How to reverse the order of an array along one or several dimensions?

[jokasimr]

What is the equivalent of np.flip(...) or arr[::-1] in scipp?

Maybe we should make this easier to find in the docs.

[YooSunYoung]

Is sort enough for your use-case?

import scipp as sc

da = sc.data.data_xy()
sc.sort(da, 'x', order='descending')

[jokasimr]

Is sort enough...

No not really. I don't want the entries sorted, I just want them in revered order.
Like this: [1, 3, 2] -> [2, 3, 1]

[jl-wynen]

What is the utility of adding a named function? As I see it, the only reason to have np.flip is to specify an axis by rank programmatically. E.g., np.flip(a, n) where n is not a constant cannot easily be done using slicing syntax. But in Scipp, we require dimension labels when slicing. So v[d, ::-1] works when d is not a constant.

So I would stick with the (almost) standard Python syntax of [::-1].

We talked about collecting usage patterns at some point. This could be added there if we actually write them down.

But can I ask you, why did this come up?

[jokasimr]

What is the utility of adding a named function? As I see it, the only reason to have np.flip is to specify an axis by rank programmatically. E.g., np.flip(a, n) where n is not a constant cannot easily be done using slicing syntax. But in Scipp, we require dimension labels when slicing. So v[d, ::-1] works when d is not a constant.

Sure, but:

sc.arange('x', 0, 10)['x', ::-1]
# IndexError: negative slice step support not implemented

It came up when I translated some numpy code to scipp.
I'm not exactly sure why the reverse was required.

[nvaytet]

But can I ask you, why did this come up?

Isn't it because we don't support negative strides?

[SimonHeybrock]

Supporting negative strides should be the way to go. Note #3188, which would probably make this much easier.