Benoit Bovy

> we can fix that in safe_cast_to_index() ...we *cannot* fix that in `safe_cast_to_index()` (or we can add a parameter to specify the desired result).

FWIW, although not Arrow-specific, the "Array of rasters" model suggested by @paleolimbot looks very much similar to @TomAugspurger's [experiment](https://discourse.pangeo.io/t/tables-x-arrays-and-rasters/1945) using Pandas & Xarray (i.e., a "pandas `RasterArray` of xarray `DataArray`s").

Thanks @TomNicholas. There is a possible fix for this implemented in #8124. I've hit some general issues with the approach taken in that PR (described in the top comment) but...

> I'm wondering if the change inside xr.merge [on this line](https://github.com/pydata/xarray/commit/26fa16a36ba04d6aaf4e6ad6f4670c9b5be0a9d4#diff-33bf32762d44e3a7b30f5695056dd49acbf9fc3a15d4cc31c1a5605f9af7ba9fR358) might also fix the other problem I'm having with indexes being automatically created - see the final error in...

Nice @ilan-gold, thanks for working on this! I haven't looked into the details yet, but I'm wondering if `ExtensionDuckArray` wouldn't be better implemented as one of the array wrapper classes...

> But I think the rest of the PR stands - we still need to do checking for dtypes as I said and will probably want some dispatch/registry mechanism for...

Sorry the two commits picked from #8124 where done before #8107 and messed up things a bit. I also picked and added the last merge commit from #8124, which should...

About the `auto_convert` option added to `as_variable()` here: this is a temporary workaround before we make a deeper and better refactoring of non-index vs. index coordinate variables. That is not...

> why is `as_variable` public API? More context in #1303

> I mean I don't think that's a huge deal Me neither (I wanted to see what others think). @TomNicholas have you tried running your notebook mentioned in https://github.com/pydata/xarray/issues/8704#issuecomment-1927264039 with...