Skip to content

Latest commit

 

History

History
36 lines (33 loc) · 2.4 KB

publish_release.md

File metadata and controls

36 lines (33 loc) · 2.4 KB
Raw

Steps for Pip

Prerequisites

Publish release

  • Summarize changes in the changelog.
  • Increment version number here and here.
  • [Optional] Remove previous versions rm dist/*
  • Create the new package: python setup.py sdist
  • [Optiona] Upload to the testpypi: twine upload --repository testpypi dist/*
  • Upload the package: twine upload dist/*. You will be asked to enter your credentials for every update.

Steps for Conda

You can build separate packages for different architectures and python versions. This is a guide for python 3.9 on a Linux machine.

Prerequisites

  • First make sure the pip package is updated.
  • Create a fresh conda environment and activate it.
  • Install some conda packages: conda install conda-build conda-verify
  • Install conda client: conda install anaconda-client
  • Register at https://anaconda.org/account/login

Publish release

  • Navigate to a different folder: TMP=$(mktemp -d); cd $TMP
  • Initialize with pip skeleton: conda skeleton pypi pyfvvdp
  • Open the file pyfvvdp/meta.yaml and make the following changes:
    1. sed -i 's/torch/pytorch/g' pyfvvdp/meta.yaml: replace all occurances of "torch" with "pytorch". Pip and conda reference PyTorch under different names.
    2. sed -i 's/pyfvvdp\/third_party/pyfvvdp.third_party/g' pyfvvdp/meta.yaml: replace "pyfvvdp/third_party" with "pyfvvdp.third_party". Not sure why conda initializes this field incorrectly, need to investigate further.
    3. update the field "recipe-maintainers".
    4. add requirement libiconv (under host and run)
  • Build the conda package: conda-build -c conda-forge --python 3.9 pyfvvdp. You can build multiple packages, one for each python version to be supported. Important Do not use multiple conda channels. This will result in broken packages.
  • Test the build by installing pyfvvdp conda install -c ${CONDA_PREFIX}/conda-bld/ -c conda-forge pyfvvdp
  • Convert the build to other architectures: E.g., conda convert --platform all ${CONDA_PREFIX}/conda-bld/linux-64/pyfvvdp-1.1.1-py39_0.tar.bz2 -o ${CONDA_PREFIX}/conda-bld
  • Upload all the packages to conda: anaconda upload ${CONDA_PREFIX}/conda-bld/*/pyfvvdp-*.tar.bz2
  • [Optional] Logout from the client: anaconda logout
  • [Optional] Clean intermediate build files: conda build purge