Chapter 15: Introduction to Pylons

While Django is currently the most popular webframework for Python, it is by no means your only choice. Where Django grew out of the needs of newsrooms to implement content management solutions rapidly - Pylons grew out of a need to build web applications in environments that may have existing databases to integrate with and the applications don’t fit neatly into the class of applications that are loosely defined in the “content management” space.

Pylons greatest strength is that it takes a best of breed approach to constructing it’s technology stack. Where everything is ‘built in’ with Django and the entire application stack is specifically designed with a single world view of how applications should be done - Pylons takes precisely the opposite approach. Pylons - the core codebase that lives in the ‘pylons’ namespace - it’s remarkably small. With the 0.9.7 release, it’s hovering around 5,500 lines of code. Django by comparison weighs in at about 125,000 lines of code.

Pylons manages to do this magic by leveraging existing libraries extensively and the Pylons community works with many other Python projects to develop standard APIs to promote interoperability.

Ultimately picking Django or Pylons is about deciding which tradeoffs you’re willing to make. While Django is extremly easy to learn because all the documentation is in one place and all the documentation relating to any particular component is always discussed in the context of building a web application - you lose some flexibility when you need to start doing things that are at the margins of what Django was designed for.

For example, in a project I’ve worked on recently, I needed to interact with a nontrivial database that was implemented in SQL Server 2000. For Django, implementing the SQL Server backend was quite difficult. There aren’t that many web developers using Django on Windows, never mind SQL Server. While the Django ORM is a part of Django, it is also not the core focus of Django. Supporting arbitrary databases is simply not a goal for Django and rightly so.

The Django project assumes you’ll do a rational thing and simply use Postgresql as your database backend. Web developers have better things to do than build backend drivers for every possible database.

Pylons on the other hand leverages SQLAlchemy. SQLAlchemy is probably the most powerful database toolkit available in Python. It only focuses on database access. The SQL Server backend was already built in a robust way for CPython and implementing the extra code for a Jython backend took 2 days - and this was from not seeing any of the code in SQLAlchemy’s internals.

That experience alone sold me on Pylons. I don’t have to rely the ‘webframework’ people to experts in databases. Similarily, I don’t have to rely on the database experts to know anything about web templating.

In short when you have to deal with the weird stuff - Pylons makes a fabulous choice - and lets be honest - there’s almost always weird stuff you’re going to have to do.

A guide for the impatient

The best way to install Pylons is inside of a virtualenv. Create a new virtualenv for Jython and run easy_install

> easy_install "Pylons==0.9.7"

Create your application

> paster create --template=pylons RosterTool

# TODO: just use the defaults for everything.  No sqlalchemy

Launch the development server

> paster serve --reload development.ini

Open a browser and connect to http://127.0.0.1:5000/

# TODO: include screenshot here

Drop a static file into rostertool/public/welcome.html

<html>
    <body>Just a static file</body>
</html>

You should now be able to load the static content by going to

http://localhost:5000/welcome.html

Add a controller

RosterTool/roster > paster controller roster

Paste will install a directory named “controllers” and install some files in there including a module named “roster.py”. You can open it up and you’ll see a class named “RosterController” and it will have a single method “index”. Pylons is smart enough to automatically map a URL to a controller classname and invoke a method. To invoke the RosterController’s index method, you just need to invoke

http://localhost:5000/roster/index

Congratulations, you’ve got your most basic possible web application running now. It handles basic HTTP GET requests and calls a method on a controller and a response comes out. Lets cover each of these pieces in detail now.

A note about Paste

While you setup your toy Pylons application, you probably wondered why Pylons seems to use a command line tool called “paster” instead of something obvious like “pylons”. Paster is actually a part of the Paste set of tools that Pylons uses.

Paste is used to build web application frameworks - not web applications - but web application frameworks like Pylons. Everytime you use “paster”, that’s Paste being called. Everytime you access the HTTTP request and response objects - that’s WebOb - a descendant of Paste’s HTTP wrapper code. Pylons uses Paste extensively for configuration management, testing, basic HTTP handling with WebOb. You would do well to at least skim over the Paste documentation to see what is available in paste.

Pylons MVC

Pylons, like Django any any reasonably sane webframework (or GUI toolkit for that matter) uses the model-view-controller design pattern.

In Pylons this maps to:

Component Implementation
Model SQLAlchemy (or any other database toolkit you prefer)
View Mako (or any templating language you prefer)
Controller Plain Python code

To reiterate - Pylons is about letting you - the application developer decide on the particular tradeoffs you’re willing to make. If using a templating language more similar to Django is better for your web designers, then switch go Jinja2. If you don’t really want to deal with SQLAlchemy - you can use SQLObject or raw SQL if you prefer.

Pylons provides some tools to help you hook these pieces together in a rational way.

Routes is a library that maps URLs to classes. This is your basic mechanism for dispatching methods whenever your webserver is hit. Routes provides similar functionality to what Django’s URL dispatcher provides.

Webhelpers is the defacto standard library for Pylons. It contains commonly used functions for the web like flashing status messages to users, date conversion functions, HTML tag generation, pagination functions, text processing - the list goes on.

Pylons also provides infrastructure so that you can manipulate things that are particular to web applications including:

  • WSGI middleware to add functionality to your application with minimal intrusion into your existing codebase.
  • A robust testing framework including a shockingly good debugger you can use through the web.
  • Helpers to enable REST-ful API development so you can expose your application as a programmatic interface.

Now let’s wrap up the hockey roster up in a web application. We’ll target a couple features:

  • form handling and validation to add new players through the web
  • login and authentication to make sure not anybody can edit our lists
  • add a JSON/REST api so that we can modify data from other tools

In the process, we’ll use the interactive debugger from both command line and through the web to directly observe and interact with the state of the running application.

An interlude into Java’s memory model

A note about reloading - sometimes if you’re doing devleopment with Pylons on Jython, Java will through an OutOfMemory error like this

java.lang.OutOfMemoryError: PermGen space
        at java.lang.ClassLoader.defineClass1(Native Method)
        at java.lang.ClassLoader.defineClass(ClassLoader.java:620)

Java keeps track of class definitions in something called the Permanent Generation heap space. This is a problem for Pylons when the HTTP threads are restarted and your classes are reloaded. The old class definitions don’t go away - they never get garbage collected.. Since Jython is dynamically creating Java classes behind the scenes, each time your develpment server restarts - you’re potentially getting hundreds of new classes loaded into the JVM.

Repeat this several times and it doesn’t take long until your JVM has run out of permgen space and it keels over and dies.

To modify the permgen heap size, you’ll need to instruct Java using some extended command line options. To set the heap to 128M, you’ll need to use “-XX:MaxPermSize=128M”.

To get this behavior by default for Jython, you’ll want to edit your Jython startup script in JYTHON_HOME/bin/jython (or jython.bat) by editting the line that reads

set _JAVA_OPTS=

to be

set _JAVA_OPTS=-XX:MaxPermSize=128M

This shouldn’t be a problem in production environments where you’re not generating new class definitions during runtime, but it can be quite frustratig during development.

Invoking the Pylons shell

Yes, I’m going to start with testing right away because it will provide you with a way to explore the Pylons application in an interactive way.

Pylons gives you an interactive shell much like Django’s. You can start it up with the following commands.

RosterTool > jython setup.py egg_info
RosterTool > paster shell test.ini

This will yield a nice interactive shell you can start playing with right away. Now lets take a look at those request and response objects in our toy application.

RosterTool > paster shell test.ini

Pylons Interactive Shell
Jython 2.5.0 (Release_2_5_0:6476, Jun 16 2009, 13:33:26)
[OpenJDK Server VM (Sun Microsystems Inc.)]

All objects from rostertool.lib.base are available
Additional Objects:
mapper     -  Routes mapper object
wsgiapp    -  This project's WSGI App instance
app        -  paste.fixture wrapped around wsgiapp

>>> resp = app.get('/roster/index')
>>> resp
<Response 200 OK 'Hello World'>
>>> resp.req
<Request at 0x43 GET http://localhost/roster/index>

Pylons lets you actually run requests against the application and play with the resulting response. Even for something as ‘simple’ as the HTTP request and response,, Pylons uses a library to provide convenience methods and attributes to make your development life easier. In this case - it’s WebOb - a derivative of Paste’s older HTTP wrapper code.

The request and the response objects both have literally dozens of attributes and methods that are provided by the framework. You’ almost certainly going to benefit if you take time to browse through WebOb’s documentation.

Here’s four attributes you really have to know to make sense of the request object. The best thing to do is to try playing with the request object in the shell.

request.GET

GET is a special dictionary of the variables that were passed in the URL. Pylons automatically converts URL arguments that appear multiple times into discrete key value pairs.

>>> resp = app.get('/roster/index?foo=bar&x=42&x=50')
>>> resp.req.GET
UnicodeMultiDict([('foo', u'bar'), ('x', u'42'), ('x', u'50')])
>>> req.GET['x']
u'50'
>>> req.GET.getall('x')
[u'42', u'50']

Note how you can get either the last value or the list of values depending on how you choose to fetch values from the dictionary. This can cause subtle bugs if you’re not paying attention.

request.POST
POST is similar to GET, but appropriatley - it only returns the variables that were sent up during an HTTP POST submisssion.
request.params
Pylons merges all the GET and POST data into a single MultiValueDict. In almost all cases, this is the one attribute that you really want to use to get the data that the user sent to the server.
request.headers
This dictionary provides all the HTTP headers that the client sent to the server.

Context Variables and Application Globals

Most webframeworks provide a request scoped variable to act as a bag of values. Pylons is no exception - whenever you create a new controller with paste - it will automatically import an attribute ‘c’ which is the context variable.

This is one aspect of Pylons which I’ve found to be frustrating. The ‘c’ attribute is code generated as an import when you instruct paste to bulid you a new controller. The ‘c’ value is not an attribute of your controller - Pylons has special global threadsafe variables - this is just one of them. You can store variables that you want to exist for the duration of the request in the context. These values won’t persist after the request/response cycle has completed so don’t confuse this with the session variable.

The other global variable you’ll end up using a lot is pylons.session. This is where you’ll store variables that need to persist over the course of several request/response cycles. You can treat this variable as aa special dictionary - just use standard Jython dictionary syntax and Pylons will handle the rest.

Routes

Routes is much like Django’s URL dispatcher. It provides a mechanism for you to map URLs to controllers classes and methods to invoke.

Generally, I find that Routes makes a tradeoff of less URL matching expressiveness in exchange for simpler reasoning about which URLs are directed to a particular controller and method. Routes doesn’t support regular expressions, just simple variable substitution.

A typical route will look something like this

map.connect('/{mycontroller}/{someaction}/{var1}/{var2}')

The above route would find the controller called “Mycontroller” (note the casing of the class) and invoke the “someaction” method on that object. Variables var1 and var2 woulld be passed in as arguments.

The connect() method of the map object will also take in optional arguments to fill in default values for URLs that do not have enough URL encoded data in them to properly invoke a method with the minimum required number of arguments. The front page is an example of this - let’s try connecting the frontpage to the Roster.index method.

Edit rostertool/config/routing.py so that there are 3 lines after #CUSTOM_ROUTES_HERE that should read something like this

map.connect('/', controller='roster', action='index')
map.connect('/{action}/{id}/', controller='roster')
map.connect('/add_player/', controller='roster', action='add_player')

While this looks like it should work, you can try running “paster serve”, it won’t.

Pylons always tries to serve static content before searching for controllers and methods to invoke. You’ll need to go to RosterTool/rostertool/public and delete the ‘index.html’ file that paster installed when you first created your application.

Load http://localhost:5000/ again in your browser - the default index.html should be gone and you should now get your welcome page.

Controllers and Templates

Leveraging off of the Table model we defined in chapter 12, let’s create the hockey roster, but this time using the postgresql database. I’ll assume that you have a postgresql installation running that allows you create new databases.

>>> from sqlalchemy import create_engine
>>> from sqlalchemy.schema import Sequence
>>> db = create_engine('postgresql+zxjdbc://myuser:mypass@localhost:5432/mydb')
>>> connection = db.connect()
>>> metadata = MetaData()
>>> player = Table('player', metadata,
...     Column('id', Integer, primary_key=True),
...     Column('first', String(50)),
...     Column('last', String(50)),
...     Column('position', String(30)))
>>> metadata.create_all(engine)

Now let’s wire the data up to the controllers, display some data and get basic form handling working. We’re going to create a basic CRUD (create, read, update, delete) inteface to the sqlalchemy model. Because of space constraints, this HTML is going to be very basic, but you’ll get a taste of how things fit together.

Paste doesn’t justs generate a stub for your controller - it will also code generate an empty functional test case in rostertool/tests/functional/ as test_roster.py. We’ll visit testing shortly.

Controllers are really where the action occurs in Pylons. This is where your application will take data from the database and prepare it for a template to render it as HTML. Let’s put the list of all players on the front page of the site. We’ll implement a template to render the list of all players. Then, we’ll implement a method in the controller to override the index() method of Roster to use SQLAlchemy to load the records from disk and send them to the template.

Along the way, we’ll touch on template inheritance so that you can see how you can save keystrokes by subclassing your templates in Mako.

First, let’s create two templates, base.html and list_players.html in the rostertool/templates directory.

base.html

<html>
    <body>
        <div class="header">
            ${self.header()}
        </div>

        ${self.body()}
    </body>
</html>

<%def name="header()">
    <h1>${c.page_title}</h1>
    <% messages = h.flash.pop_messages() %>
    % if messages:
    <ul id="flash-messages">
        % for message in messages:
        <li>${message}</li>
        % endfor
    </ul>
    % endif
</%def>

list_players.html

<%inherit file="base.html" />
<table border="1">
    <tr>
        <th>Position</th><th>Last name</th><th>First name</th><th>Edit</th>
    </tr>
    % for player in c.players:
        ${makerow(player)}
    % endfor
</table>

<h2>Add a new player</h2>
${h.form(h.url_for(controller='roster', action='add_player'), method='POST')}
    ${h.text('first', 'First Name')} <br />
    ${h.text('last', 'Last Name')} <br />
    ${h.text('position', 'Position')} <br />
    ${h.submit('add_player', "Add Player")}
${h.end_form()}

<%def name="makerow(row)">
<tr>
    <td>${row.position}</td>\
    <td>${row.last}</td>\
    <td>${row.first}</td>\
    <td><a href="${h.url_for(controller='roster', action='edit_player', id=row.id)}">Edit</a></td>\
</tr>
</%def>

There’s quite a bit going on here. The base template lets Mako define a boilerplate set of HTML that all pages can reuse. Each section is defined with a <%def name=”block()”> section and the blocks are overloaded in the subclassed templates. In effect - Mako lets your page templates look like objects with methods that can render subsections of your pages.

The list_players.html template has content that is immediately substituted into the self.body() method of the base template. The first part of our body uses our magic context variable ‘c’. Here - we’re iterating over each of the players in the database and rendering them into a table as a row. Note here that we can use the Mako method syntax to create a method called ‘makerow’ and invoke it directly within our template.

#XX: Aside for Mako Mako provides a rich set of functions for templating. I’m only going ot use the most basic parts of Mako - inheritance, variable substitution and loop iteration to get the toy application working. I strongly suggest you dive into the Mako documentation to discover features and get a better understanding of how to use the template library. ##

Next, we add in a small form to create new players. The trick here is to see that the form is being generated programmatically by helper functions. Pylons automatically imports YOURPROJECT/lib/helpers (in our case - rostertool.lib.helpers) as the ‘h’ variable in your template. The helpers module typically imports functions frmo parts of Pylons or a dependant library to allow access to those features from anywhere in the application. Although this seems like a violation of ‘separation of concerns’ - look at the template and see what it buys us? We get fully decoupled URLs from the particular controller and method that need to be invoked. The template uses a special routes function “url_for” to compute the URL that would have been mapped for a particular controller and method. The last part of our list_players.html file contains code to display alert messages.

Let’s take a look at our rostertool.lib.helpers module now

from routes import url_for
from webhelpers.html.tags import *
from webhelpers.pylonslib import Flash as _Flash

# Send alert messages back to the user
flash = _Flash()

Here, we’re importing the url_for function from routes to do our URL reversal computations. We import HTML tag generators from the main html.tags helper modules and we import Flash to provide alert messages for our pages. I’ll show you how flash messages are used when we cover the controller code in more detail in the next couple of pages.

Now, create a controller with paste (you’ve already done this if you were impatient at the beginning of the chapter)

$ cd ROSTERTOOL/rostertool
$ paster controller roster

RosterContoller should get a method very short method that reads

def index(self):
    session = Session()
    c.page_title = 'Player List'
    c.players = session.query(Player).all()
    return render('list_players.html')

This code is fairly straight forward, we are simply using a SQLAlchemy session to load all the Player objects from disk and assigning to the special context variable ‘c’. Pylons is then instructed to render the list_player.html file. Let’s take a look at that file now:

The context should be your default place to place values you want to pass to other parts of the application. Note that Pylons will automatically bind in URL values to the context so while you can grab the form values from self.form_result, you can also grab raw URL values from the context.

You should be able run the debug webserver now and you can get to the front page to load an empty list of players. Start up your debug webserver as you did at the beginning of this chapter and go to http://localhost:5000/ to se the page load with your list of players (currently an empty list).

Now we need to get to the meaty part where we can start create, edit and delete players. We’ll make sure that the inputs are at least minimally validated, errors are displayed to the user and that alert messages are properly populated.

First, we need a page that shows just a single player and provides buttons for edit and delete.

<%inherit file="base.html" />

<h2>Edit player</h2>
${h.form(h.url_for(controller='roster', action='save_player', id=c.player.id), method='POST')}
    ${h.hidden('id', c.player.id)} <br />
    ${h.text('first', c.player.first)} <br />
    ${h.text('last', c.player.last)} <br />
    ${h.text('position', c.player.position)} <br />
    ${h.submit('save_player', "Save Player")}
${h.end_form()}

${h.form(h.url_for(controller='roster', action='delete_player', id=c.player.id), method='POST')}
    ${h.hidden('id', c.player.id)} <br />
    ${h.hidden('first', c.player.first)} <br />
    ${h.hidden('last', c.player.last)} <br />
    ${h.hidden('position', c.player.position)} <br />
    ${h.submit('delete_player', "Delete Player")}
${h.end_form()}

This template assumes that there is a ‘player’ value assigned to the context and not suprisingly - it’s going to be a full blown instance of the Player object that we first saw in chapter 12. The helper functions let us define our HTML form using webhelper tag generation functions. This means you won’t have to worry about escaping characters or remember the particular details of the HTML attributes. The helper.tag functions will do sensible things by default.

I’ve setup the edit and delete forms to point to different URLs. You might want to ‘conserve’ URLs but having discrete URLs for each action has advantages - especially for debugging. You can trivially view which URLs are being hit on a webserver by reading log files. Seeing the same kind of behavior if the URLs are the same, but the behavior is dictated by some form value - well that’s a whole lot harder to debug. It’s also a lot harder to setup in your controllers because you need to dispatch the behavior on a per method level. Why not just have separate methods for separate behaviour - everybody will thank you for it when they need to debug your code in the future.

Before we create our controller methods for create, edit and delete - we’ll create a formencode schema to provide basic validation. Again - Pylons doesn’t provide validation behaviour - it just leverages another library to do so. In rostertool/controllers/roster.py

class PlayerForm(formencode.Schema):
    # You need the next line to drop the submit button values
    allow_extra_fields=True

    first = formencode.validators.String(not_empty=True)
    last = formencode.validators.String(not_empty=True)
    position = formencode.validators.String(not_empty=True)

This simply provides basic string verification on our inputs. Note how this doesn’t provide any hint as to what the HTML form looks like - or that it’s HTML at all. FormEncode can validate arbitrary Python dictionaries and return errors about them.

I’m just going to show you the add method, and the edit_player methods. You should try to implement the save_player and delete_player methods to make sure you understand what’s going on here.

from pylons.decorators import validate
from rostertool.model import Session, Player

@validate(schema=PlayerForm(), form='index', post_only=False, on_get=True)
def add_player(self):
    first = self.form_result['first']
    last = self.form_result['last']
    position = self.form_result['position']
    session = Session()
    if session.query(Player).filter_by(first=first, last=last).count() > 0:
        h.flash("Player already exists!")
        return h.redirect_to(controller='roster')
    player = Player(first, last, position)
    session.add(player)
    session.commit()
    return h.redirect_to(controller='roster', action='index')

def edit_player(self, id):
    session = Session()
    player = session.query(Player).filter_by(id=id).one()
    c.player = player
    return render('edit_player.html')

A couple of notes here. edit_player is passed in the ‘id’ attribute directly by Routes. In the edit_player method - ‘player’ is assigned to the context, but the context is never explicitly passed into the template renderer. Pylons is going to automatically take the attributes bound to the context and write them into template and render the HTML output.

With the add_player method, I’m using the validate decorator to enforce the inputs against the PlayerForm. In the case of error, the form attribute of the decorator is used to load an action against the current controller. In this case - ‘index’ - so the front page loads.

The SQLAlchemy code should be familiar to you if you have already gone through chapter 12. The lsat line of the add_player method is a redirect to prevent problems with hitting reload in the browser. Once all data manipulation has occured - the server redirects the client to a results page. In the case that a user hits reload on the result page - no data will be mutated.

Here’s the signatures of the remaining methods you’ll need to implement to make things work:

  • save_player(self):
  • delete_player(self):

If you get stuck, you can always consult the working sample code on the book website.

Adding in a JSON API

JSON integration into Pylons is very straight forward. The steps are roughly the same as adding controller methods for plain HTML views. You invoke paste, paste then generates your controller stubs and test stubs, you add in some routes to wire controllers to URLs and then you just fill in the controller code.

$ cd ROSTERTOOL_HOME/rostertool
$ paster controller api

Pylons provides a special @jsonify decorator which will automatically convert Python primitive types into JSON objects. It will not convert the POST data into an object though - that’s your responsibility. Adding a simple read interface into the player list requires only adding a single method to your ApiController

@jsonify
def players(self):
    session = Session()
    players = [{'first': p.first,
                'last': p.last,
                'position': p.position,
                'id': p.id} for p in session.query(Player).all()]
    return players

adding a hook so that people can POST data to your server in JSON format to create new player is almost as easy

import simplejson as json

@jsonify
def add_player(self):
    obj = json.loads(request.body)
    schema = PlayerForm()
    try:
        form_result = schema.to_python(obj)
    except formencode.Invalid, error:
        response.content_type = 'text/plain'
        return 'Invalid: '+unicode(error)
    else:
        session = Session()
        first, last, position = obj['first'], obj['last'], obj['position']
        if session.query(Player).filter_by(last=last, first=first,
                position=position).count() == 0:
            session.add(Player(first, last, position))
            session.commit()
            return {'result': 'OK'}
        else:
            return {'result':'fail', 'msg': 'Player already exists'}

Unit testing, Functional Testing and Logging

One of my favourite features in Pylons is its rich set of testing, and debugging. It even manages to take social networking, turn it upside down and make it into a debugger feature. We’ll get to that shortly.

The first step to knowing how to test code in pylons is to familiarize yourself with the nose testing framework. nose makes testing simple by getting out of your way. There are no classes to subclass, just start writing functions that start with the word ‘test’ and nose will run them. Write a class that has “Test” prefixed in the name and nose wll treat it as a suite of tests running each method that starts with ‘test’. For each test method, nose will execute the setup() method just prior to executing your test and nose will execute the teardown() method after your test case.

Best of all, nose will automatically huntdown anything that looks like a test and will run it for you. There is no complicated chain of testcases you need to organize in a tree. The computer will do that for you.

Let’s take a look at your first testcase - we’ll just instrument the model, in this case - SQLAlchemy. Since the model layer has no dependency on Pylons - this effectivey - a test of just SQLAlchemy.

In ROSTERTOOL_HOME/rostertool/tests, create a module called “test_models.py” with the following content

from rostertool.model import Player, Session, engine

class TestModels(object):

    def setup(self):
        self.cleanup()

    def teardown(self):
        self.cleanup()

    def cleanup(self):
        session = Session()
        for player in session.query(Player):
            session.delete(player)
        session.commit()

    def test_create_player(self):
        session = Session()
        player1 = Player('Josh', 'Juneau', 'forward')
        player2 = Player('Jim', 'Baker', 'forward')
        session.add(player1)
        session.add(player2)

        # But 2 are in the session, but not in the database
        assert 2 == session.query(Player).count()
        assert 0 == engine.execute("select count(id) from player").fetchone()[0]
        session.commit()

        # Check that 2 records are all in the database
        assert 2 == session.query(Player).count()
        assert 2 == engine.execute("select count(id) from player").fetchone()[0]

Before we can run the tests, we’ll need to edit the model module a little so that the models know to lookup the connection URL from Pylon’s configuration file. In your test.ini, add a line setting the sqlalchemy.url setting to point to your database in the [app:main] section.

You should have a line that looks something like this

[app:main]
use = config:development.ini
sqlalchemy.url = postgresql+zxjdbc://username:password@localhost:5432/mydb

Now edit the model file so that the create_engine call uses that configuration. This is as simple as importing config from pylons and doing a dictionary lookup. The two lines you want are

from pylons import config
engine = create_engine(config['sqlalchemy.url'])

and that’s it. Your model will now lookup your database connection string from Pylons. Even better - nose will know how to use that configuration as well.

From the command line, you can run the tests from ROSTERTOOL_HOME like this now

ROSTERTOOL_HOME $ nosetests rostertool/tests/test_models.py
.
----------------------------------------------------------------------
Ran 1 test in 0.502s

Perfect! To capture stdout and get verbose output, you can choose to use the ‘-sv’ option. Nose has it’s own active community of developers. You can get plugins to do coverage analysis and performance profiling with some of the plugins. Use “nosetests –help” for a list of the options available for a complete list.

Due to the nature of Pylons and it’s pathologically decoupled design, writing small unit tests to test each little piece of code is very easy. Feel free to assemble your tests any which way you want. Just want to have a bunch of test functoins? Great! If you need to have setup and teardown and writing a test class makes sense - then do so.

Testing with nose is a joy - you aren’t forced to fit into any particular structure with respect to where you tests must go so that they will be executed. You can organize your tests in a way that makes the most sense to you.

That covers basic unit testing, but suppose we want to test the JSON interface to our hockey roster. We really want to be able to invoke GET and POSt on the URLs to make sure that URL routing is working as we expect. We want to make sure that the content-type is properly set to ‘application/x-json’. In other words - we want to have a proper functional test - a test that’s not as fine grained as a unit test.

The prior exposure to the ‘app’ object when we ran the paste shell should give you a rough idea of what is required. In Pylons, you can instrument your application code by using a TestController. Lucky for you, Pylons has already create one for you in your <app>/tests directory. Just import it, subclass it and you can start using the ‘app’ object just like you did inside of the shell.

Lets take a look at a functional test in detail now. Here’s a sample you cna save into rostertool/tests/functional/test_api.py

from rostertool.tests import *
import simplejson as json
from rostertool.model.models import Session, Player

class TestApiController(TestController):
    # Note that we're using subclasses of unittest.TestCase so we need
    # to be careful with setup/teardown camelcasing unlike nose's
    # default behavior

    def setUp(self):
        session = Session()
        for player in session.query(Player):
            session.delete(player)
        session.commit()

    def test_add_player(self):
        data = json.dumps({'first': 'Victor',
            'last': 'Ng',
            'position': 'Goalie'})
        # Note that the content-type is set in the headers to make
        # sure that paste.test doesn't URL encode our data
        response = self.app.post(url(controller='api', action='add_player'),
            params=data,
            headers={'content-type': 'application/x-json'})
        obj = json.loads(response.body)
        assert obj['result'] == 'OK'

        # Do it again and fail
        response = self.app.post(url(controller='api', action='add_player'),
            params=data,
            headers={'content-type': 'application/x-json'})
        obj = json.loads(response.body)
        assert obj['result'] <> 'OK'

There’s a minor detail which you can easily miss when you’re using the TestController as your superclass. First off, TestController is a descendant of unittest.TestCase frmo the standard python unit test library. Nose will not run ‘setup’ and ‘teardown’ methods on TestCase subclasses. Instead, you’ll have to use the camel case names that TestCase uses.

Reading through the testcase should show you how much detail you can be exposed. All your headers are exposed, the response content is exposed - indeed the HTTP response is completely exposed as an object for you to inspect and verify.

So great, now we can run small unit tests, bigger functional tests - lets’s take a look at the debugging facilities provided through the web.

Consider what happens with most web application stacks when an error occurs. Maybe you get a stack trace, maybe you don’t. If you’re lucky, you can see the local variables at each stack frame like Django does. Usually though, you’re out of luck if you want to interact with the live application as thee error is occuring.

Eventually, you may locate the part of the stack trace that triggered the error, but the only way of sharing that information is through either the mailing lists or by doing a formal patch against source control. Let’s take a look at an example of that.

We’re going to startup our application in development mode. We’re also going to intentionally break some code in the controller to see the stack trace. But first, we’ll need to put some data into our app. run

Add a sqlalchemy.url configuration line as you did in the test.ini configuration, and let’s startup the application in development mode. We’re going to have the server run so that any code changes on the file system are automatically detected and the code is reloaded

$ paster serve development.ini --reload

We’ll add a single player “John Doe” as a center, and save the record

# TODO: insert screenshot of the add user interface

Now let’s intentionally break some code to trigger the debugger. Modify the RosterController’s index method and edit the call that loads the list of players. We’ll use the web session instead of the database session to try loading the Player objects.

def index(self):
    db_session = Session()
    c.page_title = 'Player List'
    c.players = session.query(Player).all()
    return render('list_players.html')

Load http://localhost:5000/ to see the error page. You should see something like this

# XXX: insert screen capture of the error page listing
'AttributeError: Session object hsa no attribute 'not_a_method'

There’s a lot of information that Pylons throws back at you. Along the top of the screen, you’ll see 4 tabs: Traceback, Extra Data, Template and Source - Pylons will have put you in the Traceback tab by default to start with. If you look at the error, you’ll see the exact line number in the source file that the error occured in. What’s special about Pylons traceback tab is that this is actually a fully interactive session.

You can select the “+” signs to expand each stackframe and a text input along with some local variables on that frame will be revealed. That text input is an interface into your server process. You can type virtually any python command into it, hit enter and you will get back live results. From here, we can see that we should have used the ‘db_session’ and not the ‘session’ variable.

Inspecting the application stack

This is pretty fantastic. If you click on the ‘view’ link, you can even jump to the full source listing of the Jython module that caused the error. One bug in Pylons at the time of writing is that sometimes, the hyper link is malformed. So while the traceback will correctly list the line number that the error occured at, the source listing may go to the wrong line.

The Pylons developers have also embedded an interface into search engines to see if your error has been previously reported. If you scroll down to the bottom of your traceback page, you’ll see another tab control with a ‘Search Mail Lists’ option. Here, Pylons will automatically extract the exception message and provide you an interface so you can literally search all the mailing lists that are relevant to your particular Pylons installation.

If you can’t find your error on the mailing lists, you can go to the next tab “Post traceback” and submit your stacktrace to a webservice on PylonsHQ.com so that you can try to debug your problems online with other collaborators. Combining unit tests, functional tests, and the myriad of debugging options afforded to you in the web debugger - Pylons makes the debugging experience as painless as possible.

Deployment into a servlet container

Deploying your pylons application into a servlet container is very straight forward. Just install snakefight from PyPI using using easy_install and you can start building WAR files.

$ easy_install snakefight
...snakefight will download and install here ...
$ jython setup.py bdist_war --paste-config test.ini

By default, snakefight will bundle a complete instance of your Jython installation into the WAR file. What it doesn’t include is any JAR files that your application depends on. For our small example, this is just the postgresql JDBC driver. You can use the –include-jars options and provide a comma separated list of JAR files.

$ jython setup.py bdist_war \
    --include-jars=postgresql-8.3-604.jdbc4.jar \
    --paste-config=test.ini

The final WAR file will be located under the dist directory. It will contain your postgreql JDBC driver, a complete installation of Jython including anything located in site-packages and your application. Your war file should deploy without any issues into any standards compliant servlet container.

Conclusion

We’ve only scratched the surface of what’s possible with Pylons, but I hope you’ve gotten a taste of what is possible with Pylons. Pylons uses a large number of packages so you will need to spend more time getting over the initial learning curve, but the dividend is the ability to pick and choose the libraries that best solve your particular problems.