v0.7.0
Performance
- 37× faster joins — LATERAL join for line-range containment
- 8× faster structural search —
(kind, name, arity)btree index bypasses GIN - 2× faster text search — file-first search with LATERAL fragment lookup
Benchmarks on 13.8M fragments (full Hex.pm index):
| Query | Before | After |
|---|---|---|
def handle_call(_, _, _) |
602ms | 78ms |
Text: TODO |
361ms | 184ms |
| DSL join: Fragment + Enum.map/2 refs | 26.4s | 705ms |
Fixed
- ParadeDB BM25 enabled by default
- BM25 index creation fixed for ParadeDB 0.21+ (bigint cast)
- Deadlock on concurrent term inserts — advisory lock per transaction
Docs
Complete documentation rewrite: all 12 guides updated with hex indexing pipeline, performance tuning, architecture details, and actual scale numbers.