Skip to content

Instantly share code, notes, and snippets.

@mdsumner
Created January 19, 2026 11:26
Show Gist options
  • Select an option

  • Save mdsumner/7e040866fc9113ea41dbebfadeeb88dc to your computer and use it in GitHub Desktop.

Select an option

Save mdsumner/7e040866fc9113ea41dbebfadeeb88dc to your computer and use it in GitHub Desktop.
I'm working on a GDAL backend for xarray: https://github.com/mdsumner/gdx I've figured out a bunch of issues that should make it perform better with dask, still early but if anyone is interested I'm using it with CMEMs Zarr stores that I catalogue with https://github.com/hypertidy/cmemsarco This combo for me is (finally!) a clean set of features that I wanted for GDAL compared to Zarr/xarray. I'm now proficient enough to actually get into the details of dask, and I've seen GDAL be more efficient with virtualized remote stores like Thredds (I think connection pooling is smarter, but of course this is stuff that Icechunk is racing ahead with too in yet another format variant so we'll see how things play out). I don't really see any Zarr work in R, but GDAL provides a very good foundation (we don't need fsspec or numcodecs etc because GDAL has those as part of its Abstraction) and gdalraster has a good framework for leveraging that. Multidimensional is really overkill for something simple like a regular array of daily data but we can't ignore the array world they will steadily replace every format or virtualize them. There will ultimately need to be a reallibrary for xarray or whatever is good enough to replace it, an actual foundation for cross lang software.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment