Benoit Bovy

Now a quick benchmark to evaluate the overhead of Python -> C++ conversion: ```cpp int dummy(PyObjectGeography obj) { // equivalent to `obj.cast()` where obj is a `py::object` obj.as_geog_ptr(); return 0;...

After more investigation, it looks like most of the conversion overhead is coming from calls to `pybind11::detail::get_type_info()` to get the C++ type from a Python type or vice-versa. This helper...

xref https://github.com/pybind/pybind11/discussions/4375 https://github.com/pybind/pybind11/issues/376

Nanobind has a low-level instance interface that could make things much easier: https://github.com/wjakob/nanobind/blob/master/docs/lowlevel.md. However, it doesn't support automatic vectorization (yet?).

Some naive questions as I'm not very familiar yet with (geo)arrow. IIUC https://github.com/paleolimbot/s2geography/pull/16 adds support for Arrow arrays so that we can easily reuse all the work done in geoarrow...

The implementation of `equals` ultimately relies on s2geometry's `S2BooleanOperation::Equals`, which provides a [precision](https://github.com/google/s2geometry/blob/713f9c27fed3085cc8dcf18a9d664c39227a0c45/src/s2/s2boolean_operation.h#L355) option. Unfortunately `Precision.SNAPPED` is not yet implemented, but I guess it would still be possible to add...

Hi @valeriupredoi, yes! We probably need to finish / get in #15 first, though.

This looks like a nice use case for the forthcoming Xarray's [custom index](https://github.com/pydata/xarray/blob/master/design_notes/flexible_indexes_notes.md) feature. How I see CRS/WCS-aware Xarray datasets with custom indexes: - A set of coordinate(s) and their...

We could also imagine ```python # returns a new dataset with both pixel and world (possibly lazy) coordinates >>> new_dataset = dataset.astro.append_world({'x': 'xw', 'y': 'yw', 'z': 'zw'}) # so that...

> In your opinion will this type of CRSIndex/WCSIndex work need #5322? If so, will it also require (or benefit from) the additional internal xarray refactoring you mention in #5322?...