
The Path
--------
WSGI puts the requested path into two variables: ``SCRIPT_NAME`` and
``PATH_INFO``. ``SCRIPT_NAME`` is everything that was used up
*getting here*. ``PATH_INFO`` is everything left over -- it's
the part the framework should be using to find the object. If you put
the two back together, you get the full path used to get to where we
are right now; this is very useful for generating correct URLs, and
we'll make sure we preserve this.
So here's how we might implement ``__call__``::
def __call__(self, environ, start_response):
fields = parse_formvars(environ)
obj = self.find_object(self.root, environ)
response_body = obj(**fields.mixed())
start_response('200 OK', [('content-type', 'text/html')])
return [response_body]
def find_object(self, obj, environ):
path_info = environ.get('PATH_INFO', '')
if not path_info or path_info == '/':
# We've arrived!
return obj
# PATH_INFO always starts with a /, so we'll get rid of it:
path_info = path_info.lstrip('/')
# Then split the path into the "next" chunk, and everything
# after it ("rest"):
parts = path_info.split('/', 1)
next = parts[0]
if len(parts) == 1:
rest = ''
else:
rest = '/' + parts[1]
# Hide private methods/attributes:
assert not next.startswith('_')
# Now we get the attribute; getattr(a, 'b') is equivalent
# to a.b...
next_obj = getattr(obj, next)
# Now fix up SCRIPT_NAME and PATH_INFO...
environ['SCRIPT_NAME'] += '/' + next
environ['PATH_INFO'] = rest
# and now parse the remaining part of the URL...
return self.find_object(next_obj, environ)
And that's it, we've got a framework.
Taking It For a Ride
--------------------
Now, let's write a little application. Put that ``ObjectPublisher``
class into a module ``objectpub``::
from objectpub import ObjectPublisher
class Root(object):
# The "index" method:
def __call__(self):
return '''
<form action="welcome">
Name: <input type="text" name="name">
<input type="submit">
</form>
'''
def welcome(self, name):
return 'Hello %s!' % name
app = ObjectPublisher(Root())
if __name__ == '__main__':
from paste import httpserver
httpserver.serve(app, host='127.0.0.1', port='8080')
Alright, done! Oh, wait. There's still some big missing features,
like how do you set headers? And instead of giving ``404 Not Found``
responses in some places, you'll just get an attribute error. We'll
fix those up in a later installment...
Give Me More!
-------------
You'll notice some things are missing here. Most specifically,
there's no way to set the output headers, and the information on the
request is a little slim.
::
# This is just a dictionary-like object that has case-
# insensitive keys:
from paste.response import HeaderDict
class Request(object):
def __init__(self, environ):
self.environ = environ
self.fields = parse_formvars(environ)
class Response(object):
def __init__(self):
self.headers = HeaderDict(
{'content-type': 'text/html'})
Now I'll teach you a little trick. We don't want to change the
signature of the methods. But we can't put the request and response
objects in normal global variables, because we want to be
thread-friendly, and all threads see the same global variables (even
if they are processing different requests).
But Python 2.4 introduced a concept of "thread-local values". That's
a value that just this one thread can see. This is in the
`threading.local <http://docs.python.org/lib/module-threading.html>`_
object. When you create an instance of ``local`` any attributes you
set on that object can only be seen by the thread you set them in. So
we'll attach the request and response objects here.
So, let's remind ourselves of what the ``__call__`` function looked
like::
class ObjectPublisher(object):
...
def __call__(self, environ, start_response):
fields = parse_formvars(environ)
obj = self.find_object(self.root, environ)
response_body = obj(**fields.mixed())
start_response('200 OK', [('content-type', 'text/html')])
return [response_body]
Lets's update that::
import threading
webinfo = threading.local()
class ObjectPublisher(object):
...
def __call__(self, environ, start_response):
webinfo.request = Request(environ)
webinfo.response = Response()
obj = self.find_object(self.root, environ)
response_body = obj(**dict(webinfo.request.fields))
start_response('200 OK', webinfo.response.headers.items())
return [response_body]
Now in our method we might do::
class Root:
def rss(self):
webinfo.response.headers['content-type'] = 'text/xml'
...
If we were being fancier we would do things like handle `cookies
<http://python.org/doc/current/lib/module-Cookie.html>`_ in these
objects. But we aren't going to do that now. You have a framework,
be happy!
WSGI Middleware
===============
`Middleware
<http://www.python.org/peps/pep-0333.html#middleware-components-that-play-both-sides>`_
is where people get a little intimidated by WSGI and Paste.
What is middleware? Middleware is software that serves as an
intermediary.
So lets
write one. We'll write an authentication middleware, so that you can
keep your greeting from being seen by just anyone.
Let's use HTTP authentication, which also can mystify people a bit.
HTTP authentication is fairly simple:
* When authentication is requires, we give a ``401 Authentication
Required`` status with a ``WWW-Authenticate: Basic realm="This
Realm"`` header
* The client then sends back a header ``Authorization: Basic
encoded_info``
* The "encoded_info" is a base-64 encoded version of
``username:password``
So how does this work? Well, we're writing "middleware", which means
we'll typically pass the request on to another application. We could
change the request, or change the response, but in this case sometimes
we *won't* pass the request on (like, when we need to give that 401
response).
To give an example of a really really simple middleware, here's one
that capitalizes the response::
class Capitalizer(object):
# We generally pass in the application to be wrapped to
# the middleware constructor:
def __init__(self, wrap_app):
self.wrap_app = wrap_app
def __call__(self, environ, start_response):
# We call the application we are wrapping with the
# same arguments we get...
response_iter = self.wrap_app(environ, start_response)
# then change the response...
response_string = ''.join(response_iter)
return [response_string.upper()]
Techically this isn't quite right, because there there's two ways to
return the response body, but we're skimming bits.
`paste.wsgilib.intercept_output
<http://pythonpaste.org/module-paste.wsgilib.html#intercept_output>`_
is a somewhat more thorough implementation of this.
.. note::
This, like a lot of parts of this (now fairly old) tutorial is
better, more thorough, and easier using `WebOb
<http://pythonpaste.org/webob/>`_. This particular example looks
like::
from webob import Request
class Capitalizer(object):
def __init__(self, app):
self.app = app
def __call__(self, environ, start_response):
req = Request(environ)
resp = req.get_response(self.app)
resp.body = resp.body.upper()
return resp(environ, start_response)
So here's some code that does something useful, authentication::
class AuthMiddleware(object):
def __init__(self, wrap_app):
self.wrap_app = wrap_app
def __call__(self, environ, start_response):
if not self.authorized(environ.get('HTTP_AUTHORIZATION')):
# Essentially self.auth_required is a WSGI application
# that only knows how to respond with 401...
return self.auth_required(environ, start_response)
# But if everything is okay, then pass everything through
# to the application we are wrapping...
return self.wrap_app(environ, start_response)
def authorized(self, auth_header):
if not auth_header:
# If they didn't give a header, they better login...
return False
# .split(None, 1) means split in two parts on whitespace:
auth_type, encoded_info = auth_header.split(None, 1)
assert auth_type.lower() == 'basic'
unencoded_info = encoded_info.decode('base64')
username, password = unencoded_info.split(':', 1)
return self.check_password(username, password)
def check_password(self, username, password):
# Not very high security authentication...
return username == password
def auth_required(self, environ, start_response):
start_response('401 Authentication Required',
[('Content-type', 'text/html'),
('WWW-Authenticate', 'Basic realm="this realm"')])
return ["""
<html>
<head><title>Authentication Required</title></head>
<body>
<h1>Authentication Required</h1>
If you can't get in, then stay out.
</body>
</html>"""]
.. note::
Again, here's the same thing with WebOb::
from webob import Request, Response
class AuthMiddleware(object):
def __init__(self, app):
self.app = app
def __call__(self, environ, start_response):
req = Request(environ)
if not self.authorized(req.headers['authorization']):
resp = self.auth_required(req)
else:
resp = self.app
return resp(environ, start_response)
def authorized(self, header):
if not header:
return False
auth_type, encoded = header.split(None, 1)
if not auth_type.lower() == 'basic':
return False
username, password = encoded.decode('base64').split(':', 1)
return self.check_password(username, password)
def check_password(self, username, password):
return username == password
def auth_required(self, req):
return Response(status=401, headers={'WWW-Authenticate': 'Basic realm="this realm"'},
body="""\
<html>
<head><title>Authentication Required</title></head>
<body>
<h1>Authentication Required</h1>
If you can't get in, then stay out.
</body>
</html>""")
So, how do we use this?
::
app = ObjectPublisher(Root())
wrapped_app = AuthMiddleware(app)
if __name__ == '__main__':
from paste import httpserver
httpserver.serve(wrapped_app, host='127.0.0.1', port='8080')
Now you have middleware! Hurrah!
Give Me More Middleware!
------------------------
It's even easier to use other people's middleware than to make your
own, because then you don't have to program. If you've been following
along, you've probably encountered a few exceptions, and have to look
at the console to see the exception reports. Let's make that a little
easier, and show the exceptions in the browser...
::
app = ObjectPublisher(Root())
wrapped_app = AuthMiddleware(app)
from paste.exceptions.errormiddleware import ErrorMiddleware
exc_wrapped_app = ErrorMiddleware(wrapped_app)
Easy! But let's make it *more* fancy...
::
app = ObjectPublisher(Root())
wrapped_app = AuthMiddleware(app)
from paste.evalexception import EvalException
exc_wrapped_app = EvalException(wrapped_app)
So go make an error now. And hit the little +'s. And type stuff in
to the boxes.
Conclusion
==========
Now that you've created your framework and application (I'm sure it's
much nicer than the one I've given so far). You might keep writing it
(many people have so far), but even if you don't you should be able to
recognize these components in other frameworks now, and you'll have a
better understanding how they probably work under the covers.
Also check out the version of this tutorial written `using WebOb
<http://pythonpaste.org/webob/do-it-yourself.html>`_. That tutorial
includes things like **testing** and **pattern-matching dispatch**
(instead of object publishing).