# with_columns_base

Documenting the current design flow for the `with_columns` decorator.

For now, it belongs to the `NodeInjector` lifecycle since it still runs the decorated function as a node but injects the dataframe with columns appended columns as one of the parameters.

The `with_columns` consists of three parts that are represented in the corresponding three abstract methods in `with_columns_base`:

1. `get_initial_nodes` -- Input node(s): Either a dataframe if `pass_datafame_as` is used or extracted columns into nodes if `columns_to_pass` and is library specific.
2. `get_subdag_nodes` -- Subdag nodes: Creating the `subdag` is outsourced to `recursive.subdag`, left flexibility to pre- and post-process since some libraries need that (see h_spark).
3. `chain_subdag_nodes` -- Merge node: The append functionality between dataframe and selected columns is library specific.

Each plugin library that can implement `with_columns` should subclass from this base class and implement the three abstract methods (four since `validate()` is also abstract). The child
classes need to override the `init` where they call out to the parent `init` and pass in `dataframe_type` which is registered in the corresponding `extensions` and has information of what
columns types are permitted for the given dataframe type.

Keeping it for now loosely coupled to the `registry` and detached from `ResultBuilder`. The API is private, should we want to switch to `registry`, the refactoring is straightforward and shouldn't get us into trouble down the road.

## NOTE
The handling of scalars and dataframe types varies between library to library. We made the decision that such a thing should not be permissible, so all the selected columns that want to be
appended to the original dataframe need to have the matching column type that is registered in the `registry` and set in the library extension modules.
