zip_no_broadcast int & string array raise ValueError: can not (unsafe) zip ListOffsetArrays with non-NumpyArray contents

[ruoyu0088]

Version of Awkward Array

2.8.10

Description and code to reproduce

The following code in the zip_no_broadcast() document raise ValueError:

import awkward as ak
one = ak.Array([[1.1, 2.2, 3.3], [], [4.4, 5.5], [6.6]])
two = ak.Array([["a", "b", "c"], [], ["d", "e"], ["f"]])
ak.zip_no_broadcast({"x": one, "y": two}).show()

the exception:

File [Lib\site-packages\awkward\operations\ak_zip_no_broadcast.py:166], in _impl(arrays, parameters, with_name, highlevel, behavior, attrs)
    163 for layout in layouts:
    164     # get the content of the ListOffsetArray
    165     if not isinstance(layout.content, ak.contents.NumpyArray):
--> 166         raise ValueError(
    167             "can not (unsafe) zip ListOffsetArrays with non-NumpyArray contents"
    168         )
    169     contents.append(layout.content)
    171 if backend.name == "typetracer":
    172     # just get from the first one
    173     # we're in typetracer mode, so we can't check the offsets (see else branch)

ValueError: can not (unsafe) zip ListOffsetArrays with non-NumpyArray contents

[ikrommyd]

This is not exactly wrong as the array two in your example above indeed does not have a numpy array content. It has a ListOffsetArray content due to how string arrays are implemented in awkward

In [8]: two = ak.Array([["a", "b", "c"], [], ["d", "e"], ["f"]])

In [9]: two.layout
Out[9]:
<ListOffsetArray len='4'>
    <offsets><Index dtype='int64' len='5'>[0 3 3 5 6]</Index></offsets>
    <content><ListOffsetArray len='6'>
        <parameter name='__array__'>'string'</parameter>
        <offsets><Index dtype='int64' len='7'>[0 1 2 3 4 5 6]</Index></offsets>
        <content><NumpyArray dtype='uint8' len='6'>
            <parameter name='__array__'>'char'</parameter>
            [ 97  98  99 100 101 102]
        </NumpyArray></content>
    </ListOffsetArray></content>
</ListOffsetArray>

A string array here is a listoffset array where the offsets tell you which characters represented as integers belong into each string. Look at this for example

In [11]: array = ak.Array(["abc", "defg", "", "h", "ij"])

In [12]: array.layout
Out[12]:
<ListOffsetArray len='5'>
    <parameter name='__array__'>'string'</parameter>
    <offsets><Index dtype='int64' len='6'>[ 0  3  7  7  8 10]</Index></offsets>
    <content><NumpyArray dtype='uint8' len='10'>
        <parameter name='__array__'>'char'</parameter>
        [ 97  98  99 100 101 102 103 104 105 106]
    </NumpyArray></content>
</ListOffsetArray>

The raw characters are here as integers [97 98 99 100 101 102 103 104 105 106] and the offsets tell the array which characters belong in each string [0 3 7 7 8 10]. So from 0 to 3 (excluding 3) are the first string "abc". 3 to 7 the second string "defg" and so on.
I don't know whether it's worth having a specific code path for string arrays in zip_no_broadcast. @pfackeldey I'll leave this to you since you wrote this function.

[pfackeldey]

Hi @ruoyu0088,

ak.zip_no_broadcast is intentionally scoped because many zip's without broadcasting are unsafe. That's why you see this error - this is on purpose.
Please use ak.zip instead, as I currently do not plan to extend to coverage of ak.zip_no_broadcast.

[ianna]

@ruoyu0088 - Thanks for submitting the issue. Indeed, this case is very special. Please, let me know if you would like me to mark it as a feature request. In the meantime, I think, this PR #3743 provides a more clear message. Thanks.