Enum mz_repr::row::encode::DatumColumnEncoder

enum DatumColumnEncoder {
    Bool(BooleanBuilder),
    U8(UInt8Builder),
    U16(UInt16Builder),
    U32(UInt32Builder),
    U64(UInt64Builder),
    I16(Int16Builder),
    I32(Int32Builder),
    I64(Int64Builder),
    F32(Float32Builder),
    F64(Float64Builder),
    Numeric {
        binary_values: BinaryBuilder,
        approx_values: Float64Builder,
        numeric_context: Context<Numeric>,
    },
    Bytes(BinaryBuilder),
    String(StringBuilder),
    Date(Int32Builder),
    Time(FixedSizeBinaryBuilder),
    Timestamp(FixedSizeBinaryBuilder),
    TimestampTz(FixedSizeBinaryBuilder),
    MzTimestamp(UInt64Builder),
    Interval(FixedSizeBinaryBuilder),
    Uuid(FixedSizeBinaryBuilder),
    AclItem(FixedSizeBinaryBuilder),
    MzAclItem(BinaryBuilder),
    Range(BinaryBuilder),
    Jsonb {
        offsets: Vec<i32>,
        buf: Vec<u8>,
        nulls: Option<BooleanBufferBuilder>,
    },
    Array {
        dims: ListBuilder<FixedSizeBinaryBuilder>,
        val_lengths: Vec<usize>,
        vals: Box<DatumColumnEncoder>,
        nulls: Option<BooleanBufferBuilder>,
    },
    List {
        lengths: Vec<usize>,
        values: Box<DatumColumnEncoder>,
        nulls: Option<BooleanBufferBuilder>,
    },
    Map {
        lengths: Vec<usize>,
        keys: StringBuilder,
        vals: Box<DatumColumnEncoder>,
        nulls: Option<BooleanBufferBuilder>,
    },
    Record {
        fields: Vec<DatumEncoder>,
        nulls: Option<BooleanBufferBuilder>,
        length: usize,
    },
    RecordEmpty(BooleanBuilder),
}

An encoder for a single column of Datums. To encode an entire row see RowColumnarEncoder.

Note: We specifically structure the encoder as an enum instead of using trait objects because Datum encoding is an extremely hot path and downcasting objects is relatively slow.

Variants

Bool(BooleanBuilder)

U8(UInt8Builder)

U16(UInt16Builder)

U32(UInt32Builder)

U64(UInt64Builder)

I16(Int16Builder)

I32(Int32Builder)

I64(Int64Builder)

F32(Float32Builder)

F64(Float64Builder)

Numeric

Fields

binary_values: BinaryBuilder

The raw bytes so we can losslessly roundtrip Numerics.

approx_values: Float64Builder

Also maintain a float64 approximation for sorting.

numeric_context: Context<Numeric>

Re-usable libdecimal context for conversions.

Bytes(BinaryBuilder)

String(StringBuilder)

Date(Int32Builder)

Time(FixedSizeBinaryBuilder)

Timestamp(FixedSizeBinaryBuilder)

TimestampTz(FixedSizeBinaryBuilder)

MzTimestamp(UInt64Builder)

Interval(FixedSizeBinaryBuilder)

Uuid(FixedSizeBinaryBuilder)

AclItem(FixedSizeBinaryBuilder)

MzAclItem(BinaryBuilder)

Range(BinaryBuilder)

Jsonb

Hand rolled “StringBuilder” that reduces the number of copies required to serialize JSON.

An alternative would be to use StringBuilder but that requires serializing to an intermediary string, and then copying that intermediary string into an underlying buffer.

Fields

offsets: Vec<i32>

Monotonically increasing offsets of each encoded segment.

buf: Vec<u8>

Buffer that contains UTF-8 encoded JSON.

nulls: Option<BooleanBufferBuilder>

Null entries, if any.

Array

Fields

dims: ListBuilder<FixedSizeBinaryBuilder>

Binary encoded ArrayDimensions.

val_lengths: Vec<usize>

Lengths of each Array in this column.

vals: Box<DatumColumnEncoder>

Contiguous array of underlying data.

nulls: Option<BooleanBufferBuilder>

Null entires, if any.

List

Fields

lengths: Vec<usize>

Lengths of each List in this column.

values: Box<DatumColumnEncoder>

Contiguous array of underlying data.

nulls: Option<BooleanBufferBuilder>

Null entires, if any.

Map

Fields

lengths: Vec<usize>

Lengths of each Map in this column

keys: StringBuilder

Contiguous array of key data.

vals: Box<DatumColumnEncoder>

Contiguous array of val data.

nulls: Option<BooleanBufferBuilder>

Null entires, if any.

Record

Fields

fields: Vec<DatumEncoder>

Columns in the record.

nulls: Option<BooleanBufferBuilder>

Null entries, if any.

length: usize

Number of values we’ve pushed into this builder thus far.

RecordEmpty(BooleanBuilder)

Special encoder for a ScalarType::Record that has no inner fields.

We have a special case for this scenario because Arrow does not allow a StructArray (what normally use to encod a Record) with no fields.

Implementations

impl DatumColumnEncoder

fn goodbytes(&self) -> usize

fn push<'e, 'd>(&'e mut self, datum: Datum<'d>)

fn push_invalid(&mut self)

fn finish(self) -> ArrayRef

Trait Implementations

impl Debug for DatumColumnEncoder

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.