Function partial_cmp_list

datafusion/common/src/scalar/mod.rs:795–833

Compares two List/LargeList/FixedSizeList/ListView/LargeListView scalars

(arr1: &dyn Array, arr2: &dyn Array)

/// Compares two List/LargeList/FixedSizeList/ListView/LargeListView scalars
fn partial_cmp_list(arr1: &dyn Array, arr2: &dyn Array) -> Option<Ordering> {
    if arr1.data_type() != arr2.data_type() {
        return None;
    }
    let arr1 = first_array_for_list(arr1);
    let arr2 = first_array_for_list(arr2);

    let min_length = arr1.len().min(arr2.len());
    let arr1_trimmed = arr1.slice(0, min_length);
    let arr2_trimmed = arr2.slice(0, min_length);

    let lt_res = arrow::compute::kernels::cmp::lt(&arr1_trimmed, &arr2_trimmed).ok()?;
    let eq_res = arrow::compute::kernels::cmp::eq(&arr1_trimmed, &arr2_trimmed).ok()?;

    for j in 0..lt_res.len() {
        // In Postgres, NULL values in lists are always considered to be greater than non-NULL values:
        //
        // $ SELECT ARRAY[NULL]::integer[] > ARRAY[1]
        // true
        //
        // These next two if statements are introduced for replicating Postgres behavior, as
        // arrow::compute does not account for this.
        if arr1_trimmed.is_null(j) && !arr2_trimmed.is_null(j) {
            return Some(Ordering::Greater);
        }
        if !arr1_trimmed.is_null(j) && arr2_trimmed.is_null(j) {
            return Some(Ordering::Less);
        }

        if lt_res.is_valid(j) && lt_res.value(j) {
            return Some(Ordering::Less);
        }
        if eq_res.is_valid(j) && !eq_res.value(j) {
            return Some(Ordering::Greater);
        }
    }

    Some(arr1.len().cmp(&arr2.len()))
}

fn flatten<'a>(array: &'a StructArray, columns: &mut Vec<&'a ArrayRef>) {
    for i in 0..array.num_columns() {