Release v0.6.7-dev

[0.6.7-dev] - 2026-05-06

🚀 Features

• (planner) Phase 1 — write LogicalPlan variants + builder + guard (#277)
• (render) Phase 2 — WriteRenderPlan + write_to_sql + id_gen (#278)
• (embedded) Phase 3 — executor wiring + ID generation + block-tracking DDL (#279)
• (tck) Phase 4 — import write feature files + write-clause docs (#280)
• (tck) Phase 5a — side-effect step + label-mutation skip tag (#281)
• (planner) Phase 5b — anonymous-node CREATE routes to __Unlabeled (#282)
• (tck) Phase 5c — lift Delete1 [3] + per-scenario triage notes (#283)
• (embedded) Phase 5d — write+RETURN side-channel + accurate counters (#284)
• (planner) Phase 5e — untyped-MATCH+write fan-out across node tables (#285)
• (ffi) Expose Phase 5d write_counters side-channel through FFI + Python (#286)
• Change default HTTP port from 8080 → 7475 (#310)

🐛 Bug Fixes

• (tck) 402/402 passing — List indexing, type validation, step regex, DELETE detection (#273)
• (deps) Bump rustls-webpki to 0.103.13 (RUSTSEC-2026-0098/0099/0104) (#274)
• ShortestPath/allShortestPaths + COUNT regression + test-stack glue (#311)
• (bolt) Neo4j Browser 5.x compatibility — handshake + bundled-query intercepts + SHOW stubs (#312)
• (ci) Scope release build to shipped crates only (no chdb on Windows) (#314)
• (release) Cfg-gate jemalloc allocator on non-MSVC targets (#315)

💼 Other

• Prep v0.6.7-dev — version bump + CHANGELOG/STATUS refresh + audit dispositioning (#313)

🚜 Refactor

• (analyzer) Drop dead pattern_resolver module (#287)
• Drop dead helper fns flagged by clippy (#288)
• Assorted clippy idiomatic-Rust fixes (#289)
• Address remaining mechanical clippy warnings (#290)
• Mechanical clippy warning cleanup (92→68) (#291)
• Clean up unnecessary_unwrap and map_entry warnings (#292)
• Triage 6 dead-code warnings (#293)
• Bundled clippy fixes (&mut, FromStr, type aliases) (#294)
• Unwrap redundant test mod blocks (module_inception) (#295)
• Silence only_used_in_recursion on Pass-API helpers (#296)
• Factor very-complex types into named aliases (#297)
• Box CypherStatement::Query payload (large_enum_variant) (#298)
• Retire 4 too_many_arguments warnings (dead + test-only) (#299)
• Retire 4 too_many_arguments warnings in variable_length_cte (#300)
• Retire 5 too_many_arguments warnings in render_plan + view_scan (#301)
• Clear remaining too_many_arguments + doc-list warnings (#302)
• Clear 9 clippy warnings in clickgraph-embedded (#305)
• Clear per-crate clippy warnings (tool + embedded) (#306)
• Delete orphan test dirs + 5 unused deprecated items (#309)

📚 Documentation

• Design plan for embedded-mode Cypher writes (#275)
• (embedded-writes) Lock Phase 0 decisions; switch SET to lightweight UPDATE (#276)
• Refresh STATUS + CHANGELOG for v0.6.7-dev cycle (#303)
• Fix all 113 rustdoc broken-link warnings (#304)

⚙️ Miscellaneous Tasks

• Adding motivation.md (#272)
• Lock in zero-warning state via -D warnings + fmt --all (#307)
• Align clickgraph-client MSRV with workspace (1.70 → 1.85) (#308)

v0.6.6-dev

What's New in v0.6.6-dev

• cg CLI tool — Agent/script-oriented CLI (clickgraph-tool crate). Translate and execute Cypher without a running server: cg sql, cg validate, cg query, cg nl (NL→Cypher via LLM), cg schema show/validate/discover/diff. Config via ~/.config/cg/config.toml. Designed for agentic callers, CI pipelines, and scripting.
• embedded feature now opt-in — clickgraph-embedded compiles without chdb by default. New Database::new_remote(schema, RemoteConfig) constructor executes Cypher against external ClickHouse with no chdb dependency.
• Agent skills — Three publishable skills for any agentic framework (Claude Code, LangChain, AutoGen, CrewAI, OpenAI function calling): /cypher, /graph-schema, /schema-discover. See skills/README.md.
• openCypher TCK: 383/402 (95.3%), 0 failures — up from 379/402. The 19 skipped scenarios cover Cypher write clauses (CREATE, SET, DELETE, MERGE).
• Type inference ANY fix — WHERE n.p1 = 12 OR n.p2 = 13 across multi-label nodes now correctly retains all candidate types. IS NULL / IS NOT NULL accesses handled without incorrectly eliminating type candidates.
• chdb resource caps — SystemConfig gains max_memory_usage_bytes; both max_threads and max_memory_usage applied at session init. chdb-rust bumped to 1.3.1.

See CHANGELOG.md for full history.

Release v0.6.5-dev

[0.6.5-dev] - 2026-03-21

🚀 Features

• Embedded write support for agentic GraphRAG workflows (#236)
• Property_types field for type-aware writable table DDL (#238)
• Hybrid remote query + local storage for embedded mode (#240)
• Kuzu API parity quick wins (#242)
• COPY FROM import for CSV, Parquet, and auto-detected formats (#243)
• Date, Timestamp, UUID Value types (#244)
• Arrow, Pandas, Polars DataFrame output for Python (#245)

🐛 Bug Fixes

• Release pipeline and Docker publish for protected branches (#235)

💼 Other

• V0.6.5-dev (#247)

📚 Documentation

• Add embedded write API to README What's New (#237)
• Embedded mode tutorials and runnable examples (#246)

🧪 Testing

• Hybrid remote query E2E tests + fix edge extraction (#241)

Release v0.6.4-dev

[0.6.4-dev] - 2026-03-14

🚀 Features

• Denormalized & coupled schema support: Full query support for schemas where node properties are embedded in edge tables via from_node_properties/to_node_properties. Includes property mapping, ORDER BY resolution, UNION aggregate column rewriting, and id() on virtual nodes (PRs #224-#228).

• OPTIONAL MATCH on denormalized schemas: New CTE + LEFT JOIN architecture for correct LEFT JOIN semantics when MATCH produces a UNION standalone node scan. Includes UnionDistribution skip for optional patterns, column reference rewriting, and join preservation through the optimizer (PRs #229-#230).

• VLP on denormalized/polymorphic schemas: Fixed exact-length VLP cycle prevention for virtual nodes (no separate table), enabling *2, *3 patterns. Range VLP (*1..3), path variables, and shortestPath all work on denormalized schemas (PR #231).

• Cross-schema pattern matrix tests: Comprehensive test suite covering 15 query patterns across 5 schema types (standard, FK-edge, denormalized, polymorphic, coupled). 151 tests passing, 0 xfails (PRs #226-#232).

🐛 Bug Fixes

• Denormalized property mapping: get_properties_with_table_alias() resolves node properties through edge table's from_node_properties/to_node_properties with direction awareness (PR #225).
• id(node) on denormalized nodes: SelectBuilder Case 5 now resolves through edge alias and mapped column instead of using the virtual node alias directly (PR #227).
• UNION branch Column qualification: Bare Column("OriginCityName") expressions from denormalized ViewScans converted to PropertyAccessExp with correct alias in GraphNode handler (PR #228).
• VLP cycle prevention: Moved extract_table_name calls inside non-denormalized branch — denormalized patterns use from_id/to_id directly (PR #231).
• UnionDistribution: Skip distributing optional GraphRel over denormalized Union to preserve LEFT JOIN semantics (PR #229).
• is_node_denormalized: Now handles Union of denormalized GraphNodes (PR #229).

🧹 Infrastructure

• jemalloc memory allocator: Reduces memory fragmentation for long-running server workloads (PR #213).
• Plan explosion guard: Prevents combinatorial blowup in multi-type VLP expansion (PR #212).
• Test cleanup: ~103 stale xfail markers removed, 25 invalid test queries converted to skips (PRs #211, #218-#223, #227, #232).

---

[0.6.3-dev] - 2026-03-05

🚀 Features

• APOC Export Procedures: Neo4j-compatible CALL apoc.export.{csv|json|parquet}.query(cypher, destination, config) for exporting query results. Supports local files, S3, GCS, Azure, and HTTP destinations. Works in HTTP server, Bolt protocol, and embedded mode.

  • Destination resolver: Maps URI schemes to ClickHouse INSERT INTO FUNCTION table functions (file(), s3(), url(), azureBlobStorage())
  • Parser fix: Standalone CALL with positional args now correctly parsed even when inner Cypher contains RETURN/UNION keywords
  • Config: Parquet compression codecs (snappy, gzip, lz4, zstd, brotli)

• Embedded mode (PR #179): Run Cypher graph queries entirely in-process via chdb — no external ClickHouse server required. Supports Parquet, CSV, Iceberg, Delta Lake, and S3-compatible storage.

  • QueryExecutor trait: Abstracts SQL execution; RemoteClickHouseExecutor (existing) and ChdbExecutor (new) are the two backends. Default behaviour is unchanged.
  • clickgraph-embedded crate: Kuzu-compatible Rust library API — Database::new(schema, config), Connection::new(&db), conn.query(cypher), result.next() → Row.
  • source: schema field: Optional per-node/relationship URI pointing to the data file. At startup, ClickGraph creates chdb VIEWs named after the schema table: field so existing SQL generation requires no changes.
  • URI schemes: file://, s3://, gs://, iceberg+s3://, iceberg+local://, delta+s3://, table_function:<raw>.
  • StorageCredentials: S3/GCS/Azure credentials applied as chdb SET commands at session init; falls back to environment variables and instance-profile credentials automatically.
  • Server embedded flag: --embedded CLI flag / CLICKGRAPH_EMBEDDED=true env var; HTTP and Bolt endpoints work as normal.
  • Tests: 9 source_resolver tests, 8 credential tests, 17 embedded unit tests, 10 e2e integration tests.
  • Docs: Embedded Mode wiki page

🚀 Features

• LDBC SNB benchmark: 14/37 → 36/37 (97%) — 22 queries promoted from adapted to official Cypher. The only remaining gap is bi-16 (CALL subquery, a known language feature gap).

  • Official queries promoted: complex-3, complex-5, complex-7, complex-10, complex-12, complex-13, bi-3, bi-8, bi-14, and others
  • Adapted queries remaining: bi-17 (multi-VLP), complex-14 (weighted shortest path via cost(path))

• GraphRAG structured output (format: "Graph") (PR #165): Query results returned as graph-structured JSON with nodes, edges, and properties — enables direct consumption by graph visualization and RAG pipelines.

• ClickHouse cluster load balancing (CLICKHOUSE_CLUSTER env var) (PR #164): Distributes queries across ClickHouse cluster nodes for horizontal read scaling.

• apoc.meta.schema() for MCP server compatibility (PR #163): Implements the Neo4j APOC procedure that MCP servers and graph tools use for schema introspection.

• LLM-powered schema discovery (:discover command) (PR #146): Server formats a discovery prompt (POST /schemas/discover-prompt), client calls LLM (Anthropic or OpenAI-compatible) to generate YAML schema from ClickHouse table metadata. Replaced the GLiNER/gline-rs approach.

• Weighted shortest path (cost(path) function) (PR #160): Supports Dijkstra-style weighted VLP traversal for queries like complex-14. WeightCteConfig carries weight info through the VLP pipeline; auto-creates bidirectional weight CTEs for undirected traversal.

• List comprehension → arrayCount() optimization (PR #153): Parses [x IN list WHERE cond | expr] syntax, maps size(ListComprehension) to ClickHouse arrayCount() — avoids correlated subqueries that fail with UNION ALL ("Cannot clone Union plan step").

• Pattern comprehension → pre-aggregated CTE approach (PR #159): Replaces correlated subqueries from size(PatternComprehension) with pre-aggregated CTEs + LEFT JOINs. Includes arrayConcat() for list concatenation (list1 + list2).

• Official complex-7 — chained map access + NOT EXISTS (PR #152): Greedy chained property parsing (a.b.c), map literal node flattening (head(collect({key: node}))), split NOT EXISTS for undirected edges.

• Official complex-3 — supertype inference + IN→OR expansion (PR #151): Supertype collapse (Post+Comment → Message), IN [col1, col2] → OR expansion for ClickHouse compatibility, 5-WITH chain support.

• Map property access (collect({score: x})[0].score → ClickHouse map subscript) (PR #147): Tracks map_keys through CTE pipeline, generates ArraySubscript for map property access with 0-based → 1-based index conversion.

• UNWIND support (ARRAY JOIN) (PR #133): Translates Cypher UNWIND to ClickHouse ARRAY JOIN.

• --log-level CLI flag for runtime log level configuration.

🐛 Bug Fixes

• Undirected edge fixes: Removed has_nested_undirected_edge guard that prevented UNION split for mid-chain undirected edges (PR #147). Fixed BidirectionalUnion for multi-pattern MATCH with bound endpoints — collapses redundant Union to single Outgoing branch (PR #148).

• VLP (variable-length path) fixes: Fixed path rewriting for reverse UNION branches (PR #135), composite ID support (PR #134, #136), *N..N exact-hop guard (PR #137), duplicate WITH RECURSIVE removal (PR #131), multi-VLP query support (PR #132), DISTINCT deduplication (PR #130), zero-lower-bound *0.. for single-type and multi-type VLPs (PR #142), CROSS JOIN removal for VLP CTEs in downstream queries (PR #145).

• OPTIONAL MATCH fixes: INNER→LEFT JOIN conversion for CTE-backed JOINs in OPTIONAL MATCH context, spurious duplicate JOIN removal, orphan JOIN removal guards, collect(node) expansion to ID-only for has() compatibility (PR #143).

• CTE/scope fixes: Bare variable resolution after WITH barrier (PR #120, #121), cte_references preservation in UNION branches (PR #122), composite alias augmentation (PR #128), buried WithClause preservation in DuplicateScansRemoving (PR #138).

• shortestPath fixes: CASE path IS NULL → ifNull(minOrNull(hop_count), -1) rewriting, spurious non-VLP JOIN cleanup, endpoint inline filter preservation (PR #157).

• Parser whitespace fix: MATCH/OPTIONAL MATCH now handle leading whitespace after $param syntax (PR #145).

• Browser click-to-expand regressions: Fixed 5 bugs from scope resolution redesign — filter_tagging crash, VLP multi-type inference, type mismatch, polymorphic label extraction, pruned MATCH detection (PR #156).

• Determinism fixes: HashSet→BTreeSet in anchor node selection, HashMap→BTreeMap in GraphSchema, sorted conversions in CTE extraction (PR #137, #139).

⚙️ Infrastructure

• Integration test cleanup: 3,068 tests passing, 57 stale xfails removed (PR #169).
• Scoping-only WITH collapse + benchmark infrastructure (PR #168): Optimizes scoping-only WITH clauses that don't need CTE materialization.
• Schema-parameterized SQL generation tests: 76 tests across 6 schema variants (PR #162).
• Browser interaction tests with full schema variant coverage (PR #161).
• Version bump to v0.6.3-dev with README cleanup (PR #167).
• Roadmap and guide updates (PR #166).

[0.6.2-dev] - 2026-02-20

⚙️ Architecture

• Scope-aware variable resolution for CTE/UNION rendering (Feb 20, 2026, PR #120): Infrastructure for correct variable resolution acro...

Release v0.6.3-dev

What's New

📊 LDBC Social Network Benchmark: 97% Coverage

36 of 37 LDBC SNB queries now pass — up from 14/37 in v0.6.2-dev. 22 queries promoted from adapted workarounds to official Cypher syntax. The sole remaining gap is bi-16 (CALL subquery — a known language feature limitation).

🚀 New Features

• GraphRAG structured output (format: "Graph") — query results as graph-structured JSON with nodes, edges, and properties for direct consumption by visualization tools and RAG pipelines
• ClickHouse cluster load balancing (CLICKHOUSE_CLUSTER env var) — distributes queries across cluster nodes for horizontal read scaling
• apoc.meta.schema() — Neo4j APOC compatibility for MCP servers and graph introspection tools
• LLM-powered schema discovery (:discover command in clickgraph-client) — generates YAML graph schema from ClickHouse table metadata via Anthropic or OpenAI-compatible LLMs
• Weighted shortest path (cost(path) function) — Dijkstra-style weighted traversal with bidirectional weight CTEs
• List comprehension ([x IN list WHERE cond | expr]) → optimized to arrayCount(), avoiding correlated subqueries
• Pattern comprehension → pre-aggregated CTE + LEFT JOIN approach, plus arrayConcat() for list concatenation
• Chained map access (a.b.c), map literal node flattening, map property subscript access
• Supertype inference (Post+Comment → Message) with IN [col] → OR expansion for ClickHouse
• UNWIND support via ClickHouse ARRAY JOIN
• --log-level CLI flag for runtime log configuration

🐛 Bug Fixes (highlights)

• Undirected edge UNION generation for mid-chain and multi-pattern MATCH
• 10+ VLP fixes: reverse branch paths, composite IDs, *0.. zero-hop, CROSS JOIN removal, deduplication
• OPTIONAL MATCH: INNER→LEFT conversion, orphan JOIN removal, collect()/IN compatibility
• CTE scope: bare variable resolution, buried WithClause preservation, composite alias augmentation
• shortestPath: CASE path IS NULL rewriting, endpoint filter preservation
• Parser: leading whitespace after $param syntax
• Browser click-to-expand: 5 regression fixes
• Deterministic output: HashSet→BTreeSet/BTreeMap across anchor selection and schema iteration

⚙️ Infrastructure

• Integration tests: 3,068 passing, 57 stale xfails removed
• Scoping-only WITH collapse optimization
• Schema-parameterized SQL generation tests (76 tests × 6 schemas)
• Browser interaction test coverage

See CHANGELOG.md for full details.

v0.6.2-dev

What's New

🐛 Bug Fixes — Neo4j Browser click-to-expand (PR #116)

Five regression bugs from the scope resolution redesign (PR #115) that completely broke Neo4j Browser click-to-expand:

• filter_tagging crash on Empty child plan after TypeInference pruning
• VLP multi-type inference — empty UNION branches generated due to stale GraphNode state
• VLP+WITH type mismatch — UInt64 vs String in JOIN condition
• extract_node_labels not polymorphic — returned only primary label for multi-type nodes
• is_return_only_query — misidentified pruned MATCH as pure RETURN → empty SQL
• HTTP OPTIONS/GET probes on Bolt port downgraded from ERROR to DEBUG noise

⚙️ Infrastructure

• Neo4j Browser demo: all 5 tables → MergeTree (data persists across restarts)
• Demo data loading consolidated to init-db.sql single entrypoint
• Demo clickgraph service → genezhang/clickgraph:v0.6.2-dev

🚀 Features (from earlier in this cycle)

• Foundational variable scope resolution redesign (PR #115)
• Schema/type inference consolidation
• ORDER BY / HAVING / LIMIT / SKIP clause extraction fixes

See CHANGELOG.md for full details.

Release v0.6.1

[0.6.1] - 2026-01-12

🚀 Features

• Integrate data_security schema, remove benchmark schemas from unified tests
• Auto-load all test schemas at session start
• Add PatternGraphMetadata POC for cleaner join inference evolution
• Phase 1 - Use cached node references from PatternGraphMetadata
• (graph_join_inference) Phase 2 - Simplified cross-branch detection using metadata
• (graph_join_inference) Phase 4 - Add relationship uniqueness constraints
• Complete fixed-length path inline JOIN optimization
• Property pruning optimization with unified test infrastructure
• Edge constraints for cross-node validation (8/8 tests passing)
• Pattern Comprehensions and Multiple UNWIND support
• Add multi-schema YAML support for loading multiple graph schemas
• Add multi-schema database setup and test scripts
• Add array subscript syntax support and complete multi-type VLP path functions
• Make MAX_INFERRED_TYPES configurable via query parameter

🐛 Bug Fixes

• Support anonymous nodes in graph patterns
• Use node ID columns for VLP CTE generation
• Optimize JOIN generation based on property usage, not node naming
• Optimize JOIN generation based on property usage, not node naming
• Permanently fix test infrastructure issues
• Add filesystem and group membership test data to setup script
• Add small-scale benchmark test data and cleanup obsolete scripts
• Migrate from schema_name='default' to USE clause convention
• Add missing matrix test schemas and USE clause support
• Add USE clause to multi-hop pattern tests
• Update social_polymorphic schema to use actual table names
• Resolve ontime schema name conflict, add benchmark schemas back for matrix tests
• Add flights to default db for ontime_benchmark - Copy flights to default database - Comprehensive matrix: +256 tests - Overall: +186 tests to 2947 - Session total: +1047 tests (+55 percent)
• Restore ontime_flights schema name for pattern matrix tests - Revert ontime_denormalized back to ontime_flights - Remove ontime_benchmark from unified test loading - Update matrix conftest to use ontime_flights - Pattern schema matrix: 0/51 to 9/51 recovery - Overall: 2758 to 2958 (+200 tests) - Session: 1900 to 2958 (+1058 tests, +55.7 percent, 85.2 percent pass rate)
• Add property_expressions schema to test loading - Fix database to default where tables actually exist - Replace CASE WHEN with if() for parsing compatibility - Add to load_test_schemas.py - Property expressions tests: 0/28 to 13/28 recovery - Overall: 2958 to 2976 (+18 tests) - Session: 1900 to 2976 (+1076 tests, +56.6 percent, 85.7 percent pass rate)
• Add schema_name to role-based query tests - Role tests now use unified_test_schema - All 5 role-based tests now pass
• Add missing property aliases to property_expressions schema
• VLP cross-branch JOIN uses node alias instead of relationship alias
• VLP transitivity check handles polymorphic relationships
• All integration tests now passing or properly marked xfail
• Add relationship labels to edge list test GraphRel structures
• Update edge list test assertions for SingleTableScan optimization
• Add proper GraphSchema to failing tests
• Thread schema through single-hop query pipeline for edge constraints
• (vlp) Fix denormalized VLP node ID selection (Dec 22 regression)
• (vlp) Complete denormalized VLP with comprehensive fixes
• VLP path functions in WITH clauses + CTE body rewriting
• Remove escaped quotes and multi_schema loader entry from conftest
• Load denormalized_flights_test schema with proper data
• VLP WHERE clause alias resolution for denormalized schemas
• Correct AUTHORED relationship schema in unified_test_multi_schema.yaml
• Multi-type VLP architectural fix - FROM alias solves all mapping issues
• Multi-type VLP JSON extraction - skip alias mapping for multi-type CTEs
• FK-edge zero-length VLP edge tuple generation
• Unify MAX_INFERRED_TYPES default to 5 for consistency
• Parameterized views apply to both node and edge tables in VLP queries
• Add anyLast() wrapping for CTE references in GROUP BY aggregations
• Rewrite CTE column references in JOINs
• VLP+WITH+MATCH pattern (ic9) - delegate to input.extract_joins() for CTE references
• Add VLP endpoint detection in find_id_column_for_alias
• Correct ontime_denormalized schema to use default database
• Skip JOINs for fully denormalized VLP patterns
• Map denormalized VLP endpoint aliases to CTE alias for rewriting
• Consecutive MATCH with per-MATCH WHERE, comment support, scalar aggregate investigation
• WITH expression scope - rewrite CASE expressions to use CTE columns

💼 Other

• Comprehensive test failure categorization (507 failures)
• V0.6.1 - WITH clause fixes, GraphRAG enhancements, LDBC progress
• Update Cargo.lock for v0.6.1 release

🚜 Refactor

• (graph_join_inference) Phase 3 - Break up infer_graph_join() god method
• [breaking] Migrate all integration tests to multi-schema format
• [breaking] Remove obsolete unified_test_schema and cleanup
• Consolidate denormalized_flights schema references

📚 Documentation

• Update README.md with v0.6.0 and accumulated features
• Update KNOWN_ISSUES.md with v0.6.0 fixes
• Archive wiki for v0.6.0 release
• Add release notes for v0.6.0
• Fix ClickHouse function prefix (ch./chagg. not clickhouse.)
• Fix composite node ID example (use nodes not edges)
• Update STATUS and investigation plan with anonymous node fix
• Update STATUS with property usage optimization and current test status
• Complete test infrastructure documentation
• Update STATUS with schema loading fix
• Update STATUS - ALL INTEGRATION TESTS PASSING! 🎉
• Add comprehensive architecture analysis for Scan/ViewScan/GraphNode relationships
• Update gap analysis - Gap #2 already implemented
• Add schema testing requirements (VLP multi-schema mandate)
• Add VLP denormalized property handling TODO
• Add session findings and feature analysis
• Clean up KNOWN_ISSUES.md and add path function limitation
• Update CHANGELOG and test infrastructure for VLP fixes
• Add multi-schema configuration documentation
• Add multi-schema setup guide
• Update TESTING.md for multi-schema architecture
• Update STATUS.md - remove load_test_schemas.py reference
• Add VS Code terminal freeze prevention to TESTING.md
• Document VLP WHERE clause bug discovery
• Update Cypher-Subgraph-Extraction.md with verified pattern support matrix
• Document max_inferred_types feature and update default to 5
• Update STATUS with LDBC progress and IC-9 CTE naming issue
• Systematic documentation cleanup and reorganization
• Streamline STATUS.md to focus on current state (2822 → 322 lines)
• LDBC benchmark baseline testing and analysis
• Update README test coverage to 3000+ tests and reorganize features
• Archive wiki documentation for v0.6.1 release

🧪 Testing

• Update test expectations for known limitations
• Add error message verification for known limitations
• (graph_join_inference) Add comprehensive unit tests for Phase 4 uniqueness constraints
• Add comprehensive VLP cross-functional testing
• Add comprehensive GraphRAG schema variation tests
• Add zero-length VLP tests for [*0..] and [*0..N] patterns

⚙️ Miscellaneous Tasks

• Update CHANGELOG.md [skip ci]
• Add lineage test schema and cleanup temporary files
• Move SCHEMA_THREADING_ARCHITECTURE.md to docs/development/
• Ignore docs1 directory in gitignore
• Clean up docs
• More doc cleanup
• More docs clean up, README
• Remove unused Flight node from unified_test_schema.yaml
• Update CHANGELOG.md [skip ci]

[0.6.0] - 2025-12-22

🚀 Features

• (functions) Add 18 new Neo4j function mappings for v0.5.5
• (functions) Add 30 more Neo4j function mappings for v0.5.5
• (functions) Add ClickHouse function pass-through via ch:: prefix
• (functions) Add ClickHouse aggregate function pass-through via ch. prefix
• (functions) Add chagg. prefix for explicit aggregates, expand aggregate registry to ~150 functions
• (benchmark) Add LDBC SNB Interactive v1 benchmark
• (benchmark) Add ClickGraph schema matching datagen format
• (benchmark) Add LDBC query test script
• (ldbc) Achieve 100% LDBC BI benchmark (26/26 queries)
• Implement chained WITH clause support with CTE generation
• Support ORDER BY, SKIP, LIMIT after WITH clause
• Implement size() on patterns with schema-aware ID lookup
• Add composite node ID infrastructure for multi-column primary keys
• Add CTE reference validation
• CTE-aware variable resolution for WITH clauses
• Fix CTE column filtering and JOIN condition rewriting for WITH clauses
• CTE-aware variable resolution + WITH validation + documentation improvements
• Add lambda expression support for ClickHouse passthrough functions
• Add comprehensive LDBC benchmark suite with loading, query, and concurrency tests
• Implement scope-based variable resolution in analyzer (Phase 1)
• Remove dead CTE validation functions
• Implement CTE column resolution across all join strategies
• Remove obsolete JOIN rewriting code from renderer (Phase 3D-A)
• Move CTE column resolution to analyzer (Phase 3D-B)
• Pre-compute projected columns in analyzer (Phase 3E)
• Add CTE schema registry for analyzer (Phase 3F)
• Use pre-computed projected_columns in renderer (Phase 3E-B)
• Implement cross-branch shared node JOIN detection
• Allow disconnected comma patterns with WHERE clause predicates
• Support multiple sequential MATCH clauses
• Implement generic CTE JOIN generation using correlation predicates
• Complete LDBC SNB schema and data loading infrastructure
• Improve relationship validation error messages
• Clarify node_id semantics as property names with auto-identity mappings
• Complete composite node_id support (Phase 2)
• Add polymorphic relationship resolution architecture
• Complete polymorphic relationship resolution data flow
• Fix polymorphic relationship resolution in CTE generation
• Add Comment REPLY_OF Message schema definition
• Add schema entity collection in VariableResolver for Projection scope
• Add dedic...

Release v0.6.0

[0.6.0] - 2025-12-22

🚀 Features

• (functions) Add 18 new Neo4j function mappings for v0.5.5
• (functions) Add 30 more Neo4j function mappings for v0.5.5
• (functions) Add ClickHouse function pass-through via ch:: prefix
• (functions) Add ClickHouse aggregate function pass-through via ch. prefix
• (functions) Add chagg. prefix for explicit aggregates, expand aggregate registry to ~150 functions
• (benchmark) Add LDBC SNB Interactive v1 benchmark
• (benchmark) Add ClickGraph schema matching datagen format
• (benchmark) Add LDBC query test script
• (ldbc) Achieve 100% LDBC BI benchmark (26/26 queries)
• Implement chained WITH clause support with CTE generation
• Support ORDER BY, SKIP, LIMIT after WITH clause
• Implement size() on patterns with schema-aware ID lookup
• Add composite node ID infrastructure for multi-column primary keys
• Add CTE reference validation
• CTE-aware variable resolution for WITH clauses
• Fix CTE column filtering and JOIN condition rewriting for WITH clauses
• CTE-aware variable resolution + WITH validation + documentation improvements
• Add lambda expression support for ClickHouse passthrough functions
• Add comprehensive LDBC benchmark suite with loading, query, and concurrency tests
• Implement scope-based variable resolution in analyzer (Phase 1)
• Remove dead CTE validation functions
• Implement CTE column resolution across all join strategies
• Remove obsolete JOIN rewriting code from renderer (Phase 3D-A)
• Move CTE column resolution to analyzer (Phase 3D-B)
• Pre-compute projected columns in analyzer (Phase 3E)
• Add CTE schema registry for analyzer (Phase 3F)
• Use pre-computed projected_columns in renderer (Phase 3E-B)
• Implement cross-branch shared node JOIN detection
• Allow disconnected comma patterns with WHERE clause predicates
• Support multiple sequential MATCH clauses
• Implement generic CTE JOIN generation using correlation predicates
• Complete LDBC SNB schema and data loading infrastructure
• Improve relationship validation error messages
• Clarify node_id semantics as property names with auto-identity mappings
• Complete composite node_id support (Phase 2)
• Add polymorphic relationship resolution architecture
• Complete polymorphic relationship resolution data flow
• Fix polymorphic relationship resolution in CTE generation
• Add Comment REPLY_OF Message schema definition
• Add schema entity collection in VariableResolver for Projection scope
• Add dedicated LabelInference analyzer pass
• Enhance TypeInference to infer both node labels and edge types
• Reduce MAX_INFERRED_TYPES from 20 to 5
• (parser) Add clear error messages for unsupported pattern comprehensions
• (parser) Add clear error messages for bidirectional relationship patterns
• (parser) Convert temporal property accessors to function calls
• (analyzer) Add UNWIND variable scope handling to variable_resolver
• (analyzer) Add type inference for UNWIND elements from collect() expressions
• Support path variables in comma-separated MATCH patterns
• Add polymorphic relationship resolution with node types
• Complete collect(node) + UNWIND tuple mapping & metadata preservation architecture
• Make CLICKHOUSE_DATABASE optional with 'default' fallback
• Add parser support for != (NotEqual) operator
• Add unified test schema for streamlined testing
• Add unified test data setup and fix matrix test schema issues
• Complete multi-tenant parameterized view support
• Add denormalized flights schema to unified test schema
• Add VLP transitivity check to prevent invalid recursive patterns

🐛 Bug Fixes

• (benchmark) Use Docker-based LDBC data generation
• (benchmark) Align DDL with actual datagen output format
• (benchmark) Add ClickHouse credentials support
• (benchmark) Align DDL and schema with actual datagen output
• (ldbc) Fix CTE pattern for WITH + table alias pass-through
• (ldbc) Fix ic3 relationship name POST_IS_LOCATED_IN -> POST_LOCATED_IN
• WITH+MATCH CTE generation for correct SQL context
• Replace all silent defaults with explicit errors in render_expr.rs
• Eliminate ViewScan silent defaults - require explicit relationship columns
• Expand WITH TableAlias to all columns for aggregation queries
• Track CTE schemas to build proper property_mapping for references
• Remove CTE validation to enable nested WITH clauses
• Prevent duplicate CTE generation in multi-level WITH queries
• Three-level WITH nesting with correct CTE scope resolution
• Add proper schemas to WITH/HAVING tests
• Correct CTE naming convention to use all exported aliases
• Coupled edge alias resolution for multiple edges in same table
• Rewrite expressions in intermediate CTEs to fix 4-level WITH queries
• Add GROUP BY and ORDER BY expression rewriting for final queries
• Issue #6 - Fix Comma Pattern and NOT operator bugs
• Resolve 3 critical LDBC query blocking issues
• (ldbc) Inline property matching & semantic relationship expansion
• (ldbc) Handle IS NULL checks on relationship wildcards (IS7)
• (ldbc) Fix size() pattern comprehensions - handle internal variables correctly (BI8)
• (ldbc) Rewrite path functions in WITH clause (IC1)
• Strip database prefixes from CTE names for ClickHouse compatibility
• Cartesian Product WITH clause missing JOIN ON
• Operator precedence in expression parser
• VLP endpoint JOINs with alias rewriting for chained patterns
• Correct NOT operator precedence and remove hardcoded table fallbacks
• Three critical shortestPath and query execution bugs
• Extend VLP alias rewriting to WHERE clauses for IC1 support
• Use correct CTE names for multi-variant relationship JOINs
• Remove database prefix from CTE table names in cross-branch JOINs
• Hoist trailing non-recursive CTEs to prevent nesting scope issues
• VLP + WITH label corruption bug - use node labels in RelationshipSchema
• Resolve compilation errors from AST and GraphRel changes
• Add fallback to lookup table names from relationship schema
• Complete RelationshipSchema refactoring - all 646 tests passing
• Add database prefixes to base table JOINs
• Use underscore convention for CTE column aliases
• Thread node labels through relationship lookup pipeline for polymorphic relationships
• Support filtered node views in relationship validation
• Add JOIN dependency sorting to CTE generation path
• Use existing TableCtx labels in multi-pattern MATCH label inference
• TypeInference creates ViewScan for inferred node labels
• QueryValidation respects parser normalization
• Populate from_id/to_id columns during JOIN creation for correct NULL checks
• (ldbc) Align BI queries with LDBC schema definitions
• Prevent RefCell panic in populate_relationship_columns_from_plan
• UNWIND after WITH now uses CTE as FROM table instead of system.one
• Replace all panic!() with log::error!() - PREVENT SERVER CRASHES
• Clean up unit tests - fix 21 compilation errors
• Complete unit test cleanup - fix assertions and mark unimplemented features
• Replace non-standard LIKE syntax with proper OpenCypher string predicates
• Add != operator support to comparison expression parser
• Preserve database prefix in ViewTableRef SQL generation
• Relationship variable expansion + consolidate property helpers
• Use relationship alias for denormalized edge FROM clause
• Re-enable selective cross-branch JOIN for comma-separated patterns
• Rel_type_index to prefer composite keys over simple keys
• WITH...MATCH pattern using wrong table for FROM clause
• Update test labels to match unified_test_schema
• Test_multi_database.py - use schema_name instead of database for USE clause
• Unify aggregation logic and fix multi-schema support
• Multi-table label bug fixes and error handling improvements

💼 Other

• Fix dependency vulnerabilities for v0.5.5
• Partial fix for nested WITH clauses - add recursive handling
• Multi-variant CTE column name resolution in JOIN conditions
• SchemaInference using table names instead of node labels

🚜 Refactor

• Fix compiler warnings and clean up unused variables
• (functions) Change ch:: to ch. prefix for Neo4j ecosystem compatibility
• Extract TableAlias expansion into helper functions
• Replace wildcard expansion in build_with_aggregation_match_cte_plan with helper
• Remove deprecated v1 graph pattern handler (1,568 lines)
• Extract CTE hoisting helper function
• Remove unused ProjectionKind::With enum variant
• Remove 676 lines of dead WITH clause handling code
• Remove 47 lines of dead GraphNode branch with empty property_mapping
• Remove redundant variable resolution from renderer (Phase 3A)
• Remove unused bidirectional and FK-edge functions
• Remove dead code function find_cte_in_plan
• Consolidate duplicate property extraction code (-23 lines)
• Remove dead extract_ctes() function (-301 lines)
• Separate graph labels from table names in RelationshipSchema
• Remove redundant WithScopeSplitter analyzer pass
• Remove old parsing-time label inference
• Consolidate inference logic into TypeInference with polymorphic support
• Replace hardcoded fallbacks with descriptive errors
• Add strict validation for system.one usage in UNWIND
• ELIMINATE ALL HARDCODED FALLBACKS - fail fast instead
• Consolidate test data setup - use MergeTree, remove duplicates

📚 Documentation

• Update wiki documentation for v0.5.4 release
• Archive wiki for v0.5.4 release
• Add UNWIND clause documentation to wiki
• Update v0.5.4 wiki snapshot with UNWIND documentation
• Update Known-Limitations with recently implemented features
• Update v0.5.4 wiki snapshot with corrected feature status
• Add 30 new functions to Cypher-Functions.md reference
• Expand vector similarity section with RAG usage
• Clarify scalar vs aggregate function categories in ch.* docs
• Add lambda expression limitation to ch.* pass-through documentation
• Split ClickHous...

Release v0.5.4

[0.5.4] - 2025-12-08

🚀 Features

• Add native support for self-referencing FK pattern
• Add relationship uniqueness enforcement for undirected patterns
• (schema) Add fixed-endpoint polymorphic edge support
• (union) Add UNION and UNION ALL query support
• Multi-table label support and denormalized schema improvements
• (pattern_schema) Add unified PatternSchemaContext abstraction - Phase 1
• (graph_join_inference) Integrate PatternSchemaContext - Phase 2
• (graph_join_inference) Add handle_graph_pattern_v2 - Phase 3
• (pattern_schema) Add FkEdgeJoin strategy for FK-edge patterns
• (graph_join) Wire up handle_graph_pattern_v2 with USE_PATTERN_SCHEMA_V2 env toggle

🐛 Bug Fixes

• GROUP BY expansion and count(DISTINCT r) for denormalized schemas
• Undirected multi-hop patterns generate correct SQL
• Support fixed-endpoint polymorphic edges without type_column
• Correct polymorphic filter condition in graph_join_inference
• Normalize GraphRel left/right semantics for consistent JOIN generation
• Recurse into nested GraphRels for VLP detection
• (render_plan) Add WHERE filters for VLP chained pattern endpoints (Issue #5)
• (parser) Reject binary operators (AND/OR/XOR) as variable names
• Multi-hop anonymous patterns, OPTIONAL MATCH polymorphic, string operators
• Aggregation and UNWIND bugs
• Denormalized schema query pattern fixes (TODO-1, TODO-2, TODO-4)
• Cross-table WITH correlation now generates proper JOINs (TODO-3)
• WITH clause alias propagation through GraphJoins wrapper (TODO-8)
• Multi-hop denormalized edge JOIN generation
• Update schema files to match test data columns
• (pattern_schema) Pass prev_edge_info for multi-hop detection in v2 path
• (filter_tagging) Correct owning edge detection for multi-hop intermediate nodes
• FK-edge JOIN direction bug - use join_side instead of fk_on_right
• Add polymorphic label filter generation for edges

🚜 Refactor

• Unify FK-edge pattern for self-ref and non-self-ref cases
• Minor code cleanup in bidirectional_union and plan_builder_helpers
• Make PatternSchemaContext (v2) the default join inference path
• Reorganize benchmarks into individual directories
• Replace NodeIdSchema.column with Identifier-based id field
• Change YAML field id_column to node_id for consistency
• Extract predicate analysis helpers to plan_builder_helpers.rs
• Extract JOIN and filter helpers to plan_builder_helpers.rs

📚 Documentation

• Update README for v0.5.3 release
• Add fixed-endpoint polymorphic edge documentation
• Add VLP+chained patterns docs and private security tests
• Document Issue #5 (WHERE filter on VLP chained endpoints)
• (readme) Minor wording improvements
• Update PLANNING_v0.5.3 and CHANGELOG with bug fix status
• Add unified schema abstraction proposal and test scripts
• Add unified schema abstraction Phase 4 completion to STATUS
• Update unified schema abstraction progress - Phase 4 fully complete
• (benchmarks) Add ClickHouse env vars and fix paths in README
• (benchmarks) Streamline README to be a concise index
• Archive PLANNING_v0.5.3.md - all bugs resolved

🧪 Testing

• Add multi-hop pattern integration tests
• Fix Zeek integration tests - response format and skip cross-table tests
• Add v1 vs v2 comparison test script
• Add unit tests for predicate analysis helpers

⚙️ Miscellaneous Tasks

• Update CHANGELOG.md [skip ci]
• Make test files use CLICKGRAPH_URL env var for port flexibility
• (benchmarks) Move social_network-specific files to subdirectory

Release v0.5.3

[0.5.3] - 2025-12-02

🚀 Features

• Add regex match (=~) operator and fix collect() function
• Add EXISTS subquery and WITH+MATCH chaining support
• Add label() function for scalar label return

🐛 Bug Fixes

• Remove unused schemas volume from docker-compose
• Parser now rejects invalid syntax with unparsed input
• Column alias for type(), id(), labels() graph introspection functions
• Update release workflow to use clickgraph binary name
• Update release workflow to use clickgraph-client binary name
• Build entire workspace in release workflow

📚 Documentation

• Archive wiki for v0.5.2 release
• Fix schema documentation and shorten README
• Fix Quick Start to include required GRAPH_CONFIG_PATH
• Add 3 new known issues from ontime schema testing
• Update KNOWN_ISSUES.md - WHERE AND now caught
• Clean up KNOWN_ISSUES.md - remove resolved issues
• Remove false known limitations - all verified working

⚙️ Miscellaneous Tasks

• Update CHANGELOG.md [skip ci]
• Release v0.5.3
• Update CHANGELOG.md [skip ci]
• Update Cargo.lock for v0.5.3
• Update CHANGELOG.md [skip ci]
• Update CHANGELOG.md [skip ci]
• Update CHANGELOG.md [skip ci]