Removal ak.layout.ListArray32 etc.

[tamasgal]

Sorry for my ignorance, I could not follow the Awkward developments in the past months, so this might be something trivial. I will investigate further today but I wanted to dump this in case someone else encounters the same issue.

As discussed here a few years ago, I am using ak.layout.ListArray32 etc. in a type check like below where I also included the current error:

    def recurse(layout):
        if layout.purelist_depth == 2:
            if startend is not None:
                np_array = _mask_startend(ak.Array(layout), *startend)
            elif minmax is not None:
                np_array = _mask_minmax(ak.Array(layout), *minmax)
            elif sequence is not None:
                np_array = _mask_sequence(ak.Array(layout), np.array(sequence))
            elif atleast is not None:
                np_array = _mask_atleast(ak.Array(layout), np.array(atleast))

            return ak.layout.NumpyArray(np_array)

        elif isinstance(
            layout,
            (
>               ak.layout.ListArray32,
                ak.layout.ListArrayU32,
                ak.layout.ListArray64,
            ),
        ):
E       AttributeError: module 'awkward' has no attribute 'layout'

I am currently updating a library which still uses Awkward v1 to v2 (multiple reasons, including NumPy v2) and apparently ak.layout does not exist anymore.

Is there a quick workaround or is a rabbit hole thing? 😆 I see that the type is now awkward.contents.listoffsetarray.ListOffsetArray so I assume I need to check for layout.content.dtype. What about ak.layout.RegularArray?

>>> km3io.tools.best_jmuon(f.trks[1])
> /Users/tamasgal/Dev/km3io/src/km3io/tools.py(301)recurse()
    295         raise ValueError(
    296             "either sequence, startend, minmax or atleast must be specified."
    297         )
    298
    299     def recurse(layout):
    300         import ipdb; ipdb.set_trace(context=13)
--> 301         if layout.purelist_depth == 2:
    302             if startend is not None:
    303                 np_array = _mask_startend(ak.Array(layout), *startend)
    304             elif minmax is not None:
    305                 np_array = _mask_minmax(ak.Array(layout), *minmax)
    306             elif sequence is not None:
    307                 np_array = _mask_sequence(ak.Array(layout), np.array(sequence))

ipdb> layout
<ListOffsetArray len='14'>
    <offsets><Index dtype='int64' len='15'>
        [336 337 339 342 345 348 351 354 357 360 363 366 369 372 375]
    </Index></offsets>
    <content><NumpyArray dtype='int32' len='239488'>[  1   2   3 ... 302 303 304]</NumpyArray></content>
</ListOffsetArray>
ipdb> type(layout)
<class 'awkward.contents.listoffsetarray.ListOffsetArray'>

[agoose77]

The layout module was previously a PyBind11 exposed C++ module. now, ak.contents is all implemented in Python. So, most of the time you can just substitute ak.layout.XXX for ak.contents.XXX. However, the Awkward classes no longer encode the index type; this comes from the buffer. So the final piece of the puzzle is ListArray32 -> ListArray with an index of ak.index.Index32.

[tamasgal]

Thanks! I will investigate and report back :)

[tamasgal]

I can confirm that simply replacing ak.layout. with ak.contents. solves the issue. In my case, I realised that I don't need to differentiate between the element types, it resolve itself downstreams ;)

That's the modified code for the sake of completeness:

    def recurse(layout):
        if layout.purelist_depth == 2:
            if startend is not None:
                np_array = _mask_startend(ak.Array(layout), *startend)
            elif minmax is not None:
                np_array = _mask_minmax(ak.Array(layout), *minmax)
            elif sequence is not None:
                np_array = _mask_sequence(ak.Array(layout), np.array(sequence))
            elif atleast is not None:
                np_array = _mask_atleast(ak.Array(layout), np.array(atleast))

            return ak.contents.NumpyArray(np_array)

        elif isinstance(layout, ak.contents.ListArray):
            if len(layout.stops) == 0:
                content = recurse(layout.content)
            else:
                content = recurse(layout.content[: np.max(layout.stops)])
            return type(layout)(layout.starts, layout.stops, content)

        elif isinstance(layout, ak.contents.ListOffsetArray):
            content = recurse(layout.content[: layout.offsets[-1]])
            return type(layout)(layout.offsets, content)

        elif isinstance(layout, ak.contents.RegularArray):
            content = recurse(layout.content)
            return ak.contents.RegularArray(content, layout.size)

        else:
            raise NotImplementedError(repr(arr))