v2.0.0

RegreSQL 2.0 is major shift. Compared to 1.0 it

• no longer tests just result and baseline
• it can detect plan and cost changes, as well as cardinality errors

Reproducible state - snapshots
Database state captured once, restored before each run, so everyone tests against identical data

• snapshot capture / build / restore / info
• snapshot build pipeline: schema + fixtures in one step, section control (pre-data/data/post-data), fixture discovery, --schema flag + auto-format
• snapshot tagging, history, and diff (--from/--to) across versions
• schema-hash validation before tests; migration-change detection
• auto-restore before test/update; skip restore if unchanged (mtime+size)
• restore_database option; disable triggers on build; post-restore VACUUM ANALYZE; --clean suggestion when the target DB isn't empty
• capture PostgreSQL server settings and planner-cost GUCs into the snapshot

Fixturize integration
https://github.com/boringSQL/fixturize

The harness layer

• Ability to run the same corpus against two PostgreSQL builds
• Testing against production's reality
• Gate on measured actuals, not estimates
• Plan-quality & policy kits
• Correctness oracles without a reference database

Break-in changes

• Removes per-test fixtures support completely

Full Changelog: v1.5.0...v2.0.0