Linked Selections Generalization

[philippjfr]

The link_selections functionality is one of the most powerful features in HoloViews and while it is tightly coupled with HoloViews internals, the basic concept is actually quite general. Specifically the core of it involves taking selections and applying them to different datasets, something that could easily be leveraged by other libraries, such as Panel, to allow selections on plots and other visual components to be synced.

The main things that are missing to make this easier are a few APIs to set this up manually:

• update_selection(name: str, dimension: str | Dimension, selection: tuple[number, number] | list[any] | constant): An API to manually add a selection along a particular dimension. By giving each selection a unique name they can be updated easily.
• compute_mask(dataframe: DataFrameLike | Dataset): Returns a set of indices of the selected rows

These two APIs would be sufficient for components in Panel and other libraries to use link_selections more effectively. Beyond that we could consider implementing a plugin API where external components can register handlers that would make it possible to call link_selections on objects of a particular type, e.g. making this possible:

table = pn.widgets.Tabulator(df)

ls = hv.link_selections.instance()

ls(table)

Internally this would effectively do the following:

def setup_tabulator(ls, table):

    def sync_selection(e):
        ls.update_selection(e.obj.name, '__index__', e.new)

    table.param.watch(sync_selection, 'selection')

    def update_selection(e):
        table.selection = ls.compute_mask(table.value)

    ls.param.watch(update_selection, 'selection_expr')

hv.link_selections.register(pn.widgets.Tabulator, setup_tabulator)