Getting Started with Paver

Often, the easiest way to get going with a new tool is to see an example in action, so that’s how we’ll get started with Paver. In the Paver distribution, there are samples under docs/samples. The Getting Started samples are in the “started” directory under there.

The Old Way

Our first sample is called “The Old Way” (and it’s in the docs/samples/started/oldway directory). It’s a fairly typical project with one Python package and some docs, and we want to be able to distribute it.

Python’s distutils makes it easy indeed to create a distributable package. We create a setup.py file that looks like this:

#<== include('started/oldway/setup.py')==>
#<==end==>

With that simple setup script, you can run:

python setup.py sdist

to build a source distribution:

# <==
# sh('cd docs/samples/started/oldway; python setup.py sdist',
#    insert_output=False)
# sh('ls -l docs/samples/started/oldway/dist')
# ==>
# <==end==>

Then your users can run the familiar:

python setup.py install

to install the package, or use setuptools’ even easier:

easy_install "TheOldWay"

for packages that are up on the Python Package Index.

The Old Way’s Docs

The Old Way project is at least a bit modern in that it uses Sphinx for documentation. When you use sphinx-quickstart to get going with your docs, Sphinx will give you a Makefile that you can run to generate your HTML docs. So, generating the HTML docs is easy:

make html

Except, in this project (as in Paver itself), we want to include the HTML files in a docs directory in the package for presenting help to the users. We end up creating a shell script to do this:

# <== include("started/oldway/builddocs.sh")==>
# <==end==>

Of course, creating a script like this means that we have to actually remember to run it. We could change this script to “buildsdist.sh” and add a python setup.py sdist to the end of the file. But, wouldn’t it be nicer if we could just use python setup.py sdist directly?

You can create new distutils commands, but do you really want to drop stuff like that in the distutils/command package in your Python library directory? And how would you call the sdist command anyway? setuptools helps, but it still requires setting up a module and entry point for this collection of commands.

Work with me here

I just know there are some people reading this and thinking “man, what a contrived example!”. Building, packaging, distributing and deploying of projects is quite custom for every project. Part of the point of Paver is to make it easy to handle whatever weird requirements arise in your project. This example may seem contrived, but it should give you an idea of how easy Paver makes it to get your tasks done.

The New Way

Let’s bring in Paver now to clean up our scripting a bit. Converting a project to use Paver is really, really simple. Recall the setup function from our Old Way setup.py:

# <== include("started/oldway/setup.py", "setup")==>
# <==end==>

Getting Started with Paver

setup.py is a standard Python script. It’s just called setup.py as a convention. Paver works a bit more like Make or Rake. To use Paver, you run paver <taskname> and the paver command will look for a pavement.py file in the current directory. pavement.py is a standard Python module. A typical pavement will import from paver.easy to get a bunch of convenience functions and objects and then import other modules that include useful tasks:

# <== include('started/newway/pavement.py', 'imports')==>
# <==end==>

Converting from setup.py to pavement.py is easy. Paver provides a special options object that holds all of your build options. options is just a dictionary that allows attribute-style access and has some special searching abilities. The options for distutils operations are stored in a setup section of the options. And, as a convenience, Paver provides a setup function that sets the values in that options section (and goes a step further, by making all of the distutils/setuptools commands available as Paver tasks). Here’s what the conversion looks like:

# <== include('started/newway/pavement.py', 'setup')==>
# <==end==>

Paver is compatible with distutils

Choosing to use Paver does not mean giving up on distutils or setuptools. Paver lets you continue to use distutils and setuptools commands. When you import a module that has Paver tasks in it, those tasks automatically become available for running. If you want access to distutils and setuptools commands as well, you can either use the paver.setuputils.setup function as described above, or call paver.setuputils.install_distutils_tasks().

We can see this in action by looking at paver help:

# <== sh('cd docs/samples/started/newway; paver help')==>
# <==end==>

That command is listing all of the available tasks, and you can see near the top there are tasks from distutils.command. All of the standard distutils commands are available.

There’s one more thing we need to do before our Python package is properly redistributable: tell distutils about our special files. We can do that with a simple MANIFEST.in:

# <== include('started/newway/MANIFEST.in')==>
# <==end==>

With that, we can run paver sdist and end up with the equivalent output file:

# <==
# sh('cd docs/samples/started/newway; paver sdist',
#    insert_output=False)
# sh('ls -l docs/samples/started/newway/dist')
# ==>
# <==end==>

It also means that users of The New Way can also run paver install to install the package on their system. Neat.

But people are used to setup.py!

python setup.py install has been around a long time. And while you could certainly put a README file in your package telling people to run paver install, we all know that no one actually reads docs. (Hey, thanks for taking the time to read this!)

No worries, though. You can run paver generate_setup to get a setup.py file that you can ship in your tarball. Then your users can run python setup.py install just like they’re used to, and Paver will take over.

But people don’t have Paver yet!

There are millions of Python installations that don’t have Paver yet, but have Python and distutils. How can they run a Paver-based install?

Easy, you just run paver minilib and you will get a file called paver-minilib.zip. That file has enough of Paver to allow someone to install most projects. The Paver-generated setup.py knows to look for that file and use it if it sees it.

Worried about bloating your package? The paver-minilib is not large:

# <==
# sh('cd docs/samples/started/newway ; paver minilib',
#    insert_output=False)
# sh('ls -l docs/samples/started/newway/paver-minilib.zip')
# ==>
# <==end==>

Paver itself is bootstrapped with a generated setup file and a paver-minilib.

Hey! Didn’t you just create more work for me?

You might have noticed that we now have three commands to run in order to get a proper distribution for The New Way. Well, you can actually run them all at once: paver generate_setup minilib sdist. That’s not terrible, but it’s also not great. You don’t want to end up with a broken distribution just because you forgot one of the tasks.

By design, one of the easiest things to do in Paver is to extend the behavior of an existing “task”, and that includes distutils commands. All we need to do is create a new sdist task in our pavement.py:

# <== include('started/newway/pavement.py', 'sdist')==>
# <==end==>

The @task decorator just tells Paver that this is a task and not just a function. The @needs decorator specifies other tasks that should run before this one. You can also use the call_task(taskname) function within your task if you wish. The function name determines the name of the task. The docstring is what shows up in Paver’s help output.

With that task in our pavement.py, paver sdist is all it takes to build a source distribution after generating a setup file and minilib.

Note

If you are depending on distutils task (via @needs), you have to call setup() before task is defined. Under the hood, setup call installs distutils/setupsools task and make them available, so do not make it conditional.

Tackling the Docs

Until the tools themselves provide tasks and functions that make creating pavements easier, Paver’s Standard Library will include a collection of modules that help out for commonly used tools. Sphinx is one package for which Paver has built-in support.

To use Paver’s Sphinx support, you need to have Sphinx installed and, in your pavement.py, import paver.doctools. Just performing the import will make the doctools-related tasks available. paver help html will tell us how to use the html command:

# <== sh('paver help paver.doctools.html')==>
# <==end==>

According to that, we’ll need to set the builddir setting, since we’re using a builddir called “_build”. Let’s add this to our pavement.py:

# <== include('started/newway/pavement.py', 'sphinx')==>
# <==end==>

And with that, paver html is now equivalent to make html using the Makefile that Sphinx gave us.

Getting rid of our docs shell script

You may remember that shell script we had for moving our generated docs to the right place:

# <== include('started/oldway/builddocs.sh')==>
# <==end==>

Ideally, we’d want this to happen whenever we generate the docs. We’ve already seen how to override tasks, so let’s try that out here:

# <== include('started/newway/pavement.py', 'html')==>
# <==end==>

There are a handful of interesting things in here. The equivalent of ‘make html’ is the @needs(‘paver.doctools.html’), since that’s the task we’re overriding.

Inside our task, we’re using “path”. This is a customized version of Jason Orendorff’s path module. All kinds of file and directory operations become super-simple using this module.

We start by deleting our destination directory, since we’ll be copying new generated files into that spot. Next, we look at the built docs directory that we’ll be moving:

# <== include('started/newway/pavement.py', 'html.builtdocs')==>
# <==end==>

One cool thing about path objects is that you can use the natural and comfortable ‘/’ operator to build up your paths.

The next thing we see here is the accessing of options. The options object is available to your tasks. It’s basically a dictionary that offers attribute-style access and can search for variables (which is why you can type options.builddir instead of the longer options.sphinx.builddir). That property of options is also convenient for being able to share properties between sections.

And with that, we eliminate the shell script as a separate file.

Fixing another wart in The Old Way

In the documentation for The Old Way, we actually included the function body directly in the docs. But, we had to cut and paste it there. Sphinx does offer a way to include an external file in your documentation. Paver includes a better way.

There are a couple of parts to the documentation problem:

  1. It’s good to have your code in separate files from your docs so that the code can be complete, runnable and, above all, testable programs so that you can be sure that everything works.
  2. You want your writing and the samples included with your writing to stand up as reasonable, coherent documents. Python’s doctest style does not always lend itself to coherent documents.
  3. It’s nice to have the code sample that you’re writing about included inline with the documents as you’re writing them. It’s easier to write when you can easily see what you’re writing about.

#1 and #3 sound mutually exclusive, but they’re not. Paver has a two part strategy to solve this problem. Let’s look at part of the index.rst document file to see the first part:

# <== include("started/newway/docs/index.rst", "mainpart")==>
# <==end==>

In The New Way’s index.rst, you can see the same mechanism being used that is used in this Getting Started guide. Paver includes Ned Batchelder’s Cog package. Cog lets you drop snippets of Python into a file and have those snippets generate stuff that goes into the file. Unlike a template language, Cog is designed so that you can leave the markers in and regenerate as often as you need to. With a template language, you have the template and the finalized output, but not a file that has both.

So, as I’m writing this Getting Started document, I can glance up and see the index.rst contents right inline. You’ll notice The # [[ [cog part in there is calling an include() function. This is the second part offered by Paver. Paver lets you specify an “includedir” for use with Cog. This lets you include files relative to that directory. And, critically, it also lets you mark off sections of those files so that you can easily include just the part you want. In the example above, we’re picking up the ‘code’ section of the newway/thecode.py file. Let’s take a look at that file:

# <== sh("cat docs/samples/started/newway/newway/thecode.py") ==>
# <==end==>

Paver has a Cog-like syntax for defining named sections. So, you just use the include function with the relative filename and the section you want, and it will be included. Sections can even be nested (and you refer to nested sections using familiar dotted notation).

Bonus Deployment Example

pavements are just standard Python. The syntax for looping and things like that are just what you’re used to. The options are standard Python so they can contain lists and other objects. Need to deploy to multiple hosts? Just put the hosts in the options and loop over them.

Let’s say we want to deploy The New Way project’s HTML files to a couple of servers. This is similar to what I do for Paver itself, though I only have one server. First, we’ll set up some variables to use for our deploy task:

# <== include('started/newway/pavement.py', 'deployoptions')==>
# <==end==>

As you can see, we can put whatever kinds of objects we wish into the options. Now for the deploy task itself:

# <== include("started/newway/pavement.py", "deploy")==>
# <==end==>

You’ll notice the new “cmdopts” decorator. Let’s say that you have sensitive information like a password that you don’t want to include in your pavement. You can easily make it a command line option for that task using cmdopts. options.deploy.username will be set to whatever the user enters on the command line.

It’s also worth noting that when looking up options, Paver gives priority to options in a section with the same name as the task. So, options.username will prefer options.deploy.username even if there is a username in another section.

Our deploy task uses a simple for loop to run an rsync command for each host. Let’s do a dry run providing a username to see what the commands will be:

# <== sh("cd docs/samples/started/newway; paver -n deploy -u kevin")==>
# <==end==>

Where to go from here

The first thing to do is to just get started using Paver. As you’ve seen above, it’s easy to get Paver into your workflow, even with existing projects.

Use the paver help command.

If you really want more detail now, you’ll want to read more about pavement files and the Paver Standard Library.