Shuffle filter parameter has numcodecs incompatible dtype

[observingClouds]

What happened

I tried to open a zarr file with xarray that I have created with CDO (and the therefore the underlying netcdf-c library). Xarray failed to open the zarr file with:

File .venv/lib/python3.12/site-packages/numcodecs/shuffle.py:31, in Shuffle._prepare_arrays(self, buf, out)
     28 else:
     29     out = ensure_contiguous_ndarray(out)
---> 31 if self.elementsize <= 1:
     32     out.view(buf.dtype)[: len(buf)] = buf[:]  # no shuffling needed
     33     return buf, out

TypeError: '<=' not supported between instances of 'str' and 'int'

Minimal example

• Create sample dataset

   import xarray as xr
   import numpy as np
   example_ds = xr.DataArray(
       np.random.rand(2, 100, 100) * 10,
       dims=["time", "lat", "lon"],
       coords={"time": np.arange(2), "lat": np.arange(100), "lon": np.arange(100)},
       name="data",
   )
   example_ds.to_zarr("example_ds.zarr", mode="w")

• Compress with cdo

  cdo -z zip copy file://<full_path_to>/example_ds.zarr#mode=xarray file:///<full_path_to>/example_ds_cdo.zarr#mode=xarray

• Try loading with xarray

  import xarray as xr
  xr.open_zarr("example_ds_cdo.zarr")

The cause
Upon further inspection, the .zarray had the following JSON:

{ "zarr_format": 2,
  "shape": [760],
  "dtype": "<f8",
  "chunks": [760],
  "fill_value": null,
  "order": "C",
  "compressor": {
    "id": "zlib", 
    "level": "1"
  },
  "filters": [{"id": "shuffle", "elementsize": "0"}]}

The elementwise filter parameter is a string, while numcodecs expects an integer.

Potential solution

If I am not mistaken, the serialization happens at

netcdf-c/plugins/NCZhdf5filters.c

Line 90 in 947035a

snprintf(json,sizeof(json),"{\"id\": \"%s\", \"elementsize\": \"%u\"}",NCZ_shuffle_codec.codecid,typesize);

and should be modified to

"{\"id\": \"%s\", \"elementsize\": %u}"

to be compatible with numcodecs.

Workaround

When modifying the .zarray to {..."filters": [{"id": "shuffle", "elementsize": 0}]}, the loading with xarray works as expected but with values being incorrect.

Environment

netCDF 4.9.3-rc1
xarray: '2025.10.1'
numcodecs: '0.16.3'

[DennisHeimbigner]

I think you are correct. However, I note that right above that in the zlib compressor the level
is similarly set to a string "1" instead of 1. Does that load correctly for you.

[observingClouds]

There is no error raised for the zlib level. However, I should say that the values for e.g. lat and lon in the minimal example (just added above) are returned correctly with e.g. ncdump -v lat file:///<full_path_to>/example_ds_cdo.zarr#mode=xarray, but in xarray (even after converting the attributes level and elementwise to integers) the values are off:

ds = xr.open_zarr("example_ds_cdo.zarr")
ds.lat.values

array([ 0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
        0.00000000e+000,  0.00000000e+000,  0.00000000e+000,
       -6.95335581e-310, -6.95346191e-310, -6.95346191e-310,
       -6.95346191e-310, -5.20000244e+002, -5.20000367e+002,
       -5.20000367e+002, -5.20000367e+002, -5.20000367e+002,
...
        3.24921980e+001,  3.25019608e+001,  3.25019608e+001,
        3.25019608e+001,  3.25019608e+001,  3.25019608e+001,
        3.25019608e+001,  3.25019608e+001,  3.25019608e+001,
        3.25019608e+001,  3.25019608e+001,  3.25019608e+001,
        3.25019608e+001])

while ncdump -v lat file:///scratch/m/m300408/example_ds_cdo.zarr#mode=xarray works just fine:

...
data:

 lat = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 
    20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 
    38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 
    56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 
    74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 
    92, 93, 94, 95, 96, 97, 98, 99 ;
}

[DennisHeimbigner]

Can you send me those two files: example_ds.zarr and example_ds_cdo.zarr
in some single file form (e.g. .tar, .gz, .zip)?

[observingClouds]

Here are the example files:

netcdf-c_issue3200_examples.tgz

[DennisHeimbigner]

Fixed by PR #3201.
This also appears to fix the problem in #3200 (comment)