Closed jbusecke closed 2 years ago
Distribute a tarball to pypi, which can then be used by individual user to build the package locally (requiring fortran compiler/setup on their end)
A while ago, I put together a GitHub Action workflow for building manylinux wheels for https://github.com/cspencerjones/xlayers. You may find this GitHub Action workflow useful if you ever decide to publish manylinux binary wheels in addition to the standard tarball.
https://github.com/cspencerjones/xlayers/blob/master/.github/workflows/pythonpublish.yaml
I think I have figured this out in the publish CI, and it works with pangeo-forge (which was the most important aspect). Closing this now.
My brain is fried for today, but let me summarize what worked and what didnt in my quest to package and move this code to both pypi and conda-forge
First of all, am I correct that I should aim for the following situation?:
Assuming that this is the case I have achieved the following today. The current state of the repo does compile and work on my local machine.
I have tried to upload a tarball to test.pypi.com, which I can download on the pangeo cloud, but have not yet managed to install. I tried to mamba install
gfortran
and then runpip install -i https://test.pypi.org/simple/ aerobulk-python
but got this bad boy of an error message:But on my local machine I was able to install and run this version with
pip install -i https://test.pypi.org/simple/ aerobulk-python --no-binary :all:
, so I assume I could make it work on the pangeo cloud too given the right tinkering.I have in parallel tried to create a conda-forge recipe, but I did not manage to actually get the code checked out using git release archives (this suggests that I might get around these issues by simply basing the recipe on a pypi tarball).
So if my assumptions above are correct, I will distribute a tarball to pypi next (which might or might not work depending on your local setup), but might be able to use that as source for the conda-recipe.
Any thought/tips are much appreciated.