Struct ExprDatabase 

pub struct ExprDatabase { /* private fields */ }

Database for storing expressions sorted by value Uses a flat sorted vector for cache-friendly range scans

Implementations

impl ExprDatabase

pub fn new() -> Self

pub fn insert_rhs(&mut self, exprs: Vec<EvaluatedExpr>)

Insert RHS expressions into the database Sorts by value for efficient range queries using partition_point

pub fn rhs_count(&self) -> usize

Get total count of RHS expressions

pub fn range(&self, low: f64, high: f64) -> &[EvaluatedExpr]

Find RHS expressions in the value range [low, high] Returns a slice of matching expressions (contiguous, cache-friendly)

pub fn find_matches( &self, lhs_exprs: &[EvaluatedExpr], config: &SearchConfig, ) -> Vec<Match>

Find matches for LHS expressions using streaming collection

This method is part of the public API for library consumers who want to perform matching without statistics collection.

pub fn find_matches_with_context( &self, lhs_exprs: &[EvaluatedExpr], context: &SearchContext<'_>, ) -> Vec<Match>

Find matches with an explicit per-run search context.

pub fn find_matches_with_stats( &self, lhs_exprs: &[EvaluatedExpr], config: &SearchConfig, ) -> (Vec<Match>, SearchStats)

Find matches with statistics collection

pub fn find_matches_with_stats_and_context( &self, lhs_exprs: &[EvaluatedExpr], context: &SearchContext<'_>, ) -> (Vec<Match>, SearchStats)

Find matches with statistics collection using an explicit per-run search context.

pub fn find_matches_turbo_with_stats_and_context( &self, lhs_exprs: &[EvaluatedExpr], context: &SearchContext<'_>, ) -> (Vec<Match>, SearchStats)

Turbo matcher: parallelize the LHS match/Newton scan across cores.

The RHS database is shared read-only. The complexity-sorted LHS set is split into contiguous bands; each Rayon worker matches its band with a private bounded pool (each sized to the full max_matches, so no band can drop a match that belongs in the global top-k). The per-band pools are then merged through one final pool to produce the global ranking.

Alternative contract

This intentionally does not honor the serial path’s byte-identical guarantee. What it does guarantee is that the single best match (rank 1) is identical to serial: the global best is rank 1 within whichever band owns its LHS (so it is never evicted), and a band’s pool sees a subset of the LHS set, so its adaptive scan radius is always at least as wide as serial’s at the same LHS — turbo therefore cannot miss a match serial found. After the merge, the global best sorts to rank 1.

Lower-ranked results may differ from serial: when more matches qualify than max_matches, each band keeps only its own top-k, so the merged tail is a valid top-k but not necessarily the same one serial would pick among equally-ranked candidates, and it may vary with thread count.

Bands are processed in ascending complexity, so the per-band complexity-ceiling early exit in [match_one_lhs] stays valid. Explicit first-hit stopping conditions are handled by the serial matcher before this matcher is called because they are not safe to apply independently in parallel bands.

Trait Implementations

impl Default for ExprDatabase

fn default() -> Self

Returns the “default value” for a type.