How to build lxml from source

To build lxml from source, you need libxml2 and libxslt properly installed, including the header files. These are likely shipped in separate -dev or -devel packages like libxml2-dev, which you must install before trying to build lxml.



The lxml.etree and lxml.objectify modules are written in Cython. Since we distribute the Cython-generated .c files with lxml releases, however, you do not need Cython to build lxml from the normal release sources. We even encourage you to not install Cython for a normal release build, as the generated C code can vary quite heavily between Cython versions, which may or may not generate correct code for lxml. The pre-generated release sources were tested and therefore are known to work.

So, if you want a reliable build of lxml, we suggest to a) use a source release of lxml and b) disable or uninstall Cython for the build.

Only if you are interested in building lxml from a checkout of the developer sources (e.g. to test a bug fix that has not been release yet) or if you want to be an lxml developer, then you do need a working Cython installation. You can use pip to install it:

pip install -r requirements.txt

lxml currently requires at least Cython 0.20, later release versions should work as well.

Github, git and hg

The lxml package is developed in a repository on Github using Mercurial and the hg-git plugin. You can retrieve the current developer version using:

hg clone git:// lxml

This will create a directory lxml and download the source into it, including the complete development history. Don't be afraid, the download is fairly quick. You can also browse the lxml repository through the web.

Building the sources

Clone the source repository as described above (or download the source tar-ball and unpack it) and then type:

python build


python bdist_egg     # requires 'setuptools' or 'distribute'

To (re-)build the C sources with Cython, you must additionally pass the option --with-cython:

python build --with-cython

If you want to test lxml from the source directory, it is better to build it in-place like this:

python build_ext -i --with-cython

or, in Unix-like environments:

make inplace

To speed up the build in test environments (e.g. on a continuous integration server), set the CFLAGS environment variable to disable C compiler optimisations (e.g. "-O0" for gcc, that's minus-oh-zero), for example:

CFLAGS="-O0"  make inplace

If you get errors about missing header files (e.g. Python.h or libxml/xmlversion.h) then you need to make sure the development packages of Python, libxml2 and libxslt are properly installed. On Linux distributions, they are usually called something like libxml2-dev or libxslt-devel. If these packages were installed in non-standard places, try passing the following option to to make sure the right config is found:

python build --with-xslt-config=/path/to/xslt-config

If this doesn't help, you may have to add the location of the header files to the include path like:

python build_ext -i  -I /usr/include/libxml2

where the file is in /usr/include/libxml2/libxml/xmlversion.h

To use lxml.etree in-place, you can place lxml's src directory on your Python module search path (PYTHONPATH) and then import lxml.etree to play with it:

# cd lxml
# PYTHONPATH=src python
Python 2.7.2
Type "help", "copyright", "credits" or "license" for more information.
>>> from lxml import etree

To make sure everything gets recompiled cleanly after changes, you can run make clean or delete the file src/lxml/etree.c.

Running the tests and reporting errors

The source distribution (tgz) and the source repository contain a test suite for lxml. You can run it from the top-level directory:


Note that the test script only tests the in-place build (see distutils building above), as it searches the src directory. You can use the following one-step command to trigger an in-place build and test it:

make test

This also runs the ElementTree and cElementTree compatibility tests. To call them separately, make sure you have lxml on your PYTHONPATH first, then run:




If the tests give failures, errors, or worse, segmentation faults, we'd really like to know. Please contact us on the mailing list, and please specify the version of lxml, libxml2, libxslt and Python you were using, as well as your operating system type (Linux, Windows, MacOS-X, ...).

Building an egg or wheel

This is the procedure to make an lxml egg or wheel for your platform. It assumes that you have setuptools or distribute installed, as well as the wheel package.

First, download the lxml-x.y.tar.gz release. This contains the pregenerated C files so that you can be sure you build exactly from the release sources. Unpack them and cd into the resulting directory. Then, to build a wheel, simply run the command

python bdist_wheel

or, to build a statically linked wheel with all of libxml2, libxslt and friends compiled in, run

python bdist_wheel --static-deps

The resulting .whl file will be written into the dist directory.

To build an egg file, run

python build_egg

If you are on a Unix-like platform, you can first build the extension modules using

python build

and then cd into the directory build/lib.your.platform to call strip on any .so file you find there. This reduces the size of the binary distribution considerably. Then, from the package root directory, call

python bdist_egg

This will quickly package the pre-built packages into an egg file and drop it into the dist directory.

Building lxml on MacOS-X

Apple regularly ships new system releases with horribly outdated system libraries. This is specifically the case for libxml2 and libxslt, where the system provided versions used to be too old to even build lxml for a long time.

While the Unix environment in MacOS-X makes it relatively easy to install Unix/Linux style package management tools and new software, it actually seems to be hard to get libraries set up for exclusive usage that MacOS-X ships in an older version. Alternative distributions (like macports) install their libraries in addition to the system libraries, but the compiler and the runtime loader on MacOS still sees the system libraries before the new libraries. This can lead to undebuggable crashes where the newer library seems to be loaded but the older system library is used.

Apple discourages static building against libraries, which would help working around this problem. Apple does not ship static library binaries with its system and several package management systems follow this decision. Therefore, building static binaries requires building the dependencies first. The script does this automatically when you call it like this:

python build --static-deps

This will download and build the latest versions of libxml2 and libxslt from the official FTP download site. If you want to use specific versions, or want to prevent any online access, you can download both tar.gz release files yourself, place them into a subdirectory libs in the lxml distribution, and call with the desired target versions like this:

python build --static-deps \
       --libxml2-version=2.9.1 \
       --libxslt-version=1.1.28 \

sudo python install

Instead of build, you can use any target, like bdist_egg if you want to use setuptools to build an installable egg, or bdist_wheel for a wheel package.

Note that this also works with pip. Since you can't pass command line options in this case, you have to use an environment variable instead:

STATIC_DEPS=true pip install lxml

To install the package into the system Python package directory, run the installation with "sudo":

STATIC_DEPS=true sudo pip install lxml

The STATICBUILD environment variable is handled equivalently to the STATIC_DEPS variable, but is used by some other extension packages, too.

If you decide to do a non-static build, you may also have to install the command line tools in addition to the XCode build environment. They are available as a restricted download from here:

Without them, the compiler may not find the necessary header files of the XML libraries, according to the second comment in this ticket:

Static linking on Windows

Most operating systems have proper package management that makes installing current versions of libxml2 and libxslt easy. The most famous exception is Microsoft Windows, which entirely lacks these capabilities. To work around the limits of this platform, lxml's installation can download pre-built packages of the dependencies and build statically against them. Assuming you have a proper C compiler setup to build Python extensions, this should work:

python bdist_wininst --static-deps

It should create a windows installer in the pkg directory.

Building Debian packages from SVN sources

Andreas Pakulat proposed the following approach.

In case dpkg-buildpackage tells you that some dependecies are missing, you can either install them manually or run apt-get build-dep lxml.

That will give you .deb packages in the parent directory which can be installed using dpkg -i.