Out-of-bounds assertion triggering on slices

[damacek]

Describe the bug

Recently, I encountered an assertion failure when passing a sliced RecordBatch to the Python FFI. This problem seems to pop out, only when working with slices of more RecordBatches containing complex data - nested lists and/or structs.

To Reproduce

Add a following test case to arrow-pyarrow-integration-testing/tests/test_sql.py:

def test_nested_struct_with_list_slice():
    """
    Test round-tripping sliced record batches with deeply nested struct types.
    
    This tests struct<struct<list<struct>>> with variable-length lists,
    ensuring that slicing at different row offsets works correctly.
    """
    # Build the nested type: struct<struct<list<struct>>>
    item_type = pa.struct([("x", pa.int64())])
    inner_struct_type = pa.struct([("items", pa.list_(item_type))])
    outer_struct_type = pa.struct([("inner", inner_struct_type)])

    # Key: variable-length inner lists (1, 2, 1 items)
    batch = pa.record_batch(
        [
            pa.array([1, 2, 3], type=pa.int64()),
            pa.array([
                {"inner": {"items": [{"x": 1}]}},
                {"inner": {"items": [{"x": 2}, {"x": 3}]}},
                {"inner": {"items": [{"x": 4}]}},
            ], type=outer_struct_type),
        ],
        names=["id", "outer"]
    )

    # Test round-trip of each sliced row
    for i in range(batch.num_rows):
        print(i)
        sliced = batch.slice(i, 1)
        result = rust.round_trip_record_batch(sliced)
        result.validate(full=True)
        assert result.to_pydict() == sliced.to_pydict()
        assert result.schema == sliced.schema

When I run pytest -v .:

        # Test round-trip of each sliced row
        for i in range(batch.num_rows):
            print(i)
            sliced = batch.slice(i, 1)
>           result = rust.round_trip_record_batch(sliced)
                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
E           pyo3_runtime.PanicException: assertion failed: (offset + length) <= self.len()

tests\test_sql.py:757: PanicException
------------------------------------------- Captured stdout call --------------------------------------------
0
1
------------------------------------------- Captured stderr call --------------------------------------------

thread '<unnamed>' (12544) panicked at C:\Code\arrow-rs\arrow-data\src\data.rs:581:9:
assertion failed: (offset + length) <= self.len()
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Expected behavior

The provided test should pass.

[Jefffrey]

Potentially related to

• Slicing a struct ArrayData with an offset double-counts the offset #7595
• Don't double-count the offset when slicing struct data from C++ #7596