Use module name (-m) to run pytest and pip

[dosas]

when using python version other than 3.8 pip and pytest were not using the correct python version

[mdellweg]

And how does python3 know the right one to choose? I believe with multiple versions of python installed in one container there is something going wrong here.

[dosas]

The switch python script takes care of that, it updates the alternative.

It is no problem to have multiple python versions as long as python -m is used.

[mdellweg]

It is no problem to have multiple python versions as long as python -m is used.

I can see how it works, but the result looks more like a space station to me than a container.

[dosas]

This PR is not about changing the architecture or workflow of oci-env. It tries to provide a fix for an 'advertised' functionality, the possibility to use a different python version than 3.8. This functionality was broken an unnoticed.

If you have general issues with this workflow or architecture this is out of scope for this PR. You are welcome to provide an alternative solution.

[mdellweg]

I guess switch python is broken then. This line is supposed to do it. https://github.com/pulp/oci_env/blob/3fd2c2486fe9f95ebcf841d6fb3fde24c345d83b/base/switch_python#L17

[dosas]

No, I disagree. The only way to have the correct python version for pip and pytest is using -m flag.

[mdellweg]

I see it adding the alternative for pip3. Looks like it's missing to put one for pip too.

[dosas]

So we have two possibilities:

1. Run pip and pytest and hope they chose the correct python version or
2. Run the correct python version and let it invoke pip and pytest

I do not see any point in favor the first one.

If it works we can additionally add pip to switch_python or change every occurrence of pip to pip3.

[mdellweg]

It is ridiculous in the first place to have multiple versions of python in one container fighting over precedence. If you can find a way to fix up switch python, that might be an acceptible stop gap, but in the long run, when you want another python, start with a different base image. When you do this, think about a user calling oci_env shell. They don't understand that pip or pytest is the wrong command to use.

[dosas]

If it is 'ridiculous' to have multiple versions of python, the logical option would be to remove the broken switch_python script.

[dkliban]

@jctanner are you using the feature that lets you switch python versions? It doesn't seem to be working correctly. We are considering removing it all together.

[jctanner]

We use it all the time with no issue https://github.com/ansible/galaxy_ng/blob/master/profiles/base/Dockerfile

[dosas]

We use it all the time with no issue https://github.com/ansible/galaxy_ng/blob/master/profiles/base/Dockerfile

That is because of the -m option :D

Are you invoking tests (unit/functional) with it?

[jctanner]

@dosas no we aren't using it for functional or unit testing. The functional and unit tests run in CI based on the ansible playbooks created by the pulp plugin template. It builds an image and runs a container via ansible to do all that work, including a matrix with s3/azure/fs backing filestores. The galaxy team does not tend to write functional tests in the galaxy_ng repo. Instead, in our CI we also have integration tests (not pulp native testing) that do use the oci profiles+image.

[cutwater]

@dosas @mdellweg I wonder why can't oci-env just use venv created by required version of python. This will effectively fix all lookup issues and will not require tampering with system level python?

[mdellweg]

We provide different base images with different python versions for this. Closing now.

@dosas @mdellweg I wonder why can't oci-env just use venv created by required version of python. This will effectively fix all lookup issues and will not require tampering with system level python?

This would still need multiple python versions to coexist in the same container.

[dosas]

For reference https://github.com/pulp/pulp-oci-images/issues/586