[aaronkanzer]

Slicing an NXfield with a negative step works when the data are in memory. It fails once the data are in a file, or in the HDF5 core memory file used for large fields. So whether field[::-1] works depends on the size of the array.

import numpy as np
from nexusformat.nexus import NXroot, NXentry, NXfield, nxload

a = np.arange(2000 * 1000.).reshape(2000, 1000)
NXroot(NXentry(NXfield(a, name="x"))).save("big.nxs", "w")
x = nxload("big.nxs")["entry/x"]
x[::-1]

Observed: ValueError: Step must be >= 1 (got -1). Writes such as x[::-2] = ... on a file opened with "rw" fail the same way. So does NXfield(shape=(200, 100), dtype=float)[::-1], because fields larger than maxsize are kept in an HDF5 core file.

Expected: the same result as a[::-1], which is what a small array returns.

Cause: NXFile.readvalue, NXFile.writevalue, _get_memdata and _put_memdata pass the index straight to h5py. HDF5 hyperslabs only accept positive steps.

Fix: a small helper turns each slice with a negative step into the matching slice with a positive step. It reads or writes that slice, then reverses the affected axes with NumPy. Indices without a negative step, and fancy indices, still go to h5py unchanged. Tests cover reads and writes on file-backed and memory-file fields. The helpers were also checked against NumPy on 5000 random int, slice and Ellipsis indices, with no mismatches.