🐛 Bug Report

When saving a cube to a netCDF file using the NETCDF4_CLASSIC format, the valid_range attribute is unsafely cast.

How To Reproduce

Steps to reproduce the behaviour:

1. Create a cube with a 64-bit int valid_range attribute with a value greater than the maximum for a 32-bit int
2. Save the cube as netCDF using format NETCDF4_CLASSIC
3. Inspect the saved file, valid_range has been unsafely cast

import iris.fileformats
import iris.cube
import numpy as np
import netCDF4

if __name__ == "__main__":
    data = np.arange(5, dtype="int64")
    cube = iris.cube.Cube(data)
    cube.attributes["valid_range"] = [0, np.iinfo(cube.dtype).max]
    print(cube.attributes)
    iris.fileformats.netcdf.save(cube, "foo.nc", netcdf_format="NETCDF4_CLASSIC")

    foo = netCDF4.Dataset("foo.nc")
    print(foo)
    foo.close()

{'valid_range': [0, 9223372036854775807]}
...
<class 'netCDF4._netCDF4.Dataset'>
root group (NETCDF4_CLASSIC data model, file format HDF5):
    valid_range: [ 0 -1]
    Conventions: CF-1.7
    dimensions(sizes): dim0(5)
    variables(dimensions): int32 unknown(dim0)
    groups: 

Notice that the data has been cast to int32 as expected, but the valid_range has been corrupted.

Expected behaviour

An error should be raised to inform that the valid_range cannot be safely cast to an allowed netCDF classic type. This would be similar to the current behaviour when a too-large value appears in the data:

import iris.fileformats
import iris.cube
import numpy as np

if __name__ == "__main__":
    data = np.arange(5, dtype="int64")
    cube = iris.cube.Cube(data)
    cube.data[-1] = np.iinfo(cube.dtype).max
    iris.fileformats.netcdf.save(cube, "foo.nc", netcdf_format="NETCDF4_CLASSIC")

ValueError: The data type of cube <iris 'Cube' of unknown / (unknown) (-- : 5)> is not supported by NETCDF4_CLASSIC and its values cannot be safely cast to a supported integer type.

Screenshots

Environment

• OS & Version: RHEL7
• Iris Version: 3.10.0