`BasicWorkflow` requires redundant specification of `inference_variables` despite adapter definition

[pipme]

While using BasicWorkflow, I encountered a minor point of confusion regarding how inference_variables are handled in conjunction with an adapter.

If I define an adapter like this:

adapter = (
    bf.Adapter()
    .rename("w", "inference_variables")
    .concatenate(["X", "y"], into="summary_variables")
    .rename("N", "inference_conditions")
)

I would expect that passing the adapter to the workflow should be sufficient:

workflow = bf.BasicWorkflow(
    ...
    adapter=adapter
)

However, it seems that inference_variables=["w"] must still be passed separately:

workflow = bf.BasicWorkflow(
    ...
    adapter=adapter,
    inference_variables=["w"]
)

From my inspection, it looks like inference_variables is only used internally by _prepare_for_diagnostics, so this extra specification feels redundant and somewhat unintuitive, especially when the adapter already defines this mapping.

Suggestion:
It would be great if the workflow could infer inference_variables directly from the adapter/approximator when possible, or at least document this design choice more explicitly to avoid confusion.

[paul-buerkner]

I fully agree, we should change this. @stefanradev93 what do you think?

[stefanradev93]

Yes, this is a good idea, which we discussed briefy on Slack. I will implement it.

[stefanradev93]

Solved in 38bef3d.