I'm a bit torn and often err on the side of installing into --user or virtualenv. What often happened is I would toy around with a project, install a few libraries, then abandon it. Months/years later I would hit an incompatibility and I wouldn't know which packages were necessary for things I use everyday and which were because of that abandoned package. Similarly, when revisiting that abandoned project I wouldn't know what a good known set of libraries were. Installing stuff globally also makes it really difficult to know what dependencies were needed to run it on a different machine.
This wasn't just for Python libraries, but also with OS packages (especially non-sanctioned RPMs). I've been very selective about what I install globally, and usually first reach for a VM or other virtual environment.
One big counter to this is when I want to make it a simple commandline tool to run. I have to enter a Virtualenv just to run my fancier `ls` command?
My solution for this problem is apk's virtual packages. I'll do something like this:
apk add -t .myproject-deps py3-foobar py3-foobaz
This creates a fake ".myproject-deps" package which depends on py3-foobar and py3-foobaz, then I can just uninstall ".myproject-deps" later and it nukes the rest. `head /etc/apk/world` is generally sufficient to get a list of projects I've forgotten about whenever I feel like some spring cleaning.
This wasn't just for Python libraries, but also with OS packages (especially non-sanctioned RPMs). I've been very selective about what I install globally, and usually first reach for a VM or other virtual environment.
One big counter to this is when I want to make it a simple commandline tool to run. I have to enter a Virtualenv just to run my fancier `ls` command?