Unclear dependencies conflict when installing from conda-forge

[drorata]

I have a conda environment defined by:

name: my_env
channels:
    - conda-forge
    - plotly
dependencies:
    - python=3.8
    - pip
    - pytest
    - "pandas<2.0"
    - "matplotlib<4.0"
    - "scikit-learn<0.23"
    - "pymc3<4.0"
    - mkl-service
    - "python-graphviz<0.14"
    - pydot
    - "bidict<0.20"
    - tqdm
    - plotly=4.6.0
    # - "dvc<1.0"
    - pip:
        - pgmpy==0.1.9
        - streamlit

Once uncommenting the dvc dependency, I get messages like:

Found conflicts! Looking for incompatible packages

After about 30 minutes I terminated the process and now wondering why is it happening and what's the best approach to add dvc to this environment. Hinted by #2782 I tried to remove tqdm from the dependencies and add dvc, but conflicts appeared again. Any ideas?

This happened on

• macOS 10.15.4
• conda version 4.8.3

NB I moved dvc to the pip section and was able to update the environment. Still it would be nice to have it install using conda(-forge).

[efiop]

@drorata I suppose when you declare python 3.8, it tries to install 3.8 package for dvc? If so, we don't currently support 3.8 on conda. conda-forge/dvc-feedstock#98

[drorata]

So in other words, you're saying that when using python 3.8 with conda, the only way at the moment to install dvc is using pip, right?

Is there a way to improve the behavior so the process will fail faster and indicate that 3.8 compatible dvc is not available?

[efiop]

So in other words, you're saying that when using python 3.8 with conda, the only way at the moment to install dvc is using pip, right?

Pip package definitely supports 3.8, yes.

Is there a way to improve the behavior so the process will fail faster and indicate that 3.8 compatible dvc is not available?

That's conda's issue, can't improve that myself :( But can look into conda-forge/dvc-feedstock#98 (we've been postponing it for awhile because of dependencies) to try to make it build for py3.8. IIRC it mostly requires pinging dep package maintainers to support 3.8 in their packages.

[drorata]

So the recommended way to install dvc in a Python 3.8 environment at the moment is using pip. Right?

[efiop]

@drorata Sorry for the delay. Correct. We have #3685 on our todo list, so should be ready soon. Closing this ticket in favor of that one.

[efiop]

@drorata 3.8 packages are now available through conda. Please give it a try 🙂 Thanks for the feedback!