Enum AdaptiveSelection 

pub enum AdaptiveSelection {
    Empty {
        nrows: usize,
    },
    All {
        nrows: usize,
    },
    SelectionVector {
        rows: Vec<u32>,
        nrows: usize,
    },
    VerbatimMask {
        mask: BitMask,
    },
    Hybrid {
        nrows: usize,
        chunks: Vec<HybridChunk>,
    },
}

A row-selection representation chosen adaptively by density.

All variants carry nrows (the size of the underlying DataFrame) so that count/contains/iteration are answerable without consulting the base.

Variants

Empty

No rows selected. nrows is the underlying frame size.

Fields

nrows: usize

All

Every row selected.

Fields

nrows: usize

SelectionVector

Sparse: ascending u32 row indices.

Fields

rows: Vec<u32>

nrows: usize

VerbatimMask

Dense: backing BitMask.

Fields

mask: BitMask

Hybrid

Mid-density chunked representation. chunks.len() == ceil(nrows / HYBRID_CHUNK_SIZE).

Fields

nrows: usize

chunks: Vec<HybridChunk>

Implementations

impl AdaptiveSelection

pub fn all(nrows: usize) -> Self

All rows selected.

pub fn empty(nrows: usize) -> Self

No rows selected.

pub fn from_bitmask(mask: BitMask) -> Self

Wrap a BitMask directly (force the dense arm). Used by call sites that already produced a mask through the existing pipeline and do not want a re-classification.

pub fn from_predicate_result(words: Vec<u64>, nrows: usize) -> Self

Construct from raw predicate-result words (LSB-first within each u64).

This is the canonical entry point from try_eval_predicate_columnar and the row-wise filter fallback. The classifier:

1. Counts set bits across words.
2. Picks Empty/All/SelectionVector/VerbatimMask per density.
3. Mid-band selections become Hybrid when nrows >= 2 * HYBRID_CHUNK_SIZE.

The caller is responsible for masking off tail bits past nrows before invoking — BitMask::all_true / BitMask::from_bools and the columnar predicate path already do this.

pub fn from_words_with_count( words: Vec<u64>, nrows: usize, count: usize, ) -> Self

Test-only and conversion helper — same as from_predicate_result but caller already knows the count.

pub fn nrows(&self) -> usize

Total number of rows in the underlying DataFrame.

pub fn count(&self) -> usize

Number of rows actually selected.

pub fn contains(&self, row: usize) -> bool

True if row is in the selection. Out-of-range rows return false.

pub fn iter_indices(&self) -> SelectionIndices<'_>

Iterate selected rows in ascending order.

pub fn materialize_mask(&self) -> BitMask

Materialize the selection as a BitMask. Always allocates O(nrows/64).

pub fn materialize_indices(&self) -> Vec<u32>

Materialize the selection as an ascending Vec<u32> of row indices.

pub fn intersect(&self, other: &AdaptiveSelection) -> AdaptiveSelection

Intersection (AND). Output mode is re-classified by density.

Panics if nrows differ — programming error (different base frames).

pub fn union(&self, other: &AdaptiveSelection) -> AdaptiveSelection

Union (OR). Output mode re-classified by density.

pub fn explain_selection_mode(&self) -> &'static str

Stable identifier of the chosen mode. Useful for debug, tests, and the planned explain_selection_mode user-facing call.

Trait Implementations

impl Clone for AdaptiveSelection

fn clone(&self) -> AdaptiveSelection

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for AdaptiveSelection

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

Formats the value using the given formatter.

impl PartialEq for AdaptiveSelection

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for AdaptiveSelection

impl StructuralPartialEq for AdaptiveSelection