Do you know why there are no changes to the backend arrays yet? Those too use *Indexer objects, and we want to break that out. However now I really there's a backwards compatibility problem there. Can you take a look at cfgrib/rioxarray/tiledb-python to see how they use these lazy arrays and indexing please?

Can you take a look at cfgrib/rioxarray/tiledb-python to see how they use these lazy arrays and indexing please?

after taking a look at the respective repos, it's my understanding both the rioxarray and cfrib projects use the explicit_indexing_adapter function. this function acts a bridge, connecting high-level *Indexer objects with backend-specific indexing methods. thanks to each backend wrapping its BackendArray within a LazilyIndexedArray (which supports the oindex[]), everything functions smoothly since we are still passing *Indexer objects within the xr.Variable.__getitem__. however, i anticipate that modifications will be necessary when we begin to directly pass the underlying tuple to methods like __getitem__, oindex, and vindex within xr.Variable.__getitem__

on the other hand, TileDB-CF-Py seems to bypass this pattern, handling the getitem method differently by using its unique logic for processing the underlying tuple. It still wraps its BackendArray within a LazilyIndexedArray though.

• https://github.com/TileDB-Inc/TileDB-CF-Py/blob/2c041de8e5f6ae585518a7e5276a6f0a5daee3bc/tiledb/cf/xarray_engine/_array_wrapper.py#L153

• https://github.com/TileDB-Inc/TileDB-CF-Py/blob/2c041de8e5f6ae585518a7e5276a6f0a5daee3bc/tiledb/cf/xarray_engine/_backend_store.py#L83-L99

• https://github.com/corteva/rioxarray/blob/e33616e6b35076749605d5c8d0ff7908e257c2a1/rioxarray/_io.py#L452C1-L456C1

• https://github.com/ecmwf/cfgrib/blob/f95912c838a42fbfc26312927a28c8305258ddb9/cfgrib/xarray_plugin.py#L153-L159

cc @dcherian