Alan D. Snow

@groutr, I think you have raised some valid points. I think the inf/nan changes are good to go. However, the empty array logic appears to need more thought/discussion. Mind moving...

> Never returns complex dtype. Mind moving these changes into a separate PR as well? I thin this topic is "complex" enough to be on its own :smile:

> What that is exactly isn't clear, but I took it to mean the these Your logic makes sense. However, since bool doesn't map to a GDAL dtype, this will...

Note one change for int8: ``` Numpy Rasterio GDAL ----- -------- ---- int8 Byte GDT_Byte (GDAL >=3.7; GDT_Int8) ``` Related: - #1595 - #1880 - #2656

Here is the order of how I am thinking we should proceed to move forward on the existing dtype PRs: ### 1: This PR looks ready to go once conflicts...

Related: - https://github.com/rasterio/rasterio/pull/3002 - https://github.com/shapely/shapely/issues/1972

Note: Forwarded from https://github.com/corteva/rioxarray/issues/681

@rbavery can you verify this does not work on your machine: ```python with rasterio.open('netcdf:/vsicurl/https://dap.ceda.ac.uk/neodc/esacci/land_surface_temperature/data/MULTISENSOR_IRCDR/L3S/0.01/v2.00/monthly/2020/11/ESACCI-LST-L3S-LST-IRCDR_-0.01deg_1MONTHLY_DAY-20201101000000-fv2.00.nc:lst') as rds: print(repr(rds.transform)) ``` ``` Affine(0.009999999728725104, 0.0, -179.99999511705187, 0.0, -0.009999999457435136, 89.9999874875217) ```

> I think the most interesting thing is that rasterio thinks that we should use the HDF5 driver to access the subdatasets I have run into this in the past...

I am not sure how I feel about `AWS_PATH_REGEX` as it isn't an official AWS environment variable. Also, I am wondering if instead maybe a list of supported domains should...