Struct SparseMaxScoreExecutor 

pub struct SparseMaxScoreExecutor<'a> { /* private fields */ }

Lazy Block-Max MaxScore executor for sparse retrieval (1-11 terms)

Combines BlockMaxScore’s cursor-based document-at-a-time traversal with BMP’s lazy block loading. Skip entries (already in memory via zero-copy mmap) drive block-level navigation; actual block data is loaded on-demand only when the cursor visits that block.

For typical 1-11 term queries with MaxScore pruning, many blocks are skipped entirely — lazy loading avoids the I/O and decode cost for those blocks. This hybrid achieves BMP’s memory efficiency with BlockMaxScore’s superior pruning for few-term queries.

Implementations

impl<'a> SparseMaxScoreExecutor<'a>

pub fn new( sparse_index: &'a SparseIndex, query_terms: Vec<(u32, f32)>, k: usize, heap_factor: f32, ) -> Self

Create a new lazy executor

query_terms should contain only dimensions present in the index. Skip entries are read from the zero-copy mmap section (no I/O).

pub async fn execute(self) -> Result<Vec<ScoredDoc>>

Execute lazy Block-Max MaxScore and return top-k results

pub fn execute_sync(self) -> Result<Vec<ScoredDoc>>

Synchronous execution — only works when the sparse index has an Inline (mmap/RAM) handle. Bypasses all async overhead for mmap-backed indexes.