My only real qualm with this is now mixing vstack with column_stack... The latter is kept because it has different semantics (particularly for 1-D arrays) than hstack

That makes me wonder if arguing for keeping row_stack is a good idea (it is just an alias, but it is a bit nicer to have consistent names in the same part of the code, sooo)

Alternatively, we could rework that code to use hstack (which may involve adding dimensions to some 1-d arrays, haven't looked too closely)

Hi! The row_stack is being removed as one of the NEP 52 goals is have each function in exactly one place (so aliases and multiple locations of the same function will be removed).
That's correct, to get the same semantics for column_stack and hstack for 1d arrays it would require:

a = np.array([1,2,3])
np.hstack([a[:,None], a[:,None]])  # np.column_stack([a, b])

numpy/numpy#24445 is looking ready to merge, does this still need work?