YantrikDB

Bugfix release closing yantrikos/yantrikdb-server#13 (filed by @mbseid).

Fixed

extract_ops_since(since_op_id=X) and extract_ops_since(since_hlc=Y) now each independently filter the oplog. Before this release, only the two-arg (Some(hlc), Some(op_id)) path filtered; both single-watermark calls fell into the catch-all _ arm and silently returned the full oplog from the start.

The match expression in crates/yantrikdb-core/src/distributed/replication.rs:55 now handles all four cursor shapes:

| (since_hlc, since_op_id) | Behaviour |
|---|---|
| (Some, Some) | Compound cursor: hlc > ?1 OR (hlc = ?1 AND op_id > ?2) |
| (Some, None) | Strict HLC boundary: hlc > ?1 |
| (None, Some) | Subquery resolves the op's hlc, then applies the compound cursor (loss-free with op_id ties) |
| (None, None) | No cursor, full scan |

@mbseid's reproduction script is now an in-tree regression test (test_extract_ops_since_single_watermark).

Impact

• Replication / sync code paths that pass only one watermark now receive the correct delta instead of the full oplog. Prevents silent over-fetch and bandwidth waste; could have masked replica divergence.
• Documented cursor semantics in the docstring so callers know which combination to pass.

Not a breaking change

Single-watermark callers were already getting incorrect (over-broad) results. Now they get correct (filtered) results.

PR: yantrikos/yantrikdb#4