v1.1.5 - Bicycle

WE PUT THE DEMON IN PRIDE MONTH

[v1.1.5] - Bicycle - 2026-06-03

Post-v1.1.4 stabilization focused on Cypher/Bolt correctness, storage resilience, and deterministic behavior under real Neo4j-driver query shapes. This range also includes broad coverage expansion; all test/coverage work is bundled under a single item below.

Changed

• Storage embedding persistence now shards oversized chunk payloads in Badger. Large embedding chunks are written/read using size-aware shard keys with backward-compatible decode logic for legacy single-value chunks. This prevents value-size failures on valid large embeddings while keeping existing data readable.
• Search query chunk scoring now uses best-match chunk score instead of average score. This improves relevance for multi-chunk documents where one strong matching chunk should dominate ranking.
• Version metadata updated to reflect the current post-v1.1.4 development state.

Fixed

• Cypher MATCH+CREATE error handling now surfaces storage failures instead of silent no-op results. Label/all-node lookups and relationship-existence checks in compound MATCH...CREATE paths now return explicit errors.
• Cypher MATCH...CREATE relationship creation regressions fixed. The following common Neo4j-compatible patterns now execute correctly again:
  • MATCH (a),(b) CREATE (a)-[:R]->(b)
  • MATCH (s) CREATE (n)-[:R]->(s)
• Cypher CREATE...SET clause-boundary parsing fixed. Trailing CREATE/WITH/RETURN after SET are now parsed as clauses (not expression text), and explicit transaction routing now matches autocommit behavior for CREATE...SET execution.
• Cypher SET x = x + "..." self-reference mutations fixed. In-place property concatenation/update now writes the computed value instead of silently no-oping.
• Cypher MATCH...CREATE projection correctness fixed. COUNT(expr) in post-create return projections now evaluates against both matched and newly created bindings, preventing false zero counts.
• Cypher MERGE return aggregation correctness fixed. COUNT(*) / COUNT(var) in MERGE return projection paths now produce deterministic Neo4j-compatible values instead of intermittent nil.
• Cypher shell parameter expression failures are now surfaced explicitly. Expression-evaluation errors are no longer swallowed in shell parameter processing.
• Cypher merge-chain correctness tightened. Post-WITH MERGE-chain clause splitting and execution were corrected so node MERGE segments are not dropped; invalid empty merge-chain shapes now return typed ErrInvalidMergeChainQuery errors.
• Cypher relationship-WHERE evaluation ordering fixed. Relationship-pattern WHERE clauses are now evaluated before single-variable shortcut routing in multi-match paths.
• Cypher chained MERGE relationship SET parsing fixed. Relationship SET parsing now stops at the next top-level MERGE boundary in chained clauses.
• Cypher MATCH...UNWIND...MERGE loop variable substitution fixed. UNWIND variables are now expanded per-item (instead of treated as literal identifiers) when UNWIND appears between MATCH and MERGE.
• Cypher/Bolt multi-hop OPTIONAL MATCH regression fixed. MATCH ... OPTIONAL MATCH ... after traversal now seeds optional matches from the full initial traversal result, preventing false zero-row outcomes.
• Bolt write counters now populate Neo4j-compatible summary stats. PULL/DISCARD completion metadata now includes write stats so ResultSummary counters are correct.
• Cypher RETURN-after-DETACH-DELETE projection fixed. Non-COUNT return expressions now resolve correctly instead of returning null placeholders.
• Storage namespaced label scans hardened. Nil-node handling in namespaced GetNodesByLabel no longer panics in edge cases.
• Storage MVCC namespace state recovery fixed after migration/reopen. Missing namespace counters are recovered from existing heads and transaction startup now primes namespace MVCC state, preventing false snapshot conflicts on migrated/reopened stores.
• Badger recovery behavior improved for stale/cold namespaces while keeping hot-path cost low.
• Search mismatched-dimension panic fixed. Query paths now handle dimension mismatch safely instead of panicking.
• CI architecture-related flaky float precision assertion fixed.

Tests

• Bundled test/coverage work: expanded deterministic regression and branch coverage across cypher, storage, search, server, nornicdb, heimdall, and config, including many new correctness-focused edge-case tests.

Technical Details

• Range covered: v1.1.4..main
• Commits in range: 33 (non-merge)
• Repository delta: 220 files changed, +23,707 / -1,496 lines

Full Changelog: v1.1.4...v1.1.5