Today I've been spending far too long installing matplotlib into a Python virtualenv on Ubuntu 12.04, but I finally have it working as I want now. matplotlib lets you make nice data plots with a few lines of Python, but it has several dependencies and this makes it trickier than some packages to get going.

The first obstacle is getting numpy installed. I suspect there were some dependencies I had to install to be able to successfully pip install numpy in my virtualenv. It was only a couple of hours ago, but I have forgotten completely if I had to install anything to get numpy working. Not very helpful, but if you search for any error messages that come up here, you should find the relevant Stack Overflow posts that helped me solve that initial problem.

You also need freetype and libpng installed to be able to pip install matplotlib. I had libpng12-0 already installed, but needed libpng12-dev; likewise, I had libfreetype6 installed, but also needed the development package, libfreetype6-dev.

That wasn't too difficult to solve and matplotlib seemed to install fine. As described elsewhere, one problem I had was that matplotlib was generating plots with savefig, but not showing them when I was trying to show() them. I changed the backend to GTK as the solution suggested, but matplotlib then complained that with an ImportError that "Gtk* backend requires pygtk to be installed".

So, how to solve this. First of all, you'll need to pip uninstall matplotlib in your virtualenv if you've installed matplotlib already. You also need to sudo apt-get install python-gtk2-dev.

I found this post discussing using PyGTK in a virtualenv which helped fix this. Activate the virtualenv you want to use matplotlib in, then do:

ln -sf /usr/lib/python2.7/dist-packages/{glib,gobject,cairo,gtk-2.0,,pygtk.pth} $VIRTUAL_ENV/lib/python2.7/site-packages

This creates symbolic links to the required globally installed Python packages and solves the problem of them being inaccessible. You have to do this before you pip install matplotlib. Otherwise it won't detect at compile time that you have the backend GTK dependencies in the virtualenv. Finally, you need to set the backend to GTKAgg in ~/.config/matplotlib/matplotlibrc (I had to copy matplotlibrc from /lib/python2.7/site-packages/matplotlib/mpl-data in the virtualenv.) If you run this test plot, it should now save it and display it in an interactive window.

Update: Going away and coming back to this problem, I found that using the --system-site-packages option when creating a new virtualenv should also allow the GTK backend to work. The method I've detailed above has the advantage of copying over the minimum of required packages.