requirements

developers/users want both flexibility and reproducibility

• at pip install time, want libsolv to choose depends versions according to their requirements

• at distribute time, want pip freeze into requirements.txt to reproduce in 3rd party env

• github software POV: need flexible combination for different projects

• develop project POV: need reproducibility, via env, lockfile, hash export/import

2 types of depends

fixed pkgname + hash

• suitable for ELF NEEDED lib.so
• not suitable mechanism for python depends
• at invoke time: need search hash for updated libs

Comparing to RPM, Nix turns runtime depends to build time depends, e.g. sed -e "s| awk | ${gawk}/bin/awk |"

• pro: scalable by avoiding global version range conflicts, or dimond-depends
• con: maintain costs on per-app fixups

global libsolv algorithm is no longer applicable for ELF lib.so depends: /opt/epkg/store/hash per-dir multi-version and rpath fixed the dimond-depends problem, which eliminates version range conflicts in another dimension

libsolv pkgname + version range

• suitable for PATH cli command, PYTHONPATH python module
• at invoke time:
  • prepend current $env/bin to PATH: work for 'cmd', wont work for '/bin/cmd'
  • bind mount /bin: work in linux, WSL2, wont work in WSL, mac?

optional depends

• conditional: +dep if selinux
• alternative: dep1|dep2, e.g. mawk|gawk, vim|nvim

TODOs

https://conda-forge.org/ https://anaconda.org/conda-forge

scipy 2019: want nix improvements for python

• allow imperative installation of python packages
• maintain multiple versions of major packages in ixpkgs
• declarative standard for python instead of setup.py, setup.cfg, requirements.txt, MANIFEST.in etc.

epkg + pypi

• epkg for fixed c/c++ hash
• multi-version python/gcc etc. in one repo
• make local 'epkg build' fast/easy:
  • source yaml is depends in version range
  • • lock file is version baseline, =>实例化构建
  • binary is reusable hash
  • how to auto get lock file? libsolv.. OS default baseline https://wiki.nixos.org/wiki/Python
• replace pyenv/virtualenv/conda, deep integration with pip
  • https://ioflood.com/blog/pyenv/
  • https://prefix.dev/blog/pixi_a_fast_conda_alternative
  • https://prefix.dev/blog/pixi_launch
  • https://taras.glek.net/post/trying-pixi-modern-python-packaging/
  • https://pip.pypa.io/en/stable/cli/pip_install/
  • https://pip.pypa.io/en/stable/topics/vcs-support/
• how to ensure OS self-contained depends for python modules, from other OS packages?
• how to meet AI project requirements like conda?

epkg env for C+python projects

epkg env create mypyproj
epkg env activate mypyproj
epkg install C packages
epkg install python packages
epkg install pip
pip install pypi modules
epkg env export # C+python+pypi
epkg env import