Skip to content

v0.2.0

Latest
Latest

Choose a tag to compare

View all tags
@rjwalters rjwalters released this 15 Jun 20:28
· 638 commits to main since this release
v0.2.0
1c4d56b

This release introduces a Raft-based replication track (new vibesql-consensus crate) and turns on the MVCC visibility + on-demand GC subsystem end-to-end. 369 commits since v0.1.4 also push SQLite compatibility forward — full SQLITE_MAX_TRIGGER_DEPTH (1000) trigger semantics, window-function correctness, ALTER TABLE RENAME COLUMN with trigger-body rewrite — and bring EXPLAIN QUERY PLAN output much closer to sqlite3 parity.

Replication & Consensus

A new vibesql-consensus workspace crate built on openraft introduces single-group whole-database replication. The MVCC state machine applies committed transactions from the Raft log; HTTP and prepared-statement writes can now be routed through consensus.

  • Durable Raft log + vote persistence (#5357)
  • In-process channel network + multi-node test cluster (#5364)
  • TCP transport + cluster config + make test-cluster (#5368)
  • Network snapshot transfer + durable snapshots + purge safety (#5369, #5374)
  • MVCC state machine — apply committed txns from the Raft log (#5378, #5384)
  • Freeze nondeterministic SQL at propose time (#5377, #5382)
  • Freeze volatile DEFAULTs + validate replicated trigger bodies (#5396)
  • Linearizable leader reads + stale-leader fencing (#5387)
  • Bounded-staleness follower reads + read-your-writes tokens (#5390)
  • Reject residual TIME→TIMESTAMP casts over non-target operands (#5398, #5405)
  • Boot TCP test clusters on pre-bound ephemeral listeners (#5507, #5514)

Replicated server surface

The server now exposes the consensus surface across its HTTP and PostgreSQL-wire endpoints.

  • Replicated operational surface — prepared-stmt writes, /health role, HTTP gating (#5411)
  • Interactive transactions (BEGIN…COMMIT) in replicated sessions (#5402)
  • Mid-transaction RYW + savepoints in replicated sessions (#5413)
  • Route HTTP /api/query + CRUD-create writes through consensus (#5423)
  • Route HTTP CRUD by-id endpoints through consensus (#5441)
  • Route HTTP GraphQL queries + mutations through consensus (#5447)
  • Route HTTP blob storage writes through consensus (#5457)
  • Buffer prepared EXECUTE inside a replicated transaction (#5430)
  • Route replicated sessions' writes through MvccRaftNode (#5395)
  • Feed HTTP SSE subscriptions from applied consensus entries (#5452)
  • Coalesce apply-path change events for replicated subscriptions (#5456, #5467)
  • Names-only replicated Describe to avoid materializing rows (#5503)
  • Prune replicated subscription re-queries by changed PK (#5496)
  • Resolve real column names in replicated SELECT results (#5428)

MVCC

VibeSQL's MVCC subsystem now covers the read path end-to-end, with on-demand garbage collection wired to VACUUM.

  • Stamp xmin/xmax on INSERT/UPDATE/DELETE behind mvcc_enabled (#5193)
  • TxnSnapshot + Row::visible_to predicate (Phase 1b) (#5184)
  • Add xmin/xmax to Row + bump vbsql to v7 (Phase 1a) (#5142)
  • Phase 1d read-path visible_to filtering + FK deferred-replay coordination (#5209)
  • Extend visible_to to index scans, PK lookup, UNIQUE scans (#5204, #5222)
  • Widen snapshots for autocommit + in-txn self-write visibility (#5223)
  • On-demand vacuum_mvcc for old-version garbage collection (#5224)
  • Thread visibility filter through DML read sites (#5205, #5225)
  • Enable SIMD/columnar fast paths under mvcc_enabled (#5227)
  • Parse VACUUM / VACUUM INTO and map execution to MVCC GC (#5226)

SQL Compatibility

Triggers

Trigger semantics got a heavy pass to align with sqlite3 3.51.

  • Honor sqlite3_limit(SQLITE_LIMIT_TRIGGER_DEPTH) at runtime (#5605)
  • Reach SQLite's full SQLITE_MAX_TRIGGER_DEPTH (1000) via on-demand stack growth (#5610)
  • Raise recursion depth cap 16 → 700 toward SQLite (#5533)
  • Honor PRAGMA recursive_triggers = off (#5550)
  • Schema-aware trigger firing for temp vs. main triggers (#5592)
  • Parse CREATE TEMP/TEMPORARY TRIGGER + schema-qualified names (#5532)
  • Parse CREATE TRIGGER: IF NOT EXISTS, reject FOR EACH STATEMENT, surface body syntax errors (#5495)
  • Reject CREATE TRIGGER with bound parameter in body / WHEN (#5499)
  • Accept unparenthesized UPDATE OF column list (#5580)
  • Accept subquery in trigger WHEN clause (#5586)
  • Support RAISE() in trigger bodies (#5415)
  • Reject RAISE() outside trigger programs at parse time (#5424)
  • Resolve NEW/OLD in trigger WHEN-clause subqueries (#5585, #5593)
  • Resolve NEW.rowid / OLD.rowid (oid / rowid) in trigger bodies (#5519)
  • Live columnar values for BEFORE/AFTER UPDATE triggers (#5543, #5546)
  • Live count(*) for combined BEFORE+AFTER DELETE triggers (#5542)
  • Interleave BEFORE/AFTER row triggers per-row in multi-row DML (#5504)
  • Cascade-drop triggers ON a dropped table (temp lifecycle) (#5597)
  • Cascade-drop INSTEAD OF triggers when their view is dropped (#5604)
  • List temp views + temp triggers in sqlite_temp_master (#5582)
  • Transactional DDL rollback + DROP TRIGGER IF EXISTS (#5506)
  • UPDATE OR REPLACE/IGNORE conflict resolution + DELETE triggers on WITHOUT ROWID (#5522)
  • Fire child UPDATE triggers on ON DELETE SET NULL/SET DEFAULT cascades (#5518)
  • Statement-end orphan FK check after cascade-fired RAISE(IGNORE) (#5498)
  • Auto-commit statement atomicity for cascade / RAISE(ABORT) and multi-row DML (#5473)
  • OLD/NEW in INSERT…SELECT trigger bodies + trigger*.test triage (#5493)
  • Rewrite trigger table refs + verbatim SQL on ALTER TABLE RENAME (#5510)

Window Functions

  • Match SQLite last-pass row order and sort GROUP BY output by key (#5311)
  • Propagate PARTITION BY evaluation errors instead of NULL-keying (#5305)
  • Honor NULLS FIRST/LAST and collation in sort and RANGE frames (#5297)
  • Resolve aggregate args of embedded window functions via hidden columns (#5280)
  • Evaluate window functions mixed with aggregates in one SELECT (#5265)
  • Window IN-subqueries and ordinal ORDER BY through wildcards (#5256)
  • window1.test residuals — nested scope binding, UNION-arm misuse, text coercion (#5273)
  • WHERE push-down into window subqueries + covering-index window-sort EQP (#5349)
  • Pre-compute aggregates in OVER (PARTITION BY/ORDER BY/frame) (#5127)
  • Reject non-window aggregates in ORDER BY of window queries (#5094, #5119)
  • Detect window-function misuse in positional GROUP BY (#5111)
  • COUNT()/COUNT(*) over post-aggregate frames now counts all frame rows (#5089)
  • Preserve REAL type in sum() over float inputs
  • GROUP_CONCAT dispatch and RANGE ORDER BY validation (#5058)
  • Thread case_sensitive_like PRAGMA through all filter paths
  • Recurse through all expression variants for window detection (#5095, #5121)

ALTER TABLE

  • ALTER TABLE RENAME COLUMN with trigger-body column rewrite (#5602)
  • Abort RENAME COLUMN on ambiguous trigger-body column ref (#5607)
  • Edit verbatim sqlite_master.sql in place for DROP COLUMN / RENAME TO (#5635)
  • Sync catalog schema on ALTER + edit verbatim SQL in place (#5633)
  • Match sqlite3 wording for ALTER TABLE RENAME TO collision (#5560)

Foreign Keys

  • Partial-index awareness unblocks fkey1-6.0/6.1/6.2 (#5213)
  • Wire sqlite3_db_status DBSTATUS_DEFERRED_FKS bridge (#5211)
  • Persist DEFERRABLE clause and skip auto-save inside txn (#5185)
  • Self-ref multi-row INSERT + tcl-shim multi-table parse (#5180)
  • Collation-aware comparison in CASCADE/SET NULL/SET DEFAULT helpers (#5176)
  • Self-FK + deferral introspection (Phase C3 of #5085) (#5141)
  • Deferred FK violation queue + COMMIT-time enforcement (Phase C2 of #5085) (#5133)
  • Catalog metadata + PRAGMA defer_foreign_keys (Phase C1 of #5085) (#5124)
  • Emit "foreign key mismatch" error class for invalid FK targets (#5120)
  • Fire child row triggers on FK ON DELETE/UPDATE CASCADE (#5463)

Parser

  • Parse VACUUM / VACUUM INTO (#5226)
  • Support WITH clause before VALUES statement (#5358)
  • Support date() and time() as scalar function calls (#5316)
  • Accept TEMP/TEMPORARY modifier on CREATE TRIGGER (#5229)
  • Support partial-index WHERE clauses in CREATE INDEX (#5109)
  • Support RESTRICT, post-constraint DEFAULT, empty statements (#5088)

EXPLAIN QUERY PLAN

EQP output moved much closer to sqlite3 parity this cycle.

  • Expand views / subqueries in EQP output (windowpushd) (#5354)
  • Flatten plain views in EQP output (#5360)
  • Render blocked view bodies as CO-ROUTINE blocks in EQP (#5366)
  • Emit temp B-tree annotations for GROUP BY/DISTINCT/compound (#5370)
  • Ordering-index scan lines + compound ORDER BY shape (#5372)
  • Suppress ORDER BY temp B-tree for rowid-alias natural-order scans (#5380)
  • Emit EQP window-sort entries for OVER-clause sorting passes (#5244)
  • Suppress statement ORDER BY temp-B-tree entry when covered by first window sort key (#5249)
  • Restrict window-sort index suppression to the innermost pass (#5250)

Storage & Indexes

  • Schema-aware index manager — same-named indexes coexist across schemas (#5540, #5555)
  • Schema-aware spatial indexes — same-name cross-schema coexistence (#5558, #5576)
  • Schema-aware indexes + sqlite_temp_master (#5513, #5539)
  • HNSW threshold-triggered graph compaction to restore recall (#5461)
  • Vector index per-row maintenance — keep IVFFlat/HNSW in sync on non-compacting INSERT/UPDATE/DELETE (#5453)
  • Rebuild IVFFlat/HNSW vector indexes on table compaction (#5449)
  • Make full-tree index rebuild reversible mid-transaction (#5443)
  • Undo disk-backed (spilled) index mutations on ROLLBACK (#5433)
  • Scope spilled-index undo to statement savepoint (RAISE(ABORT)) (#5458)
  • Copy-on-write Operations snapshot for txn rollback (#5426)
  • Preserve original index-name case in SQL dump (#5579, #5590)
  • Quote identifiers in SQL-dump index DDL round-trip (#5567)
  • Index-ordered scans wrong after mid-statement tombstone delete (#5524, #5530)
  • Repopulate catalog index metadata on binary load (#5220)
  • Preserve original-case identifier echo across dump/reload (#5618, #5621)
  • Persist CREATE TRIGGER in SQL-dump format (#5086)

Planner & Optimizer

  • Select partial indexes via structural predicate implication (#5331)
  • Preserve NULL semantics in scalar-comparison decorrelation (#5281)
  • Implicit-outer-aggregate-collapse + correlated window subquery (#5134)
  • Scope-aware outer-aggregate-collapse for nested FROM subqueries (#5139)

Performance

  • Add HNSW recall@k benchmark to the suite for regression tracking (#5476)
  • Add explicit SIMD intrinsics for filtered SUM on f64 (#5042)
  • Incremental columnar maintenance for INSERT/UPDATE (#5043)
  • Extend vectorized execution to LEFT/RIGHT JOINs and ORDER BY (#5039)
  • Early bail-out checks in columnar JOIN dispatch (#5051)

Web Demo

  • Surface HNSW recall@k on the trends dashboard (#5481)
  • Restore expectedRows/expectedCount payloads for example validation (#5259)
  • Restore sqllogictest SQL payloads (#5257)
  • Pin web-demo as standalone pnpm workspace and resync lockfile (#5252)

Infrastructure

  • New vibesql-consensus workspace crate — published to crates.io alongside the rest of the workspace.
  • Consolidate rusqlite under [workspace.dependencies] (#5171, follow-up clean-up #5237)
  • Migrate pnpm overrides to pnpm-workspace.yaml (#5169)
  • Loom 0.10.0 installation (#5168)
  • Add VibeSQL-specific /loom:release skill — interactive release flow with awareness of the four version-bearing files and the dual crates.io / PyPI tag-triggered workflows.
  • Add scripts/version.sh for version sync + release tagging (#5356)
  • pyo3 0.28 → 0.29 (#5379), arrow 58.1 → 59.0 (#5242), rusqlite 0.39 → 0.40 (#5163), rstar 0.12 → 0.13 (#5162)
  • Routine dependabot grouped bumps across web-demo, benchmarks, and root.
Assets 2
Loading