janus-core has a large and strict dependency footprint; pinning aiida-mlip to a narrow range of janus-core versions has a drastic impact on its compatibility with other aiida plugins. It also needlessly bloats the installation of the package.
The janus-core code@computer is actually interacted with through command-line parameters, and may exist in its own (e.g. Conda) environment and/or on a another computer. The aiida_mlip plugin code doesn't seem to import from janus_core at all, so it is not technically a required dependency.
I appreciate that aiida-mlip versions are associated with a particular range of supported janus-core versions. But the pyproject.toml dependency doesn't actually do anything to enforce that the user points to this particular janus-core installation with their code@computer; I suggest this is handled with runtime validation instead.
janus-corehas a large and strict dependency footprint; pinningaiida-mlipto a narrow range ofjanus-coreversions has a drastic impact on its compatibility with other aiida plugins. It also needlessly bloats the installation of the package.The janus-core code@computer is actually interacted with through command-line parameters, and may exist in its own (e.g. Conda) environment and/or on a another computer. The aiida_mlip plugin code doesn't seem to import from
janus_coreat all, so it is not technically a required dependency.I appreciate that
aiida-mlipversions are associated with a particular range of supportedjanus-coreversions. But the pyproject.toml dependency doesn't actually do anything to enforce that the user points to this particularjanus-coreinstallation with theircode@computer; I suggest this is handled with runtime validation instead.