pgvector compatibility: vector(n) type, distance operators and INSERT support

.github/workflows/ci.yml

@@ -92,6 +92,9 @@ jobs:
           - name: functions-full
             flags: "--features datafusion-pg-functions/full"
             optional: false
+          - name: pgvector
+            flags: "--features datafusion-postgres/pgvector"
+            optional: false
           - name: postgis
             flags: "--features datafusion-postgres/postgis"
             optional: true
@@ -132,6 +135,24 @@ jobs:
     - uses: actions/checkout@v6
     - uses: insightsengineering/disk-space-reclaimer@v1
     - uses: cachix/install-nix-action@v31
+    - name: Initialize podman
+      run: nix develop --command podman info >/dev/null 2>&1 || true
+    - run: |
+        cd tests-integration
+        nix develop --command bash test.sh --skip-postgis --skip-pgvector
+
+  # pgvector has its own dedicated integration job so it can run (and fail)
+  # independently of the base suite. `--skip-postgis` builds with the pgvector
+  # feature and runs the full python suite including the pgvector tests.
+  integration-pgvector:
+    name: Integration tests (pgvector)
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    needs: [test]
+    steps:
+    - uses: actions/checkout@v6
+    - uses: insightsengineering/disk-space-reclaimer@v1
+    - uses: cachix/install-nix-action@v31
     - name: Initialize podman
       run: nix develop --command podman info >/dev/null 2>&1 || true
     - run: |
@@ -154,7 +175,7 @@ jobs:
       run: nix develop --command podman info >/dev/null 2>&1 || true
     - run: |
         cd tests-integration
-        nix develop --command bash test.sh
+        nix develop --command bash test.sh --skip-pgvector
 
   msrv:
     name: MSRV

.gitignore

@@ -2,6 +2,7 @@
 .direnv
 .envrc
 .vscode
+.idea
 .aider*
 /tests-integration/test_env
 

arrow-pg/Cargo.toml

@@ -16,6 +16,9 @@ rust-version.workspace = true
 default = ["arrow"]
 arrow = ["dep:arrow"]
 datafusion = ["dep:datafusion"]
+# pgvector compatibility: teach the type mapping/encoding layer about the
+# pgvector `vector` wire type. See datafusion-pg-catalog `pgvector` feature.
+pgvector = []
 postgis = ["postgres-types/with-geo-types-0_7", "dep:geoarrow", "dep:geoarrow-schema", "dep:postgis", "dep:geo-postgis", "pgwire/pg-type-postgis", "dep:geo-traits"]
 
 [dependencies]