0.23.1

New Features 🎉

• Inline tokenizers can now be chained into relevance tuning casts (pdb.boost, pdb.const, pdb.fuzzy, pdb.slop). For example:

  sqlCopy

  SELECT * FROM mock_items
  WHERE description &&& 'shoes'::pdb.whitespace::pdb.boost(3);
  

Performance Improvements 🚀

• Tighten Block-Max WAND pruning in the single-scorer path, improving query performance when many documents share the same score.
• ||| (OR) queries over a single term now use Block-Max WAND pruning.

Stability Improvements 💪

• Replicate missed v0.23.0 migrations in v0.23.1 so upgrades from earlier versions apply cleanly.
• Return parameterized selectivity for non-Const BM25 operator arguments, improving plans for prepared statements.
• Fix a panic in terms aggregations that could occur when a missing key was combined with an include/exclude filter on certain segments.
• Correctly handle the missing sentinel in terms aggregations when combining exclude with missing on columns with zero unique values.

The full changelog is available on the GitHub Release.