Improve test coverage

[Galileo-Galilei]

Test coverage is very poor at the moment. the goal is too get as fast as possible at 100% coverage with test for "normal" use and to add edge case when a bug is encountered. Once coverage will reach 100%, we will enforce keeping this 100% coverage for merging on master in the CI chain.

I'll edit this post with the test cases I'll come up with:

• [x] cli:
  • [x] cli_utils
  • [x] cli
• [x] context:
  • [x] config
  • [x] mlflow_context
• [x] hooks:
  • [x] node_hook
  • [x] pipeline_hook
• [x] io:
  • [x] mlflow_dataset:
    • [x] Test whether that saving and loading works in a local mlruns folder
    • [x] Test whether both "string" (e.g. "pandas.CSVDataSet" and "classes" (e.g CSVDataSet) both work to ensure possible configuration in catalog.yml
    • [x] Test that you can save in a specific crun id, wcovering the edge case when a mlflow run is already active
• [x] mlflow:
• [x] pipeline:
  • [x] pipeline_ml:
    • [x] test that PipelineML class is preserved through context manipulation (especially different kinds of filtering)
    • [x] test that catalog extraction works in different setup : with MemoryDataSet, with non local filepath, with empty catalog, with unspecified model_input_name...
    • [x] test instantiation for pipeline with the degree of freedom is ok for pipeline with degree of freedom <1, ==1 or >1.
  • [x] modular_pipeline_ml
• [x] template
• [x] utils

[Galileo-Galilei]

Hello @kaemo, I have finally completed the unit testing phase and reach 100% coverage. I merged the results on the develop branch. You have detailed explanation of the changes in the CHANGELOG.

By now, the tests are only unit test and do not contain integration test. If the formatting of some objects change in a future kedro release, some test will not break even if the the new kedro release is no longer compatible with the plugin (I have specifically in mind the run_params argument of the pipeline hook, but template changes can cause the same type of problem).

I think that integration tests are something really different and they should be tracked in another issue. I close this issue, but feel free to reopen it if needed.