Why filtering on regular array works differently?

[wiso]

I read some data from uproot and the resulting array is represented as a regular array

values.layout is

<RegularArray size='1024' len='1104'>
    <content><NumpyArray dtype='uint16' len='1130496'>[  0   0   0 ... 100  98  98]</NumpyArray></content>
</RegularArray>

Values are

[[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ..., 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ..., 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
 [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, ..., 97, 85, 93, 100, 76, 108, 85, 94, 109, 87]]
--------------------------------------------------------------------------------
backend: cpu
nbytes: 2.3 MB
type: 3 * 1024 * uint16

if I apply a filter

values[values != 0] I get a 1D array

which is not what I expect.

I solved with

values = ak.from_regular(values)

but this seems awkward

[ianna]

@wiso - Thanks for reporting it! Conceptually, this is something Awkward could (and arguably should) handle at the slicing layer:

When applying a jagged mask to a RegularArray, the result could be “promoted” to a list-like structure instead of flattened.

That would make values[values != 0] always behave consistently.

So yes — you’re absolutely right it is awkward:

• users should not need to care about RegularArray.
• The slicing interface is the right abstraction layer to hide it.