From e82ca10aab5f6f82f57830da68f7ab51de684fc6 Mon Sep 17 00:00:00 2001
From: Andrew Chen <48723787+chuenchen309@users.noreply.github.com>
Date: Mon, 13 Jul 2026 08:57:42 +0800
Subject: [PATCH] fix: use uuid4 (not uuid1) for auto-generated ids in async
 constructors

PGVector's sync constructor path (`__from`) mints auto-generated document
ids with `uuid.uuid4()`, but the async path (`__afrom`, used by
`afrom_texts`/`afrom_embeddings`/`afrom_documents`) used `uuid.uuid1()`.

uuid1 encodes the host MAC address and a timestamp, so ids generated via
the async API (a) diverge from the random ids the sync API produces and
(b) leak the server's hardware address and record-creation time into the
stored/returned ids. `aadd_embeddings` itself already uses uuid4, so this
lone uuid1 was an oversight rather than intentional.

Align `__afrom` with the sync path and the rest of the module by using
uuid4.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
---
 langchain_postgres/vectorstores.py     |  2 +-
 tests/unit_tests/v1/test_afrom_uuid.py | 43 ++++++++++++++++++++++++++
 2 files changed, 44 insertions(+), 1 deletion(-)
 create mode 100644 tests/unit_tests/v1/test_afrom_uuid.py

diff --git a/langchain_postgres/vectorstores.py b/langchain_postgres/vectorstores.py
index a3743e4b..066ac234 100644
--- a/langchain_postgres/vectorstores.py
+++ b/langchain_postgres/vectorstores.py
@@ -722,7 +722,7 @@ async def __afrom(
         **kwargs: Any,
     ) -> PGVector:
         if ids is None:
-            ids = [str(uuid.uuid1()) for _ in texts]
+            ids = [str(uuid.uuid4()) for _ in texts]
 
         if not metadatas:
             metadatas = [{} for _ in texts]
diff --git a/tests/unit_tests/v1/test_afrom_uuid.py b/tests/unit_tests/v1/test_afrom_uuid.py
new file mode 100644
index 00000000..fdb31619
--- /dev/null
+++ b/tests/unit_tests/v1/test_afrom_uuid.py
@@ -0,0 +1,43 @@
+"""Auto-generated ids from the async constructors must be random UUIDv4.
+
+These are pure-Python tests: constructing ``PGVector`` with ``async_mode=True``
+defers ``__post_init__``, so no database connection is opened, and
+``aadd_embeddings`` is stubbed out to capture the ids the constructor generates.
+"""
+
+import uuid
+from typing import Any, List, Optional
+from unittest.mock import patch
+
+import pytest
+
+from langchain_postgres.vectorstores import PGVector
+from tests.unit_tests.fake_embeddings import FakeEmbeddings
+
+# A well-formed connection string. ``create_async_engine`` is lazy, so this does
+# not open a socket, and ``aadd_embeddings`` is patched before any I/O happens.
+FAKE_ASYNC_CONNECTION = "postgresql+psycopg://user:pw@localhost:5432/db"
+
+
+@pytest.mark.asyncio
+async def test_afrom_texts_generates_uuid4_ids() -> None:
+    """``afrom_texts`` should mint random (v4) ids, matching the sync path."""
+    captured: dict = {}
+
+    async def _capture(
+        *args: Any, ids: Optional[List[str]] = None, **kwargs: Any
+    ) -> None:
+        captured["ids"] = ids
+
+    with patch.object(PGVector, "aadd_embeddings", _capture):
+        await PGVector.afrom_texts(
+            texts=["foo", "bar", "baz"],
+            embedding=FakeEmbeddings(),
+            connection=FAKE_ASYNC_CONNECTION,
+        )
+
+    ids = captured["ids"]
+    assert ids is not None and len(ids) == 3
+    # uuid1 embeds the host MAC address and a timestamp; the sync `from_texts`
+    # uses uuid4, so the async path must too.
+    assert all(uuid.UUID(i).version == 4 for i in ids)