Nested Array type inspection, best practices

[gipert]

I'm working on a routine that accepts only a specific type of Array, e.g.

>>> ak.Array([[[1, 2], [3]], [[7], []], [[9], [3, 8, 5]]])
<Array [[[1, 2], [3]], ..., [[9], [3, ...]]] type='3 * var * var * int64'>

strictly characterized by:

• generic dimensionality
• fully jagged
• non-record-type
• non-option-type

each of these features must be present at each axis.

I'm looking for a way to check the array sanity (i.e. the presence of all these characteristics) at all depths at the same time. I presume one needs to recursively walk the array structure? If yes, what is the recommended way to do so?

[agoose77]

Awkward Arrays have three levels of type information:

1. Types ("this is a ragged list", "this is a regular dimension", "this is an option", ...)
2. Forms ("this is a ragged list with starts and stops", ...)
3. Contents ("this is a ragged list with starts and stops whose values are ...", ...)

It sounds like you only care about the first one, ak.Array.type.

We would write this kind of test using recursion, e.g.

import awkward as ak


def assert_valid(type_):
    if isinstance(type_, ak.types.ArrayType):
        return assert_valid(type_.content)
    elif isinstance(type_, ak.types.ScalarType):
        raise TypeError("Expected ArrayType or its content")

    assert not isinstance(
        type_,
        (
            ak.types.RegularType,
            ak.types.OptionType,
            ak.types.UnionType,
            ak.types.UnknownType,
            ak.types.RecordType,
        ),
    )
    if isinstance(type_, ak.types.ListType):
        assert_valid(type_.content)
    else:
        assert isinstance(type_, ak.types.NumpyType)


array = ak.Array([[1, 2, 3]])
assert_valid(array.type)

These type objects form a closed set, so we can exhaustively test each one. You should take a look at the type object of an array to see what its properties are!

Does this enforce the rules you care about?

[gipert]

Seems so thanks!

Just as a clarification, why wouldn't this:

def assert_valid(type_):
    if isinstance(type_, ak.types.ArrayType | ak.types.ListType):
        return assert_valid(type_.content)
    elif isinstance(type_, ak.types.ScalarType):
        raise TypeError("Expected ArrayType or its content")

    assert not isinstance(
        type_,
        (
            ak.types.RegularType,
            ak.types.OptionType,
            ak.types.UnionType,
            ak.types.UnknownType,
            ak.types.RecordType,
            ak.types.NumpyType,
        ),
    )

be equivalent to your function?

[agoose77]

It would be! The way I wrote the function reflects my thinking on our type hierarchy: ArrayType and ScalarType are top-level types that shouldn't be seen inside a type tree.