Hello dear numpy community. I would like to propose an enhancement to the numpy.lib.recfunctions.structured_to_unstructured() function. I have submitted a PR already: <https://github.com/numpy/numpy/pull/23652> The PR also contains a short example. The structured_to_unstructured() function already returns views, if all elements in the structured array have the same dtype (or have sub-arrays of the same dtype) and are contiguous. I propose to relax the requirement of being contiguous, and instead only require that the offset between the fields is constant (so that it can be turned into a stride). This will for example allow performing structured_to_unstructured when selecting a subset of fields, e.g. from a vertex containing x, y, z, red, green, and blue (which would formerly return a copy): `structured_to_unstructured(vertex[['x', 'y', 'z']])` Additional changes: a) For views, ndarray (not an array subclass) is always returned: structured_to_unstructured() adds a new dimension to the array, but some subclasses behave strangely if that is done. For example, the numpy.matrix class always tries to keep 2 dimensions, which causes issues here. Maybe we want to change structured_to_unstructured() to always return an ndarray, even in the case of a copy? So far, the documentation does not specify if subclasses are preserved. b) The `base` attribute of the array will be more deeply nested than before. The only way I can think of to work around this, is to use the __array_interface__ directly.