I’ve been out of the Python loop for a few years, but my last impression was that packaging and distribution of Python modules was far from a solved problem. Has this changed?
I'm always wondering if it has to do with the language and or just the package manager itself. Rust's cargo, Node's npm and probably quite a few others exist that work amazingly well.
Genuinely curious. I use Python daily and I rarely encounter problems with it. There is some getting used to in the beginning, but I went through a similar phase when I started using npm and cargo as well.
What I can say is that I had to go through a lot of experimentation myself to arrive at the tools I use now (pyenv + Poetry). And if anything, maybe the lack of one way that is adopted by everyone in the community is the problem.
That is, I think, the big problem. Just using pip has never really worked so people have built tools on top of pip, in parallel with pip and replacements for pip. Basically there is no one way to do things.
Besides, I have 15 years of Python behind me, I code and train in Python for a living, and I still use just pip.
pip and venv are packaged with recent versions of Python and do the job fine.
There is no good heuristic for picking one package over the other when they occupy a similar problem space. This is mostly a cultural problem, the community's efforts are lackluster.
virtualenv is not installed by default, making bootstrapping into a separate prefix that is independent from the system installation unnecessarily aggravating.
Semiautomatic packaging tools (e.g. pypi into rpm) produce low kwalitee packages and manual intervention more often needed compared to similar languages.
Worse are languages that simply don't have much manpower behind them in absolute numbers, e.g. CL/quicklisp. Given Python's mindshare, the results are subpar.
It is, but it's called venv these days:
python3 -m venv my_venv
I _hate_ this working on Java/Maven projects. Why do I need to run all the tests in order to install dependencies? The tests should have been run at _packaging_ time. I'm installing binary dependencies, why should I not trust that they have been tested?
These are fundamentally separate concerns. If you're worried about the robustness of your dependencies, review them first.
> There is no good heuristic for picking one package over the other when they occupy a similar problem space. This is mostly a cultural problem, the community's efforts are lackluster.
Stars on Github? Issues opened vs closed? Stack Overflow? The community is active in all sorts of places and it's not that hard to get a feel for the prevailing best tool for a given job. It just takes a little effort. The community's responsibility is to maintain those parts which they have created, not give you recommendations.
> virtualenv is not installed by default
python -m venv
> Semiautomatic packaging tools (e.g. pypi into rpm)
If you want to package your application and its dependencies as an OS package, that's fine. Do it as a single unit and isolate it from the rest of the system. We have venv (or pyenv if you need a different version of Python) to solve this.
I've experienced every headache there is with Python packaging. It's been bad, it is now better but still rather quirky, but it does require some domain knowledge. If you know how to use it, you will rarely experience serious problems.
I was trying to upgrade a package. I first looked for a "pip upgrade/update <package>"--doesn't exist. I then tried "pip install <package>" to see if it would offer to upgrade--it didn't. I think found "pip install --upgrade <package>".
I'm sure I'd learn these things if I did them all the time, but it's some slightly poor defaults (so slightly bad and impactful I can't imagine them being changed), and some discoverability and friction I don't see in package managers I use way less often.
First it forces the node_module in your repository, which mean you have to configure tons of tools to ignore it or you'll have a bad time.
Then there is no way to have comfortable local command, so plenty of time you gotta have sudo -g. Recent npm version now have a tool for that but the experience is meh.
Plus there is no builtin tool to manage several versions of node, so you gotta add nvm on top of that.
And then if you work on the front end, you can't even use whatever you npm installed stuff, you gotta setup the entire webpack shenaigan, making it literally the hardest stuff to setup. It's so complicated we rely on black box such as create-react-app to do the job for us.
export PATH=$PATH:./node_modules/.bin/
or put the command into `scripts` of package.json, and use `npm run-script foo`.
Have never used npm -g, and never will. Global packages are owned by the system package manager.
E.G:
- people telling you to sudo pip install
- people telling you to use virtualenv instead of venv
- people not telling you to do python -m
- people not telling you about setup.cfg
Don't run before you know how to walk.
you need python (and the distutils module) and you can use the get-pipenv.py
and separating venv management from pip is what causes The Mess