'Ubuntu 21.04, Virtualenv and its configuration of Python

EDIT:

  • In addition to the behaviour outlined below, the Python3.10 based environment seems to be ignoring packages installed with the pip -e option (development mode).
  • Specifically, a package installed in development mode is not listed in pip freeze (which, it does in the Python3.9 based virtual environment) and a simple import <package_name> fails. If the package is installed normally (i.e. not in development mode) everything works as expected.

I am using virtualenv to create a virtual environment based around Python 3.10. While virtualenv finishes without errors and does seem to activate, it still fails to pick up its own Python unless the PYTHONPATH environment variable has been set manually.

I am not sure if the situation I am faced with is due to Ubuntu's way of incorporating Python or the way virtualenv is setting up the environment so that it picks up a local interpreter. Here is what I have gathered this far:

  1. My base system is an Ubuntu 21.04. It has its own Python 3 (Python3.9.5) installation which I have not touched at all except installing the python3-virtualenv package using apt.

  2. With Python 3.10, I installed the python3.10-dev package and proceeded to create a virtual environment in the usual way:

    1. > virtualenv -p python3.10 the_env/
    2. > source the_env/bin/activate
  3. Although this looks OK so far, this environment does not have any information about its own site-packages directory. Not even the one that virtualenv is supposed to be creating which includes pip. In this installation, if you try to > pip --version you simply get an error that the pip package does not exist (the pip "executable" location is picked up correctly, but because the interpreter does not know anything about its site-packages it fails to start pip properly).

Long story short, I created two environments, one based on Python3.9 (which works perfectly) and one based on Python3.10 (which does not work) and did a very simple test in each environment:

  • > python -m site
    • On the Python3.9 environment, sys.path includes a path that leads all the way to this particular environment's site-packages
    • On the Python3.10 environment, sys.path does not include that particular path but still includes the typical paths you expect to find (e.g. those pointing to the interpreter itself and the top environment directory but not the specific path that points to the site-packages location.

Following this, I defined a PYTHONPATH manually, before activating the environment which points exactly to the site-packages for that particular environment and everything worked as expected.

I suspect that this might be something to do with the fact that my system's Python is 3.9 which means that the USER_SITE variable is valid while in the case of Python3.10, it is not (because, I do not have a use for it, this is just a virtual environment I am creating). So, I suspect that this might be throwing off the way the site module determines where things are.

As I am not sure, I would like to ask the following:

  1. Could this be something to do with the way Ubuntu handles the Python installation that might just be creating this small problem with virtual environments?

  2. Could the problem be with virtualenv that does not explicitly specify a PYTHONPATH?

  3. Could this behaviour be something of a corner case of the site module?



Solution 1:[1]

What worked for me is an installation from source. After unpacking the source code, as a summary:

$ ./configure --enable-optimizations --with-ensurepip=install --prefix=/path/to/install/to/
$ make -j
$ make test
$ make install
$ /path/to/install/to/bin/python3.10 -m venv /path/to/test
$ source /path/to/test/bin/activate
$ pip list
Package    Version
---------- -------
pip        21.2.4
setuptools 58.1.0
WARNING: You are using pip version 21.2.4; however, version 21.3.1 is available.
You should consider upgrading via the '/path/to/test/bin/python3.10 -m pip install --upgrade pip' command.
$ python -m pip install --upgrade pip
[...]
$ pip list
Package    Version
---------- -------
pip        21.3.1
setuptools 58.1.0

Edit

As mentioned below, I believe the reason why the procedure above just works is simply that when building and installing Python from source, the installation simply will be correct, which I suspect the 3.10 installation in Ubuntu 21.04 is not.

I made my Python installation from source somewhere under my home directory, in order to not risk messing things up, and I also did not permanently modify the PATH.

Setting the path to $PATH:/path/to/install/to/bin should be fine however, either permanently, or just when running mkvirtualenv.

Doing that, my new installation even seems to integrate seamlessly with my system's virtualenvwrapper.

I don't really need to be on the Python leading edge myself, but if I did, I would definitely make myself independent of Ubuntu's latest development by using the procedure described above.

By the way, if I update pip directly in the Python installation built from source, I will no longer get the message about the older pip (see above) when I install new virtual environments.

Edit 2

Bug reported to Ubuntu: https://bugs.launchpad.net/ubuntu/+source/python3.10/+bug/1955742

Edit 3

Also, since Ubuntu 21.04 has end of life in a month or so, upgrading to 21.10 could really make sense. I have just tried that, and it seems that Python 3.10 virtual environments work just fine in that release.

Solution 2:[2]

With the advent of Ubuntu 22.04 (which caused a few minor issues with some specific python virtualenv setups I had) and having already spent some time figureing this out, I ended up switching to using pyenv.

Pyenv made it very easy to install any python version and any number of virtual environments within it which can be further customised in complete isolation.

The only thing to be careful of is installing all necessary python prerequisites before installing a specific version to avoid missing functionality from some packages (e.g. not including the lzma library will generate an ominous warning from pandas ("your python installation is incomplete..."). This can be ignored if you are not using that functionality or otherwise, easily fixed.

I think that this is a better option overall if you have to manage different versions and specific configurations for python, even across distros.

This solution is very close to the one suggested before ("install from source") so I will be accepting that one and leave my contribution as additional information about this problem.

Sources

This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.

Source: Stack Overflow

Solution Source
Solution 1
Solution 2 A_A