Maybe just put _data = data in the case _ below? I'd actually prefer if we make it more explicit like:

case "arg" | "file" | "geojson" | "grid" | "stringio":
    _data = data

So that if there's a new kind in the future, we know to add it explicitly.

I'd actually prefer if we make it more explicit like:

case "arg" | "file" | "geojson" | "grid" | "stringio":
    _data = data

So that if there's a new kind in the future, we know to add it explicitly.

We can't do it like this, because we're currently writing case statements like case "image" if data.dtype != "uint8":, which only catches the case when kind="image" and data.dtype != "uint8".

We can refactor the codes to:

case "arg" | "file" | "geojson" | "grid" | "stringio":
    _data = data
case "image":
    _data = data
    if data.dtype != "uint8":
        # do something
case "matrix":
    _data = data
    if data.dtype.kind not in "iuf":
    # do something

or

case "arg" | "file" | "geojson" | "grid" | "stringio":
    _data = data
case "image" if data.dtype != "uint8":
    raise ...
case "matrix" if data.dtype.kind not in "iuf":
    ...
case _:
    _data = data # this is necessary to catch cases like kind="matrix" and dtype in "iuf"  

I still prefer the current codes, which set _data to data by default and the match-case statements only need to deal with exceptions.

So that if there's a new kind in the future, we know to add it explicitly.

If there is a new kind in the future, _data=data is still the default. If needs special handling and we forget to add it, then the related tests will fail anyway.

Ah ok, that makes sense, let's keep _data = data at the top then.