release: 0.6.0 — openCypher TCK 91.5%→97.7% (+239)

.metis/initiatives/GQLITE-I-0049/initiative.md

@@ -81,3 +81,34 @@ sub-steps once started (human check-in before decomposing, per Metis HITL).
 
 - 2026-05-30: Initiative created. Structure approved (initiative + task-per-cluster).
   Starting with Temporal (T-A).
+- 2026-06-02: **WithOrderBy (T-D) cluster CLOSED (+10, PR #91).** `WithOrderBy1`
+  [45] (all 10 type examples). Two root causes: (1) a non-aggregating WITH's
+  ORDER BY over a *kept* input variable was only applied on the outer SELECT, so a
+  downstream aggregating WITH (`collect`) never saw sorted rows — now also pushed
+  into the CTE body via `order_expr_all_live_input` gate in `transform_with.c`
+  (SQLite preserves an ordered subquery's order through `json_group_array`);
+  (2) `gql_order_cmp_func`'s temporal compare used `parse_temporal_ns` whose
+  `epoch*1e9` overflows int64 ~year 2262, so year-9999 datetimes sorted *smallest*
+  — new overflow-safe `cmp_temporal_strings()` compares `(epoch_seconds, sub_ns)`.
+  Verified zero regressions, 3776→3786, unit 944/944, functional clean.
+- 2026-06-02: **Merge (T-C) — deep architectural blockers identified, deferred.**
+  Investigated the undirected MATCH+MERGE+RETURN cluster (Merge5 [12]/[13]).
+  Root cause of row-doubling: the RETURN-after-MERGE re-query builds a synthetic
+  combined MATCH (original MATCH pattern + MERGE pattern). For undirected
+  `(a)-[r]-(b)` whose endpoints reuse MATCH variables, the transform re-emits the
+  shared endpoints into a *reset* FROM and drops the MATCH nodes' property
+  constraints, so the undirected OR matches each edge twice. A property-transplant
+  fix (copy MATCH node props onto MERGE endpoints, drop redundant standalone nodes)
+  was prototyped but blocked by a *second* transform bug: a fresh-fresh undirected
+  connected pattern in the re-query drops endpoint property joins entirely (the
+  parsed-equivalent query emits them fine — the re-query path resets the builder
+  mid-emission). Reverted (no commit). Additional Merge blockers: (a) MERGE
+  processes only the FIRST match row (`break` in `executor_merge.c` ~line 1088), so
+  multi-row MATCH+MERGE can't bind/return per-row (Merge8 [1] expects +3 rels);
+  (b) `handle_create_return` is a hand-rolled projection that doesn't handle
+  aggregate RETURN items (`count(*)`/`count(a)` → `?column?`/null) — gates
+  Merge5 [4], Merge9 [3]. These need transform/executor rework, not a contained
+  fix; left for a dedicated effort.
+- 2026-06-02: Comparison2 [3] confirmed blocked by boolean-type-preservation
+  (`WHERE <bool-var>` and `WHERE b = true` both filter all rows). List12 [1] is the
+  entity-in-list property access deep item (`x.name` on collected nodes → null).

CHANGELOG.md

@@ -4,6 +4,58 @@ All notable changes to GraphQLite are documented here. Format loosely follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versions follow
 [Semantic Versioning](https://semver.org/).
 
+## [0.6.0] — 2026-06-03
+
+A large openCypher conformance release. TCK pass rate moves from **91.5%**
+(3549 executable scenarios) to **97.7%** (3788 / 3876), **+239 scenarios** with
+zero regressions among previously-passing scenarios. 944/944 unit tests pass;
+functional tests clean.
+
+### Added — openCypher coverage
+
+- **Pattern & path matching completeness** — variable-length path hydration,
+  `relationships()` / `nodes()` on varlen and null paths, path-wide relationship
+  uniqueness, multi-rel OPTIONAL MATCH with full-pattern (all-or-none) semantics,
+  bidirectional bracketed relationship patterns, and four-label conjunctions in
+  expression context.
+- **Temporal** — closed the non-DST temporal cluster: duration arithmetic and
+  ISO round-trip normalization (months not normalized across days; 1 month =
+  30.436875 days), `duration.between`, component accessors, `date`/`time`/
+  `datetime`/`localdatetime` construction and selection, named-zone handling, and
+  UTC-instant ordering. Cross-type ordering now uses an orderability comparator
+  with a NaN sentinel and an overflow-safe `(seconds, nanoseconds)` temporal
+  comparison (far-future years no longer wrap int64 epoch-nanoseconds).
+- **Quantifiers** — `all` / `any` / `none` / `single` over list pipelines now
+  evaluate consistently; non-deterministic CTEs (`rand()`/`RANDOM()` in WITH
+  projections) are emitted `AS MATERIALIZED` so a multiply-referenced list is
+  evaluated once, fixing algebraic-identity scenarios.
+- **WITH / ORDER BY / aggregation** — `ORDER BY` on a non-aggregating WITH now
+  flows to a downstream aggregating WITH (`… ORDER BY x WITH collect(y)` collects
+  in sorted order); computed (non-aggregate) WITH keys (e.g. `a.n % 3 AS m`) are
+  now GROUP BY keys; `LIMIT`/`OFFSET` inline into the WITH CTE body.
+- **Existential subqueries** — `EXISTS { pattern }` brace form.
+- **Validation** — compile-time `SyntaxError` diagnostics (pattern expressions in
+  `SET` RHS, relationship-uniqueness, label/type 3-valued-logic predicates, and
+  more).
+
+### Fixed
+
+- Multi-row `MATCH … CREATE` / `MATCH … DELETE … CREATE` runs the write per
+  matched row; inter-pattern variable references in `CREATE` (`{x: a.id}`)
+  resolve against earlier patterns; `RETURN` is pre-captured before `DELETE`.
+- `REMOVE` on a null/unbound variable is a no-op; `RETURN *` excludes synthetic
+  anonymous aliases.
+- General-expression projection in `CREATE … RETURN` and `SET` (list/JSON props).
+
+### Notes
+
+Remaining gaps (~2.3%) concentrate in DST-aware timezone arithmetic, nested
+existential subqueries, multi-row MERGE binding, and boolean subtype preservation
+(SQLite has no native boolean type). See
+[`docs/testing/semantic-coverage-matrix.md`](docs/testing/semantic-coverage-matrix.md).
+Windows extension tests (`test_*_timestamp*`) remain non-blocking pending an
+MSYS2/MinGW `julianday('now')` regression (GQLITE-T-0205).
+
 ## [0.5.0] — 2026-05-22
 
 Significant TCK conformance release plus the first-time Windows extension

README.md

@@ -43,6 +43,25 @@ g.dijkstra("alice", "bob")
 - **Zero configuration** — Works with any SQLite database, no server required
 - **Multiple bindings** — Python, Rust, and raw SQL interfaces
 
+## openCypher Conformance
+
+GraphQLite is validated against the official [openCypher Technology Compatibility
+Kit (TCK)](https://github.com/opencypher/openCypher) — the canonical conformance
+suite for the Cypher query language. Current coverage:
+
+| Area | Scenarios | Passing |
+|------|-----------|---------|
+| **Overall** | **3,876** | **97.7%** |
+| Expressions (temporal, lists, maps, comparison, literals, …) | 2,599 | 98.0% |
+| Clauses (MATCH, WITH, MERGE, CREATE, SET, DELETE, UNWIND, …) | 1,247 | 97.2% |
+| Use cases (triadic selection, subgraph counting) | 30 | 100% |
+
+Run it yourself with `angreal test tck`. The remaining gaps are tracked in
+[`docs/testing/semantic-coverage-matrix.md`](docs/testing/semantic-coverage-matrix.md)
+and concentrate in a few deep areas (DST-aware timezone arithmetic, nested
+existential subqueries, multi-row MERGE). Booleans, strings, null handling,
+CREATE/SET/DELETE/REMOVE, UNION, and SKIP/LIMIT are at 100%.
+
 ## Documentation
 
 **[Full Documentation](https://colliery-io.github.io/graphqlite/)** — Tutorials, how-to guides, and API reference

bindings/python/src/graphqlite/__init__.py

@@ -8,7 +8,7 @@
 from .utils import escape_string, sanitize_rel_type, CYPHER_RESERVED
 from ._platform import get_loadable_path
 
-__version__ = "0.5.0"
+__version__ = "0.6.0"
 __all__ = [
     "BulkInsertResult",
     "Connection", "connect", "wrap", "load", "loadable_path",

bindings/rust/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "graphqlite"
-version = "0.5.0"
+version = "0.6.0"
 edition = "2021"
 description = "SQLite extension for graph queries using Cypher"
 license = "MIT"