5 Commits

Include renames

Author	SHA1	Message	Date
codingetandClaude	d7e31dfce9	docs: codify the FK-safe stream ordering as a wire-format invariant ... The postgres importer inserts records sequentially with FK enforcement live, so it depends on records arriving in dependency order. That requirement was implicit in each backend's export table list; make it explicit and enforced. - Add EXPORT_KIND_ORDER (user, abode, resident, apikey, note) as a documented single source of truth, with the FK dependency chain spelled out on its doc comment, and note the ordering guarantee on the ExportEnvelope wire-format doc. Derive the isExportKind set from it. - Both backends' export() now iterate EXPORT_KIND_ORDER via a loader map (postgres omits note by leaving it out of the map), so emission order is tied to the constant and can't drift. - Tests: a change-detector on EXPORT_KIND_ORDER, plus assertions that both the sqlite and postgres (fake) exports emit record kinds grouped in FK-safe order (kind rank non-decreasing down the stream, after the leading meta). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>	2026-07-23 00:14:15 +00:00
codingetandClaude	3a56dcd9e5	feat: add export/import to the postgres backend; import auto-detects backend ... The export/import plan predated the postgres backend. Bring it up to parity: - PostgresInterface implements Exportable + Importable, mirroring the sqlite backend. Export is a signal-checked async generator (one query per table); import drives a transaction via `WrappedPool.multi`, which rolls back on any error/abort and commits only after the whole stream is consumed cleanly. - The `note` kind is skipped on postgres (its note CRUD is still unimplemented, so a pg database holds none) — a full dump from sqlite imports its user/abode/resident/apikey records and drops notes. - Unlike sqlite (PRAGMA foreign_keys=off), postgres keeps FK enforcement; the FK-safe insertion order keeps a full dump valid, and truly-dangling partial dumps will (correctly) fail. - abode-import now resolves the backend via getDbInterface instead of constructing SqliteInterface directly; isImportable keeps it from ever running over the remote (api) interface, which has no import. - Add isImportable(); make postgres selectClientApikeys' where optional. - Tests: exercise the real PostgresInterface export/import paths against an in-memory fake WrappedPgClient (no pg service in CI) — NDJSON shape, meta.source, filtering, note-skip, insert dispatch, and error/abort rollback. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>	2026-07-22 23:36:21 +00:00
codingetandCodex	31d4636dde	ci: add pull request quality gates ... Co-Authored-By: gpt-5.6-terra <noreply@openai.com>	2026-07-22 21:50:29 +00:00
codingetandClaude	b9fa79ff1f	fix: address review findings on postgres backend ... - PostgresMigrator: run JS migration parts on the transactional client (WrappedPgTx) instead of the pool, preserving migration atomicity - pool.ts: guard ROLLBACK so a failing rollback no longer masks the original error (also applied in the migrator loop) - pool.ts: share all/get/run between WrappedPool and WrappedPgTx via a common base class - sql.ts: document that toPositional precludes JSONB ?/?|/?& operators - PostgresInterface: implement deleteSession, required by BackendDbInterface since the logout-invalidation change on master Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>	2026-07-20 02:03:47 +00:00
codingetandClaude	4521274a27	feat: add PostgreSQL backend ... Mirrors the node:sqlite sub-backend structure with full migration support. Uses native pg types (UUID, JSONB, TIMESTAMPTZ) and $1/$2 parameterisation via internal ? placeholders converted at execution time. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>	2026-07-20 02:00:10 +00:00