tinysearch Files

This release of tinysearch adds a completely new indexing algorithm: a compact exact inverted index designed to make search feel immediate.

Compared with the probabilistic Xor8 backend, the new default stores exact postings and supports prefix matching across titles, body text, and metadata. Once a query term reaches three characters, tinysearch can return useful results while the user is still typing. Exact postings also remove the compounded false positives inherent in probabilistic filters.

That richer behavior has a size cost, but the compressed difference stays modest. On my endler.dev corpus:

Compared with the current Xor8 build, the exact index adds 7,848 bytes after gzip or 4,856 bytes after Brotli. In return, searches get immediate partial-word matches across every indexed field and exact matching semantics.

Xor8 remains fully supported and is still a good choice when the smallest possible artifact matters more than body and metadata prefix search. Select it with --indexer xor8 in the CLI or IndexKind::Xor8 in the library. Existing Xor8 storage files remain readable.

The indexing API is more explicit as well. SearchIndex is now opaque, backend construction uses the IndexBackend trait with ExactIndexBackend and Xor8IndexBackend, and building and serialization are separate steps:

:::rust
let index = search.build_index(&posts)?;
let bytes = search.serialize_index(&index)?;

build_and_serialize_index remains available as a deprecated compatibility helper.

The generated WASM demo now updates search results and suggestions while typing. This release also tightens the lint policy, hardens compact-index decoding, updates dependencies and CI maintenance, and teaches the WASM optimization path about modern bulk-memory modules.

Full changelog: https://github.com/tinysearch/tinysearch/compare/v0.10.0...v0.11.0