Add `.data_view_info()` to `@render.data_frame` renderer

[schloerke]

[I'd like to have access to the selected rows]

Good news...

With a combination of

py-shiny/shiny/render/_data_frame.py

Lines 405 to 412 in 3b742a2

	@reactive.calc
	def self__input_data_view_indices() -> list[int]:
	data_view_indices = self._get_session().input[
	f"{self.output_id}_data_view_indices"
	]()
	return data_view_indices
	self._input_data_view_indices = self__input_data_view_indices

and

py-shiny/shiny/render/_data_frame.py

Lines 492 to 502 in 3b742a2

	cell_selection = self.input_cell_selection()
	if cell_selection is not None and cell_selection["type"] == "row":
	# Use a `set` for faster lookups
	selected_row_indices_set = set(cell_selection["rows"])
	# Subset the data view indices to only include the selected rows
	data_view_indices = [
	index
	for index in data_view_indices
	if index in selected_row_indices_set
	]

, maybe we expose them via .data_view_indices(*, selected: bool = False)? (Similar to .data_view(*, selected: bool = False))

It could return a {rows: list[int], cols: list[int]} object containing all rows/cols being used (given selected value). We could add a type to describe how the values were selected, similar to the CellSelection type

py-shiny/shiny/render/_data_frame_utils/_selection.py

Lines 162 to 185 in 3b742a2

	class CellSelectionRow(TypedDict):
	type: Literal["row"]
	rows: ListOrTuple[int]
	class CellSelectionCol(TypedDict):
	type: Literal["col"]
	cols: ListOrTuple[int]
	class CellSelectionRect(TypedDict):
	type: Literal["rect"]
	# Attempt to type the list of size 2, but it is not type enforced
	rows: tuple[int, int] | Annotated[list[int], 2]
	cols: tuple[int, int] | Annotated[list[int], 2]
	# For receiving selection info from JS:
	CellSelection = Union[
	CellSelectionRow,
	CellSelectionCol,
	CellSelectionRect,
	CellSelectionNone,
	]

.

For the return value, I still don't know if rows and cols should always exist (containing all values). If they always exist, it is easier to code as a user. If they do not exist, it is much clearer as to what is being subsetted but more annoying to integrate with as a user. I'm currently leaning towards always including rows and cols containing all values being used. Ex: multi row selection would return {type: "row", rows: [1,3,5], cols: [0,1,2,3,4,5]} assuming we selected rows 1, 3, and 5 and there were 6 columns.

class DataViewIndices(TypedDict):
    type: Literal["row", "col", "rect", "none"]
    rows: ListOrTuple[int]
    cols: ListOrTuple[int]

# class data_frame...
def data_view_indices(self, *, selected: bool = False) -> DataViewIndices:
    ...

---

@pstorozenko Glad you like .data_view()!

It's not a problem for now, but if one day we shiny switches to some dataframe representation without creature like pd.Index it may be.

Can I get a copy of your crystal ball? 😆

I'd love to move to something like https://github.com/data-apis/dataframe-api-compat , however to get typing support, their typing specification is defined deep within another repo: https://github.com/data-apis/dataframe-api/tree/main/spec/API_specification/dataframe_api .

I do not understand why the types being used within dataframe-api-compat are not shipped with the package. /rant

Originally posted by @schloerke in #1345 (comment)