Struct GroupSegmentStats 

pub struct GroupSegmentStats {
    pub group: u32,
    pub row_count: u64,
    pub live_count: u64,
    pub null_count: u64,
    pub sort_min: Option<Value>,
    pub sort_max: Option<Value>,
    pub columns: HashMap<Arc<str>, ColumnStats>,
}

Per-group segment statistics for scan pruning.

One record per owner group: row and live counts, the property null-cell total, endpoint sort bounds, and per-column statistics. Column statistics reuse the attribute-column caliber exactly: they are built with compute_stats_streaming and serialized with ColumnStats metadata, so no second statistics structure exists. Bounds only ever widen across refreshes so pruning stays conservative for every snapshot; counts are exact-current for observability. Topology itself never holds nulls, so a zero null total with bitmap-backed property nulls is the normal case.

Fields

group: u32

row_count: u64

live_count: u64

null_count: u64

sort_min: Option<Value>

sort_max: Option<Value>

columns: HashMap<Arc<str>, ColumnStats>

Implementations

impl GroupSegmentStats

pub fn collect( group: u32, row_count: u64, live_count: u64, endpoints: &[u32], column_values: &HashMap<Arc<str>, Vec<Option<Value>>>, column_encodings: &HashMap<Arc<str>, EncodingType>, ) -> GroupSegmentStats

Collect fresh statistics for one group from its endpoint values and per-column value slices. The endpoint column uses the same streaming aggregation as attribute columns.

pub fn widen_with(&mut self, fresh: &GroupSegmentStats)

Merge fresh statistics into these bounds, widening only.

Minimums move down, maximums move up, distinct estimates merge; counts take the fresh values. A group that once held a value keeps covering it, so segment pruning stays conservative for every snapshot while counts stay exact-current for observability.

pub fn may_contain(&self, predicates: &[ScanPredicate]) -> bool

Whether this segment may contain rows matching the pushed predicates.

Conservative pre-filter: returns false only when a merged predicate range provably excludes the widened column bounds. Missing statistics or missing bounds never prune. Null-only reasoning never prunes either: nulls use bitmap semantics and old snapshots may still hold values where the flushed epoch shows nulls.

pub fn encode(&self) -> Vec<u8>

pub fn decode( data: &[u8], cursor: &mut usize, ) -> Result<GroupSegmentStats, StorageError>

Trait Implementations

impl Clone for GroupSegmentStats

fn clone(&self) -> GroupSegmentStats

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for GroupSegmentStats

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

Formats the value using the given formatter.