virtualenv

Portland Python Users Group

What is virtualenv?

virtualenv creates isolated Python environments.

• A directory on the filesystem
• Contains Python libraries and scripts
• Optionally insulated from the system's installed packages
• Without requiring you to compile your own Python

PYTHONPATH?

You can already import "uninstalled" modules by manipulating PYTHONPATH:

PYTHONPATH=lib:more_libs:/usr/local/lib/otherlibs python something.py

• Works, but you always get the system packages too
• Installing complex packages into a custom location can be a pain
• Have to always remember to set PYTHONPATH
• Awkward with more than one custom directory on the path

virtualenv / virtualenv.py

http://pypi.python.org/pypi/virtualenv

• It can be downloaded and installed into the system Python
• Or download the archive and just keep virtualenv.py

One step install

$ mkdir foo
$ ls foo
$ python virtualenv.py foo
New python executable in foo/bin/python
Installing setuptools............done.

One step install

$ mkdir foo
$ ls foo
$ python virtualenv.py foo
New python executable in foo/bin/python
Installing setuptools............done.
$ ls foo
bin/  lib/
$ ls foo/bin foo/lib
foo/bin:
activate  easy_install*  easy_install-2.5*  python*

foo/lib:
python2.5/

Activating the virtualenv

$ which python
/usr/bin/python
$ source foo/bin/activate
(foo)$ which python
~/projects/talks/Lightning/virtualenv/foo/bin/python
(foo)$ echo $PYTHONPATH

(foo)$

Managing installation

When you use the python in your virtualenv's bin/ directory, all package installations will be routed into the virtualenv.

• Including packages with C extensions
• And packages with scripts
• Or using easy_install
• With no further ado

Installs with setup.py

(foo)$ tar zxf psycopg2-2.0.6.tar.gz
(foo)$ cd psycopg2-2.0.6
(foo)$ python setup.py install
running install
running build
[...]
(foo)$ python -c 'import psycopg2'
(foo)$ /usr/bin/python -c 'import psycopg2'
Traceback (most recent call last):
  File "<string>", line 1, in ?
ImportError: No module named psycopg2

Installs with easy_install

(foo)$ easy_install nose
Searching for nose
Reading http://pypi.python.org/simple/nose/
Reading http://somethingaboutorange.com/mrl/projects/nose/
Best match: nose 0.10.1
Downloading http://[...]/nose-0.10.1.tar.gz
[...]
(foo)$ which nosetests
~/projects/talks/Lightning/virtualenv/foo/bin/nosetests

Installing development versions

(foo)$ svn co somepackage
[...]
(foo)$ cd somepackage
(foo)$ python setup.py develop

Okay, now what do I do with it?

1. Keep your source code in your virtualenv

$ cd ~/projects/myproj
$ source bin/activate
(myproj)$ ls
bin/  lib/  myproj/  setup.py  tests/
(myproj)$ <hack hack hack>

Okay, now what do I do with it?

2. Don't keep your source code in your virtualenv

$ source ~/environments/myproj/bin/activate
(myproj)$ cd ~/projects/myproj
(myproj)$ ls
myproj/  setup.py  tests/
(myproj)$ <hack hack hack>

Okay, now what do I do with it?

3. Use the virtualenv without activating it

$ ~/projects/foo/bin/python -V
Python 2.5.1

This is great for the #! line of scripts

Okay, now what do I do with it?

4. Use it in mod_wsgi, mod_python or other embedded interpreters

Just put the virtualenv's site-packages directory on the path

import site
site.addsitedir('/home/jek/projects/foo/lib/python-2.5/site-packages')

Okay, now what do I do with it?

5. Bundle it with your source code

Include a bin/virtualenv.py in your source control system, and run it on new checkouts to set up a development environment.

virtualenv also provides an API that allows you to generate a version of the virtualenv.py script customized with any additional logic you might like, such as installing project pre-requisites into the virtualenv.

virtualenv for code library development

• A virtualenv is an isolated version of a specific Python interpreter.
• If you need to develop and test against multiple Python versions and want isolation, each interpreter version needs it's own virtualenv.
• But it's easy to manage. I use a very convenient system of shell aliases, bopping back and forth between 2.3, 2.5, jython, 2.6 trunk, etc.

What virtualenv 1.0 doesn't do

• Work out of the box with Python 2.3.

  You need to find the 2.3-compatible version of subprocess and stick it in the path somewhere before running virtualenv.py.

• Put pydoc in the virtualenv

  You can (awkwardly) (foo)$ python `which pydoc` or just put a pydoc with the correct #! line in your bin/ (it's a 3 line script)

What else does virtualenv do?

It works on Windows too.

virtualenv / virtualenv.py

http://pypi.python.org/pypi/virtualenv

Lots of on-line resources: mailing list, bug tracker, IRC

Increasingly, I see virtualenv recommended as a best practice for developing with web frameworks. If you do web development, you may find additional support in your framework community.

Summary

• A simple, powerful tool
• Environments are cheap and disposable
• Deploy services with a known set of libraries on any system
• Easily test your code against new versions of libraries
• Develop multiple projects simultaneously without conflicts

Thanks!