Enum Form 

#[non_exhaustive]
pub enum Form {
    Flat,
    Constant,
    Sequence,
    Dictionary,
    BitPacked,
    Rle,
}

Which physical form a vector is in.

An operator asks this once per vector and then takes the path it wants, which is the one branch per vector that the whole design is willing to spend.

Not exhaustive, and that is a decision rather than an oversight. Encoded is the fifth form and it arrives at layer three with the specialization contract. If this enum were exhaustive, the day it lands is the day every kernel in the workspace stops compiling, and the pressure at that moment would be to add an arm to each of them in a hurry rather than to think about what each one should do with an encoded vector. A required fallback arm means each kernel already has a correct answer for a form it has never seen, and specializing it is then a change that can be made one kernel at a time with a benchmark next to it.

Variants (Non-exhaustive)

Non-exhaustive enums could have additional variants added in future. Therefore, when matching against variants of non-exhaustive enums, an extra wildcard arm must be added to account for any future variants.

Flat

One value per position.

Constant

One value, repeated.

Sequence

A start and a step, computed rather than stored.

Dictionary

Codes into a smaller vector of distinct values.

BitPacked

Integers stored in as many bits as the range of the column needs, offset from a base.

The form a narrow integer column is in. A ClickBench ResolutionWidth is a SMALLINT whose values live between 0 and 2560, which is twelve bits, so the column is three quarters of the size it was and the pages behind it are three quarters of the reads. What it costs is a shift and a mask per value, which is why this is worth it at storage and at rest and is not a form anything should be building in the middle of a pipeline.

Rle

One value per run, with the row each run ends at.

The form a clustered column is in. hits is written in time order, so EventDate is a few hundred runs over a hundred million rows, and a sum over it is a few hundred multiplications rather than a hundred million additions. Dictionary says which distinct values there are and this says where they stop, and a column can want either one without wanting the other.

Trait Implementations

impl Clone for Form

fn clone(&self) -> Form

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for Form

impl Debug for Form

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

Formats the value using the given formatter.

impl Eq for Form

impl Hash for Form

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for Form

fn eq(&self, other: &Form) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for Form