issue_comments: 622342323

Currently, the error that is raised by coerce_nc3_dtype is not covered by tests.

A test could look something like this:

```python def test_dtype_coercion_error(self): for dtype in _nc3_dtype_coercions: if dtype == "bool": continue

        # Create some data that cannot be coerced (cause it's a downcast)
        dtype = np.dtype(dtype)
        maxval = np.iinfo(dtype).max
        x = np.array([0, 1, 2, maxval], dtype=dtype)
        test_data = Dataset({"x": ("t", x, {})})

        # Make sure a roundtrip fails
        with pytest.raises(ValueError, match="could not safely cast"):
            self.roundtrip(test_data)

```

... but I'm not sure ... * whether this would be desirable to test * whether this PR should take care of this or not * where in test_backend.py I would have to put the test case such that it's invoked only for netcdf3. * I already found the NetCDF3Only class and requires_scipy decorator, but don't know which test class is responsible here ...