perf[vortex-array]: don't materialize const arrays in to_arrow_dictionary

[asubiotto]

Executing a Constant array into an arrow dictionary would materialize the full array and then perform the cast. Since we only have one value, we can just cast that value and create the codes array. This results in a 15% performance improvement on our real-world queries.

[joseph-isaacs]

Can we add a test that 1 its correct (equal to the expected value) and 2 that the values count of the array dict is 1

[asubiotto]

Is there a better way to match on vtables? The other thing I though of doing is an if/else chain on array.is

[joseph-isaacs]

Not now, but we should add this.

[joseph-isaacs]

happy to merge once you have those tests

[asubiotto]

TFTR! Will work on adding some tests

[codspeed-hq]

Merging this PR will degrade performance by 32.7%

❌ 1 regressed benchmark
✅ 1137 untouched benchmarks
⏩ 1265 skipped benchmarks¹

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Simulation	chunked_bool_into_canonical[(1000, 10)]	43 µs	63.9 µs	-32.7%

---

_{Comparing asubiotto/constantdict (00403b6) with develop (0d0fc63)}

Footnotes

1. 1265 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[joseph-isaacs]

Mind writing these test in terms of assert_eq(expected, actual)

#[test]
fn dict_to_dict_with_null_codes() -> VortexResult<()> {
    let dict = DictArray::try_new(
        PrimitiveArray::from_option_iter(vec![Some(0u8), None, Some(1)]).into_array(),
        VarBinViewArray::from_iter_str(["a", "b"]).into_array(),
    )?;
    let actual = execute(dict.into_array(), &dict_type(DataType::UInt8, DataType::Utf8))?;

    let expected: DictionaryArray<UInt8Type> =
        vec![Some("a"), None, Some("b")].into_iter().collect();

    assert_eq!(expected, actual.as_dictionary::<UInt8Type>().clone());
    Ok(())
}

[joseph-isaacs]

Its less code and tests more stuff

[asubiotto]

Good point, will also use rstest