Function zero_array

benchmarks/src/bin/gen_wide_data.rs:108–127

Build a zero-filled array of the given datatype. Used for the suffix-renamed padding columns. Zero-filled rather than all-null so the parquet reader can't shortcut on null-array statistics — the wide-schema slowdown reproduces ~35 % wider with zeros than with nulls.

(dt: &DataType, n: usize)

/// wide-schema slowdown reproduces ~35 % wider with zeros than with
/// nulls.
fn zero_array(dt: &DataType, n: usize) -> ArrayRef {
    match dt {
        DataType::Int32 => {
            Arc::new(Int32Array::from_iter_values(std::iter::repeat_n(0i32, n)))
        }
        DataType::Int64 => {
            Arc::new(Int64Array::from_iter_values(std::iter::repeat_n(0i64, n)))
        }
        DataType::Float64 => Arc::new(Float64Array::from_iter_values(
            std::iter::repeat_n(0.0f64, n),
        )),
        DataType::Date32 => {
            Arc::new(Date32Array::from_iter_values(std::iter::repeat_n(0i32, n)))
        }
        DataType::Utf8 => {
            Arc::new(StringArray::from_iter_values(std::iter::repeat_n("", n)))
        }
        _ => panic!("zero_array: unsupported datatype {dt:?}"),
    }
}

/// Eight-column base schema. All fields nullable so the schema is
/// uniform across base and zero-filled replicated copies.