Packaging with manylinux + auditwheel pip wheels vs. Conda

755 Views Asked by At

Description

So I am looking to package a fairly complex python application that requires scientific libraries. This question is somewhat similar to the stackoverflow pip vs conda discussion but it does not go into detail about the differences now that binary packaging for linux wheels is available since 2016. I have seen that pypi/cryptography uses manylinux and distribute binaries via pypi wheels. Another package mpi4py only releases for binary packages for conda only because of the difficulty. Even going as far as to say that binary wheels are not appropriate for the task.

Question

What does shared library packaging via wheels look like vs conda? Is shared library packaging via wheels even worth it as of 2018?

Requirements

My package requires

  • openblas
  • fftw3
  • openmpi
  • lammps

In all the my static binary is about 100Mb so it does require a lot of shared libraries. And many I also point out that installation is a huge pain even for myself... I can't imagine what it would be like for someone trying to install it themselves. So far I have a working docker container.

0

There are 0 best solutions below