VS code debugger doesn't load /opt/continuum/project into sys.path

[saraneem]

From European Investment Bank:

When we try to use the vs code debugger for a file located in a subfolder, /opt/continuum/project isn't loaded into sys.path. This means that the local import doesn't work.

Here is a concrete example:

In /opt/continuum/project I create a file test.py and I create a function 'func' in it.
Then I create a folder, with an init.py file and another file 'test2.py'. If I try to import "from test import func", it will fail if I try to debug the file.

It works well if we use conda develop /opt/continuum/project but this configuration can't be saved.

ZD ticket: https://anaconda.zendesk.com/agent/tickets/21015

[AlbertDeFusco]

I'll take a look.

[AlbertDeFusco]

I'm looking at the PNG from the zendesk ticket and this what I see

project/
├── __init__.py
├── anaconda-project.yml
├── folder/
│   ├── __init__.py
│   └── second.py
└── test.py

# folder/second.py

from test import func

The folder/second.py file attempts to import from a file in the main project directory called test.py, but that is not allowed. Please note that the same error message occurs on desktop VSCode.

@mcg1969 , @philippjfr , @jlstevens , I'm inclined to say that this is not a supported use case of Python imports and a different directory structure should be adopted. I'm also not inclined to have ae5-vscode or the session extension adjust sys.path or $PYTHONPATH. Are any of you more aware of what they're attempting to do?

[mcg1969]

I agree with your analysis and verdict far as we understand it. I don't know more about the use case.

[philippjfr]

I agree with that assessment too. I've never tried it but is it possible to ship a package inside an anaconda project with a setup.py and then install it (equivalent to pip install -e .)? I'm guessing at their use case here but I think that would address their needs but even if it worked that would be kind of a pain to set up for a novice.

[jlstevens]

I've never tried it but is it possible to ship a package inside an anaconda project with a setup.py and then install it (equivalent to pip install -e .)

Even if that actually works, I would consider it a flaming hack! Dependencies should be installed as part of prepare and not installed ad hoc in this way. On the other hand, I wonder if it is hack that would work well enough for EIB that we can tell them about even if we don't like it/endorse it.

I agree with @mcg1969 that we need to know more about the use case before suggesting anything.