Conflicting test imports

[nikeethr]

RE: "random" test failure

I suspect this has to do with pytest import modes: https://pytest.org/en/7.4.x/explanation/pythonpath.html#import-modes

Hypothesis:

with the default i.e. prepend, having two tests directory in the mono repo with __init__.py specified will cause conflicts. It could be that pipeline package was the only package with __init__.py in its tests folder (needs verification). Introducing __init__.py to site_archive_nci/tests caused two "tests" modules to be recognised, with site_archive_nci taking precedence, since it is installed last (due to prepend - LIFO). This may have caused helper functions in pipeline to not be recognized.

Regardless of precedence, we want the ability for package tests to have utility modules since they are in some instances easier to write and organise than fixtures.

There are a few options to work around this, but its beyond the scope of this issue.

Originally posted by @nikeethr in #106 (comment)

[nikeethr]

The task is to basically verify the "hypothesis" in the description and come up with a solution for it - if the hypothesis correctly identifies the problem (if not report back and we can close this issue, raise a different one for further investigation).

A couple potential options/hints (noting that there could be better ideas):

1. split the pipeline to run tests per package separately.
   • note: this also has the added advantage of loading dependencies independently causing things to fail faster reducing the load on the CI pipeline
2. investigate the use of append and import modes as mentioned in the documentation in the description: https://pytest.org/en/7.4.x/explanation/pythonpath.html#import-modes
3. use fixtures instead of utility modules (only pipeline has a utility module - fake_* under tests, so its pretty reasonable to replace with fixtures.)