.github/workflows/tests-linux.yml (2)

222-224: Use standard POSTGRES_USER env variable

Official images expect POSTGRES_USER. POSTGRES_USERNAME is non‑standard and may be ignored, leading to surprises.

-          POSTGRES_USERNAME: postgres
+          POSTGRES_USER: postgres

---

218-229: Harden service healthcheck

Add a health start period to reduce flakiness on cold starts.

         options: >-
           --health-cmd pg_isready
+          --health-start-period 30s
           --health-interval 10s
           --health-timeout 5s
           --health-retries 5

src/driver/sap/SapDriver.ts (1)

669-675: Graceful downgrade maps vector types to varbinary on non‑Cloud HANA

Good fallback, but add an integration test to assert DDL generation and round‑trip behavior when isReleaseVersionOrGreater(...,"4.0") is false.

Would you like me to add a test that boots a simulated pre‑4.0 environment and verifies real_vector/half_vector normalize to varbinary?

test/functional/database-schema/vectors/postgres/vector-similarity.ts (2)

25-37: Unnecessary clear; DB is already reloaded per test

reloadTestingDatabases resets schema. The extra repository.clear() is redundant.

-        await postRepository.clear() // Clear existing data
+        // DB is reloaded before each test; explicit clear is unnecessary

---

47-50: Cast param to vector to avoid type inference quirks

Being explicit prevents “could not determine data type” edge cases.

-                    `SELECT id, embedding FROM "post" ORDER BY embedding <-> $1 LIMIT 2`,
+                    `SELECT id, embedding FROM "post" ORDER BY embedding <-> ($1)::vector LIMIT 2`,

Apply similarly to other raw queries using vector/halfvec params.

docs/docs/entity/1-entities.md (1)

183-243: Add explicit pgvector extension note; duplication checked

• Append under the PostgreSQL bullet in the Note block of docs/docs/entity/1-entities.md:

> **Note**: 
> - **PostgreSQL**: Vector columns require the `pgvector` extension to be installed.
+>   Enable it in your database:
+>   `CREATE EXTENSION IF NOT EXISTS vector;`

• Duplication check: only one "### Vector columns" header found at line 183 in docs/docs/entity/1-entities.md.

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

src/driver/postgres/PostgresDriver.ts (1)

756-764: Hydration: add halfvec and optional dimension check (use length as dimensions)

• halfvec deserialization is missing.
• Optional validation against declared length (dimensions) improves fail-fast behavior.

Apply:

-        } else if (columnMetadata.type === "vector") {
-            if (
-                typeof value === "string" &&
-                value.startsWith("[") &&
-                value.endsWith("]")
-            ) {
-                if (value === "[]") return []
-                return value.slice(1, -1).split(",").map(Number)
-            }
+        } else if (
+            columnMetadata.type === "vector" ||
+            columnMetadata.type === "halfvec"
+        ) {
+            if (
+                typeof value === "string" &&
+                value.startsWith("[") &&
+                value.endsWith("]")
+            ) {
+                const arr =
+                    value === "[]"
+                        ? []
+                        : value.slice(1, -1).split(",").map((v) => Number(v))
+                const expected =
+                    columnMetadata.length != null && columnMetadata.length !== ""
+                        ? parseInt(String(columnMetadata.length), 10)
+                        : undefined
+                if (expected != null && arr.length !== expected) {
+                    throw new TypeORMError(
+                        `Invalid ${columnMetadata.type} length for column "${columnMetadata.databaseName}": expected ${expected}, got ${arr.length}`,
+                    )
+                }
+                return arr
+            }

If throwing feels too strict for hydration, we can log or skip the check; DB constraints should already prevent mismatches.

src/driver/postgres/PostgresQueryRunner.ts (1)

3494-3504: Vector/Halfvec length parsing: consider array forms and minor regex tightening

Current regex won’t match array types like vector(3)[]. While length is later handled for arrays in the generic withLength block, you can make this block self-sufficient and avoid double work by also accepting an optional [].

Apply this diff if you prefer to capture arrays here too:

-                                ].match(/^(?:vector|halfvec)\((\d+)\)$/)
+                                ].match(/^(?:vector|halfvec)\((\d+)\)(?:\[\])?$/)

src/driver/postgres/PostgresDriver.ts (1)

496-507: Auto-installing pgvector: OK, but message can clarify halfvec

The warning mentions “vector column” only. Consider “vector/halfvec” to be explicit. Functionality is fine.

Apply:

-                    "At least one of the entities has a vector column, but the 'vector' extension (pgvector) cannot be installed automatically. Please install it manually using superuser rights",
+                    "At least one of the entities has a vector/halfvec column, but the 'vector' extension (pgvector) cannot be installed automatically. Please install it manually using superuser rights",

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro

docker-compose.yml (2)

52-52: Avoid personal registry; pin by digest and update comment context

• Prefer an org-owned image (or build in-repo) and pin by SHA256 digest for supply‑chain and reproducibility.
• The comments above still refer to the generic postgis image; update to reflect the custom pgvector image.

Apply this diff to refresh the comment:

-    # postgis is postgres + PostGIS (only). If you need additional extensions,
-    # it's probably time to create a purpose-built image.
+    # Custom image includes PostGIS and pgvector (v0.8.0).
+    # Using a purpose-built image to run vector/halfvec tests.

---

65-65: Same as pg14: org-owned image and digest pinning

Mirror the pg14 guidance here: host under the TypeORM org (or build locally) and pin by digest.

src/driver/postgres/PostgresDriver.ts (1)

756-767: Dimension and NaN checks on hydration for symmetry with serialization

Add the same finite/length validation when parsing strings.

Apply this diff:

             if (
                 typeof value === "string" &&
                 value.startsWith("[") &&
                 value.endsWith("]")
             ) {
-                if (value === "[]") return []
-                return value.slice(1, -1).split(",").map(Number)
+                if (value === "[]") return []
+                const arr = value
+                    .slice(1, -1)
+                    .split(",")
+                    .map((s) => Number(s))
+                if (arr.some((n) => !Number.isFinite(n))) {
+                    throw new TypeORMError(
+                        `Invalid ${columnMetadata.type} element on hydration: all elements must be finite numbers`,
+                    )
+                }
+                const expected =
+                    columnMetadata.length != null
+                        ? Number(columnMetadata.length)
+                        : undefined
+                if (
+                    expected != null &&
+                    Number.isFinite(expected) &&
+                    arr.length !== expected
+                ) {
+                    throw new TypeORMError(
+                        `Invalid ${columnMetadata.type} length on hydration: expected ${expected}, got ${arr.length}`,
+                    )
+                }
+                return arr
             }

docker-compose.yml (1)

55-56: Port collisions: postgres-14 and postgres-17 both expose 5432 on host

Running both containers together will fail. Use distinct host ports.

Apply this diff:

   postgres-17:
@@
     ports:
-      - "5432:5432"
+      - "5433:5432"

Also applies to: 68-69

src/driver/postgres/PostgresDriver.ts (1)

496-507: Surface extension install errors to aid debugging

Include the error message in the warning; today failures are silent aside from a generic note.

Apply this diff:

-        if (hasVectorColumns)
-            try {
+        if (hasVectorColumns)
+            try {
                 await this.executeQuery(
                     connection,
                     `CREATE EXTENSION IF NOT EXISTS "vector"`,
                 )
-            } catch (_) {
-                logger.log(
+            } catch (e: any) {
+                logger.log(
                     "warn",
-                    "At least one of the entities has a vector column, but the 'vector' extension (pgvector) cannot be installed automatically. Please install it manually using superuser rights",
+                    `At least one of the entities has a vector column, but the 'vector' extension (pgvector) cannot be installed automatically. Please install it manually using superuser rights. Error: ${e?.message ?? e}`,
                 )
             }

Configuration used: CodeRabbit UI

Review profile: CHILL

Plan: Pro