Tags give the ability to mark specific points in history as being important
-
v0.5.2
df07711d · ·v0.5.2 — Source-table-direct baseline, UNION SQL gen, optimize hint, ANALYZE TABLE Key changes: - Source-table-direct battery runner (no managed edges) - UNION / UNION ALL SQL generation for SQL-only backends - execution_mode='sql' on direct runner (no silent fallbacks) - optimize parameter on engine + /*+ optimize=false */ query hint - --analyze flag for COMPUTE STATISTICS on all tables - Spark installed on a remote host (Java 17 + PySpark 4.1.1) Performance (SF10, cold-cache): - Direct: 199/201, 19.7 min (2.2x faster than managed) - Managed: 201/201, 43.8 min
-
v0.5.1-perf-baseline
da828ede · ·Pre-table-per-rel-type baseline — single edges table + node_id_map + views 201/201 LDBC battery, PK/FK RELY, Liquid Clustering, ANALYZE TABLE. FK elimination works on direct queries but blocked through views. Performance baseline for comparison with table-per-rel-type architecture.
-
v0.5.1
c80873ca · ·201/201 LDBC battery on Databricks SF10 — ZERO FAILS Complete LDBC SNB battery pass on managed edges (143.9M edges, 27.2M nodes). Every query in the C-series generates correct SQL and executes successfully. Session arc: 107 -> 201 over two working days. Key deliverables: - Single SQL path (tree serializer, 148 lines) - ~9600 lines legacy code deleted - Iterative UNION ALL for bounded VL paths - Managed-edge views with source-table property LEFT JOIN - SQLRuntimeError vs SQLGenerationError distinction - Optimizer fusion fix for VL MATCH groups - Pre-VL CTE alias rewrite for all variables - sql_fallback warning for observable performance cliffs
-
v0.5.1-rc1
fe2bbc94 · ·200/201 LDBC battery on Databricks SF10 (C85 sole residual - managed edges property gap) Session deliverables: - 107 -> 200/201 on Databricks SF10 LDBC battery - ~9600 lines dead code deleted (legacy interpreter + sql_builder.py) - Single SQL path (tree serializer only) - Iterative UNION ALL for bounded VL paths - SQLRuntimeError vs SQLGenerationError distinction - sql_fallback warning for observable performance cliffs - Spark dialect consolidated - Optimizer fusion fix for VL MATCH groups - Pre-VL CTE alias rewrite for all variables
-
-
v0.4.20
f074668e · ·Release v0.4.20: filter_predicate + CYCLE detection Features: - filter_predicate on NodeMapping: Map multiple graph labels to the same base table using a discriminator column filter. Injected on reads at 5 sites; writes skip it. Supports YAML config. - CYCLE clause for PostgreSQL recursive CTEs (14+): Prevents infinite loops on cyclic graphs. Body filter adds NOT is_cycle alongside depth guard. Optimizer preserves cycle_columns through all passes. Tests: 26 new tests across 4 files. All 632 Cypher + 134 SPARQL passing.
-
-
v0.4.16
47982312 · ·v0.4.16 - Query battery 183/183, 0 timeouts (SF10, 157s) - Selective anti-join deferral past WITH when CTE has folded LIMIT - Fix unlabeled-node UNION branch joins (C149 UNRESOLVED_COLUMN) - LabelCheckExpr translation (content:Post → __label__ = 'Post') - Anti-join end-node FK fix, SemiJoin elimination fix - Transitive predicate inference, SemiJoin IR node - Native LEFT ANTI JOIN, ~3600 lines deleted
-
v0.4.15
528493b6 · ·feat: Tree-based SQL pipeline as default backend, SPARQL parity for tree pipeline
-
-
-
-
-
v0.4.3
90ec8606 · ·ProAct inference adapters - SqlRulesAdapter: SQL-based inference with automatic SQL Warehouse routing - EnsembleAdapter: Combine multiple adapters with weighted_average, max, noisy_or, voting - SparqlRulesAdapter: SPARQL-based inference rules - SQL Warehouse integration provides 3-5x speedup on complex queries - Standardized documentation naming to lowercase-kebab-case
-
v0.4.0
ea6973ba · ·v0.4.0 - Mutation support milestone Features: - Complete SPARQL 1.1 Update support (INSERT/DELETE DATA, MODIFY, CLEAR/DROP GRAPH) - Cypher mutations (CREATE, SET, DELETE, DETACH DELETE, MERGE, REMOVE) - Mutable Pandas backend with MutableDataFrameMixin - Mutable Spark backend with Delta Lake support - ID generation strategies (uuid, sequence)
-
-
-
-
-
v0.3.71
6e3b9e95 · ·Release v0.3.71: Support arbitrary length MATCH chains with WITH pushdown - _merge_with_prior_results reconstructs node-like dicts for new variables - _extract_bound_ids handles flat SQL result format (var.id keys) - Enables MATCH-WITH-MATCH-WITH-MATCH... patterns of any length - New tests: TestArbitraryMatchChains in test_with_pushdown.py - Verified on Databricks with LDBC SF10 (triple MATCH chain in 5.85s)