Options for packaging your Python code: Wheels, Conda, Docker, and more

pythonspeed.com

23 points·plessthanpt05··5 comments

There is also zip app [1] if you already have python installed

[1] https://docs.python.org/3/library/zipapp.html

This seems to be focused on packaging applications, and suggests that wheel is not good enough if you have a dependency on a C library. I am surprised since I thought you could embed a C library in a wheel.

I would like to know, is wheel+pypi an acceptable way to go to package and distributed a Python library that includes a C library (so/dll)?

I get the impression that big libraries like numpy and tensorflow use this method to pretty good effect. It seems easier than trying to compile the C libraries on the target machine, but is it quite difficult to achieve multiplatform support this way?

It depends on what you intent to do by including a C library. It’s a good choice you want to use that library in your Python modules. If you expect users to link to that library directly, wheels would be a poor choice that leads to never ending headaches (not wheel’s fault, but mainly due to how shared libraries are designed).

> The Conda package system packages both Python packages and C shared libraries and the Python interpreter into Conda packages.

This doesn't sound correct. Conda packages don't actually include the entire Python interpreter in them right?

A prerequisite to installing Conda packages is to have Anaconda/Miniconda installed which brings in the Python interpreter and a given set of packages depending on the version you install but that doesn't mean that each Conda package will be carrying the interpreter with it.

That's not what I meant, but rather "Python is a Conda package too, a dependency." I'll rephrase to make it clearer.

Reply on news.ycombinator.com