Skip to content

Latest commit

 

History

History
48 lines (38 loc) · 2.04 KB

File metadata and controls

48 lines (38 loc) · 2.04 KB

Scanner Core Parity

This document tracks the parity gate for scanner core logic consolidated into crates/scanner-engine.

Runtime Integration Status

The runtime-facing integration surface currently keeps configuration, summary, and parity plumbing local to gossip-scanner-runtime:

  • crates/gossip-scanner-runtime provides typed scan_fs and scan_git entrypoints plus sink-aware scan_fs_with_runtime and scan_git_with_runtime helpers.
  • crates/scanner-rs-cli exposes the workspace scanner-rs binary with scan fs|git shape and --execution-mode defaulting to direct.
  • crates/gossip-worker exercises the same runtime surface through a config-resolved worker entrypoint that can dispatch either to local scans or to the distributed worker path.
  • crates/gossip-scanner-runtime owns local ScanReport, ScanCheckpoint, CancellationToken, commit-pipeline, commit-sink, commit-model, result-translation, result-committer, checkpoint-aggregator, coordination-sink, and distributed worker-loop types.
  • crates/scanner-engine owns the detection pipeline: vectorscan prefilter, regex, transform decode, offline validation, and finding emission.

At the moment the runtime validates scan inputs, routes filesystem and git requests through live family runtimes, and exposes a receipt-driven distributed worker loop for filesystem leases. End-to-end parity work remains focused on the surfaces with pinned fixtures and canonicalized outputs:

  • CLI parsing and summary rendering
  • event-sink formatting
  • JSONL canonicalization in parity.rs
  • local-vs-distributed filesystem finding-set parity after JSONL path normalization
  • deterministic identity derivation via translate_item_result

As additional receipt-driven distributed paths land, the same public runtime API can widen detection-path parity work without another caller-facing surface change.

Throughput Policy

Throughput gates apply to runtime paths that execute the scanner engine:

  • median absolute throughput delta <= 2.0%
  • per-case absolute throughput delta <= 5.0%