fix: serialize extension arrays after provisioning

[jgpruitt]

Summary

• bind batch_upsert parallel arrays with postgres.js's built-in text[] serializer
• cast the parameterized arrays to uuid[], public.ltree[], tstzrange[], and the configured pgvector array type in SQL
• avoid stale custom-array serializer maps when createIndex() installs extensions after a pool connection starts
• add a fresh-template0, single-connection regression covering prepared statement reuse, temporal values, nulls, escaping, halfvec, and vector
• document the fix in the changelog

Root cause

postgres.js discovers custom array serializers when a connection starts. When createIndex() installs ltree and pgvector on that already-open connection, its type map remains stale. Bare JavaScript arrays then fall back to comma-separated string conversion for the newly resolved custom array OID and PostgreSQL rejects the value with 22P02.

prepare: false does not fix this on a genuinely fresh database; reconnecting after extension installation only appeared to do so.

Validation

• ./bun run check
• core database integration suite: 56/56 passed
• fresh-extension regression against PostgreSQL 18
• packed npm artifact reproduction with default postgres.js settings on a fresh database

[Copilot]

🟢 Approval recommended

The change is narrowly scoped, aligns with the batch_upsert routine’s parameter types, and is validated by a deterministic regression test covering the reported failure mode.

Pull request overview

This PR fixes a postgres.js type-discovery edge case where extension-installed array types (e.g., ltree[], vector[]/halfvec[]) aren’t recognized on an already-open connection, causing batch array parameters to serialize incorrectly. The solution forces batch arrays through postgres.js’s built-in text[] serializer and lets PostgreSQL cast to the routine’s target array types, plus adds a regression test to prove correctness on a single-connection pool with prepared statement reuse.

Changes:

• Bind batch_upsert parallel arrays via sql.array(..., textOID) and cast to uuid[], public.ltree[], tstzrange[], and public.vector[]/public.halfvec[] in SQL.
• Add a fresh-template0 regression test that deterministically reproduces the extension-installation-on-same-connection scenario and verifies escaping/null/temporal/vector behavior.
• Document the fix in the changelog.

File summaries

File	Description
packages/core/test/support/db.ts	Adds optional max pool size support to force a single-connection pool for deterministic regression coverage.
packages/core/test/fresh-extension-arrays.integration.test.ts	New integration regression ensuring extension array types serialize correctly even when installed after connection startup.
packages/core/src/write.ts	Uses sql.array with text[] + explicit casts to avoid stale per-connection custom array serializer maps.
CHANGELOG.md	Documents the fixed behavior for upsert/upsertMany on the same pool connection used by createIndex.

Review details

• Files reviewed: 4/4 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.