Enum Condition 

pub enum Condition {
    Pk(Vec<u8>),
    BitmapEq {
        column_id: u16,
        value: Vec<u8>,
    },
    BitmapIn {
        column_id: u16,
        values: Vec<Vec<u8>>,
    },
    Ann {
        column_id: u16,
        query: Vec<f32>,
        k: usize,
    },
    FmContains {
        column_id: u16,
        pattern: Vec<u8>,
    },
    FmContainsAll {
        column_id: u16,
        patterns: Vec<Vec<u8>>,
    },
    Range {
        column_id: u16,
        lo: i64,
        hi: i64,
    },
    RangeF64 {
        column_id: u16,
        lo: f64,
        lo_inclusive: bool,
        hi: f64,
        hi_inclusive: bool,
    },
    SparseMatch {
        column_id: u16,
        query: Vec<(u32, f32)>,
        k: usize,
    },
    MinHashSimilar {
        column_id: u16,
        query: Vec<u64>,
        k: usize,
    },
    IsNull {
        column_id: u16,
    },
    IsNotNull {
        column_id: u16,
    },
}

One predicate over the row-id space.

Variants

Pk(Vec<u8>)

Primary-key exact match (encoded key bytes).

BitmapEq

Low-cardinality equality via the roaring bitmap index.

Fields

column_id: u16

value: Vec<u8>

BitmapIn

Multi-value equality via the roaring bitmap index (Phase 13.5). Resolves to the union of bitmap[col].get(v) for each value — the index- accelerated equivalent of col IN (v1, v2, …) or a semi-join’s runtime value set.

Fields

column_id: u16

values: Vec<Vec<u8>>

Ann

Semantic search via the binary-quantized ANN index.

Fields

column_id: u16

query: Vec<f32>

k: usize

FmContains

Arbitrary substring via the FM index (no tokenization).

Fields

column_id: u16

pattern: Vec<u8>

FmContainsAll

Multi-segment FM intersection for LIKE '%seg1%seg2%...' (Priority 12). Resolves to the intersection of FM lookups for each segment — a much tighter superset than the single longest segment. DataFusion still re-applies the real wildcard semantics (Inexact pushdown).

Fields

column_id: u16

patterns: Vec<Vec<u8>>

Range

Inclusive integer range (served by scanning the int column, later by the learned PGM index / page-index pruning). Exclusive bounds (>,<) are expressed exactly via ±1 in the translator.

Fields

column_id: u16

lo: i64

hi: i64

RangeF64

Floating-point range with per-bound inclusivity (exact for >/</>=/ <=/BETWEEN), served the same way as Condition::Range.

Fields

column_id: u16

lo: f64

lo_inclusive: bool

hi: f64

hi_inclusive: bool

SparseMatch

SPLADE-style sparse retrieval: top-k row ids by sparse dot product over shared tokens. query is a sparse vector (token id → weight).

Fields

column_id: u16

query: Vec<(u32, f32)>

k: usize

MinHashSimilar

MinHash/LSH set-similarity: candidate row ids whose set is similar to query (a set of 64-bit token hashes), ranked by estimated Jaccard, truncated to k. Approximate (LSH recall) — the caller re-verifies.

Fields

column_id: u16

query: Vec<u64>

k: usize

IsNull

Rows where column_id is NULL. Resolved by decoding the column and collecting null positions — a column scan, but no row materialization. Page-stat aware: pages with null_count == 0 are skipped.

Fields

column_id: u16

IsNotNull

Rows where column_id is NOT NULL. The complement of Self::IsNull. Page-stat aware: pages with null_count == row_count are skipped.

Fields

column_id: u16

Trait Implementations

impl Clone for Condition

fn clone(&self) -> Condition

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Condition

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

Formats the value using the given formatter.