Unit test workflow failed setting up conda environment: "Warning: Pip subprocess error"

[prjemian]

The code passed unit tests locally but the GitHub actions workflow failed while setting up the conda/pip environment using mamba. (Pip could not find the named conda environment for py36 & py38).

Warning: Pip subprocess error:
Not a conda environment: /usr/share/miniconda/envs/hklpy-test-3.8
/usr/share/miniconda/envs/${ENV_NAME}/.tmp6z870pyp: line 3: /usr/share/miniconda/envs/hklpy-test-3.8/bin/python: No such file or directory

Originally posted by @prjemian in #100 (comment)

This is a new failure (since last commit a few months ago) of the workflow. The failure happened before the repo's code was installed, while constructing the conda environment when adding the pip requirements. Pip could not find the environment directory.

[prjemian]

Reported previously in the setup-miniconda repo we're using in our workflow. Suggested remedies are:

1. don't use python-version (hint: use sed to rewrite environment.yml and specify python version)
2. continue with python-version but separate the pip install into a separate workflow command

We're using python-version in our workflow now.

[prjemian]

We're already using sed to specify the python version:

hklpy/.github/workflows/conda_unit_test.yml

Lines 20 to 23 in c77c3cd

	- name: set environment name in YAML file
	shell: bash -l {0}
	run: |
	sed -i.bak "s/name: ${PACKAGE}/name: ${ENV_NAME}/g" environment.yml

but also specifying python-version in the config:

hklpy/.github/workflows/conda_unit_test.yml

Line 35 in c77c3cd

python-version: ${{ matrix.python-version }}

Comment-out line 35 and give it a try.

[prjemian]

Commit 96f672c did not solve the problem. Plan B is to separate the pip install to a separate command.

Need to rewrite environment.yml and remove the pip requirements, writing them to a requirements.txt file, then add the pip install step (after activating the new conda environment).

If the sed can be replaced with a python script, this will be easy. It requires that some python version be available at that part of the script. Test that next. Since this is a test, comment all the later steps for now.

[prjemian]

Result (python 3.8.5 is available as python or python3):

  which python
  which python3
  python --version
  python3 --version
  shell: /usr/bin/bash -l {0}
  env:
    ENV_NAME: hklpy-test-3.8
    PACKAGE: hklpy
/usr/bin/python
/usr/bin/python3
Python 3.8.5
Python 3.8.5

[prjemian]

The root problem was in the workflow with activate-environment: ${ENV_NAME}.

Replacing that with activate-environment: test-hklpy has a successful outcome for all Python versions tested 3.6, 3.7, 3.8). For now, all will use the conda environment name test-hklpy