Imports isolation in tests
When working on PR #708, I accidentally removed this inline import:
https://github.com/NatLibFi/Annif/blob/fb512928b455f785f4799e0c6b96db5ecbfb12b3/annif/project.py#L205
This made annif index command fail:
...
File "/home/local/jmminkin/git/Annif/annif/cli.py", line 322, in run_index
results = project.suggest_corpus(documents, backend_params).filter(limit, threshold)
File "/home/local/jmminkin/git/Annif/annif/project.py", line 238, in suggest_corpus
return annif.suggestion.SuggestionResults(suggestions)
AttributeError: module 'annif' has no attribute 'suggestion'
But this was not detected by tests for index command. Seems that some commands that are tested before import annif.suggestion, and this import remains for index command.
The tests should probably be some way more isolated to be able to detected this kind of possibility for bugs.
I think that could be a consequence of running the CLI tests via CliRunner. It simulates a CLI execution environment, but isn't quite the same as running the commands separately from a real CLI. If we wanted to test that, we'd need something like a shell script (or possibly even a bats test suite) that runs the annif commands separately.