I've created what's called a "buildout" for lxml, and I figured I'd let everybody know how cool that is.

Buildout is a new system for constructing Python applications and its constituent parts, based on setuptools and eggs. Much more information can be found here:

http://cheeseshop.python.org/pypi/zc.buildout/

Now before you read all that (it's pretty technical), let me explain what problem the lxml buildout tries to solve.

Recently I tried using lxml's XMLSchema validator against an XML file and a schema. It gave a validation error, but to the best of my knowledge, the XML file should've been valid. Thinking it might perhaps be a bug that was fixed in more recent versions of libxml2, I set out to build and install libxml2 and libxslt locally, compiled lxml against it, and confirmed it: the document does validate against the schema with the most recent release of libxml2. This was indeed a bug in libxml2.

Now downloading, building and installing libxml2 and libxslt and linking lxml against it is quite a bit of work. Buildout to the rescue! Now it's just this:

$ svn co https://infrae.com/svn/buildout/lxml-recent/trunk lxml-recent
$ cd lxml-recent
$ python bootstrap/bootstrap.py
$ bin/buildout

and then wait. It'll download, configure and compile libxml2, libxslt, will downoad the lxml source and compile it against the libraries.

In the end you end up with a script that invokes the Python interpreter with the right paths set up so it can import lxml. Here's how to start it:

$ bin/lxmlpython

Though you can also manipulate the PYTHONPATH yourself in your scripts to put the lxml development egg on it, and then you can use the regular Python you used to run bootstrap.py with.

The only thing needed is a buildout.cfg - that's basically what you're checking out above, along with a standard bootstrap script and a required recipe (zc.recipe.cmmi) that hasn't been released as an egg to cheesehop yet.

Let's go through each section of buildout.cfg:

[buildout]
develop = zc.recipe.cmmi
parts = libxml2 libxslt lxml lxmlpython

Each buildout.cfg file has a top section called [buildout].

This one has a line develop. It says that I have one development egg, zc.recipe.cmmi. If the recipe were uploaded to the cheeseshop, this line can go away entirely, but right now it hasn't been released as an egg yet, so I need the local checkout of zc.recipe.cmmi to be available as an egg.

The next line, parts, lists what parts need to be built to finish the whole buildout. I want libxml2, libxslt, lxml and lxmlpython. The rest of the buildout.cfg file contains information on how to build the specified parts. Each part uses a recipe, such as zc.recipe.egg, zc.recipe.cmmi, etc.

The first part specifies how to build libxml2. We use the configure;make;make install recipe (zc.recipe.cmmi), which we just made available as a (development) egg. Here is the section that builds libxml2:

[libxml2]
recipe = zc.recipe.cmmi
url = ftp://xmlsoft.org/libxml2/libxml2-2.6.26.tar.gz
extra_options = --without-python

The tarball we want to download, compile and install is specified with the url option. This is enough for many tarballs: the tarball would be compiled and installed as a part in the parts directory under the buildout.

We want to do one extra thing and pass the special option to the configure script of libxml2: do not compile libxml2 with its default Python bindings. After all, lxml replaces those default binding bindings with something better!

The next section for [libxslt] is very similar. We do pass another extra option along to configure to make it link to the libxml2 part we just built:

[libxslt]
recipe = zc.recipe.cmmi
url = ftp://xmlsoft.org/libxml2/libxslt-1.1.17.tar.gz
extra_options =  --with-libxml-prefix=${buildout:directory}/parts/libxml2/
              --without-python

Now we want to build the lxml egg in the [lxml] section. We want to compile lxml 1.1.1 - the system will automatically download its sources from the cheeseshop. We want to pass a few extra options to the lxml build process (its setup script), to make sure it links against the libxml2 and libxslt parts we just built:

[lxml]
recipe = zc.recipe.egg:custom
eggs = lxml == 1.1.1
include-dirs = ${buildout:directory}/parts/libxml2/include:${buildout:directory}/parts/libxslt/include
rpath = ${buildout:directory}/parts/libxml2/lib:${buildout:directory}/parts/libxslt/lib

Finally, we want to build a Python interpreter with the lxml egg as an available import on the PYTHONPATH:

[lxmlpython]
recipe = zc.recipe.egg
interpreter = lxmlpython
eggs = lxml

That's all that's needed to make it download, compile and build lxml and its requirements.

Many buildout recipes exist: we already saw zc.recipe.egg (which we use here just to build an interpreter that has the egg on its path), and zc.recipe.egg:custom to build the lxml egg with some custom options, and zc.recipe.cmmi to do configure;make;make install.

At Infrae we also built a recipe to deploy OpenOffice in server mode, and allowing us to connect to it with PyUNO with the Python interpreter we want, for document conversion purposes.

Other recipes available can install Zope 3 and the ZODB.

This expanding pool of recipes will allow us to put together extremely complicated applications in a reproducable fashion. The idea of buildout is to have control and reproducability - to know exactly what is installed and not to have to install all Python libraries into a global site-packages.

As I said before, buildout uses setuptools and easy_install internally, so the whole egg story just works; it expands on that.

I think this way of building lxml has more benefits than just my single use case of trying it with the latest version of libxml2. It'll allow one to deploy lxml against a known-good version of libxml2 and libxslt, at least on a linux system. It should also help testing and developing lxml - it becomes a lot more easy to test and develop against different versions of libxml2.

I hope this short introduction makes people interested in buildout!