Struct ScanState 

pub struct ScanState {
    pub matches: BinaryHeap<Reverse<RawMatch>>,
    pub credential_interner: HashSet<Arc<str>>,
    pub metadata_interner: HashSet<Arc<str>>,
    pub static_intern: Option<Arc<StaticInterner>>,
    pub ml_score_cache: HashMap<(String, String), f64>,
    pub ml_cache_order: VecDeque<(String, String)>,
    pub ml_cache_bytes: usize,
    pub ml_pending: Vec<MlPendingMatch>,
}

Internal state for a single scan operation (tracks matches and ML cache).

Fields

matches: BinaryHeap<Reverse<RawMatch>>

Matches collected for this chunk, prioritized by confidence. Uses Reverse to make it a min-heap so we can easily pop the LOWEST confidence.

credential_interner: HashSet<Arc<str>>

Interner for credentials found in this chunk to save memory on duplicates.

metadata_interner: HashSet<Arc<str>>

Static string cache for detector metadata. Uses HashSet<Arc<str>> (not HashMap<String, Arc<str>>) so a cache miss allocates ONLY the Arc<str> - the prior shape also allocated a String to serve as the HashMap key, paying twice for what’s a single dedup slot. HashSet::get(&s) works via Arc<str>: Borrow<str>, no allocation on hits.

Hit ONLY by dynamic strings now: the scanner-wide StaticInterner (vyre CHD perfect hash) handles every (detector_id, detector_name, service, source_type) lookup without per-scan allocation.

static_intern: Option<Arc<StaticInterner>>

Optional reference to the scanner’s frozen static-string interner. When Some, intern_metadata checks here first before falling through to the per-scan metadata_interner. Lock-free on read so concurrent rayon workers share one instance without contention.

ml_score_cache: HashMap<(String, String), f64>

ml_cache_order: VecDeque<(String, String)>

ml_cache_bytes: usize

ml_pending: Vec<MlPendingMatch>

Detector matches queued for batch ML scoring at the end of the scan.

Implementations

impl ScanState

pub fn intern_credential(&mut self, s: &str) -> Arc<str>

Intern a credential string, returning an Arc<str>.

pub fn intern_metadata(&mut self, s: &str) -> Arc<str>

Intern a metadata string (detector_id, name, service, source_type, …).

Lookup order:

1. Scanner-wide StaticInterner (vyre CHD perfect hash) for detector metadata that’s frozen at scanner construction - O(1), no allocation, no lock contention.
2. Per-scan metadata_interner HashSet for dynamic strings (file paths, commit SHAs, author names, dates).

pub fn with_static_intern(intern: Arc<StaticInterner>) -> Self

Construct a ScanState that consults the scanner-wide static interner first. Use this from any path that has a &CompiledScanner in scope; falls back to default() for stand-alone unit tests.

pub fn push_match(&mut self, m: RawMatch, limit: usize)

Push a match to the state, maintaining priority and capacity. High-confidence secrets will displace lower-confidence findings.

pub fn into_matches(self) -> Vec<RawMatch>

Drain all matches into a sorted vector. Dedups identical findings (same detector + same credential + same offset) - two engines can produce the same finding for the same pattern (e.g. ac_map’s literal hit + homoglyph fallback variant both fire on plain ASCII because the homoglyph char-class includes the original char). The caller only wants one of them in the result set.

Trait Implementations

impl Default for ScanState

fn default() -> ScanState

Returns the “default value” for a type.