Conda environment revision rollback limits with mixed pip installations

Rollback Behavior and Limitations

No, the conda revision system does not provide a native mechanism to synchronize, remove, or revert pip-installed packages during a rollback. The --revision flag operates exclusively on the conda metadata history; because pip manages its own package state independently, pip-installed dependencies are invisible to the conda rollback process.

Conflict Behavior and Import Resolution

When a conda rollback reinstalls a package that conflicts with a pip-installed version, the environment enters a "shadowed" state. Both versions may physically exist within the site-packages directory, but Python's import mechanism will only load one.

• Import Priority: Generally, the version located first in sys.path is imported. In most conda environments, conda-managed packages take precedence over those installed via pip.
• Runtime Risk: If a conda rollback restores an older version of a library while a newer, incompatible pip version remains, you may encounter ImportError, AttributeError, or silent runtime failures if the pip package has modified shared dependencies.

Recovery Workflow

Since conda cannot automate pip recovery, you must manage the pip state manually. To ensure a consistent environment after a rollback, follow these steps:

1. Export current pip state: Before rolling back, capture the current pip dependencies if you need a reference point.

   pip freeze > pre_rollback_requirements.txt

2. Perform the conda rollback:

   conda install --revision [N]

3. Clean pip packages: Manually uninstall pip packages that were added after the target revision or use a fresh requirements file to reset the pip state.

   pip uninstall -r pre_rollback_requirements.txt

4. Reinstall verified pip dependencies: Install only the specific pip packages required for that specific conda revision point.

Verification

To verify if a pip package is shadowing a restored conda package, run the following command to check the installation path of the imported module:

python -c "import [package_name]; print([package_name].__file__)"

If the path points to a site-packages location not managed by the restored conda revision, the pip version is still present and potentially conflicting.

Diagnostic Detail: Are you using a conda-forge based environment or the defaults channel? This can affect how conda handles the initial installation of packages that are also available via pip.