fix: allow nan values in payload in local mode

qdrant_client/local/local_collection.py

@@ -76,31 +76,12 @@ def _to_jsonable_python(x: Any) -> Any:
         return ENCODERS_BY_TYPE[type(x)](x)
 
 
-def convert_nan_inf_to_null(obj: Any) -> Any:
-    if isinstance(obj, float) and (np.isnan(obj) or np.isinf(obj)):
-        return None
-
-    if isinstance(obj, dict):
-        return {k: convert_nan_inf_to_null(v) for k, v in obj.items()}
-
-    if isinstance(obj, str) or isinstance(obj, bytes) or isinstance(obj, range):
-        return obj
-
-    # pydantic converts iterables to lists
-    if isinstance(obj, Iterable):
-        return [convert_nan_inf_to_null(v) for v in obj]
-
-    return obj
-
-
 def to_jsonable_python(x: Any) -> Any:
-    # breaks congruence with remote if pydantic<2.7, since it does not convert nan/inf to null
-    x = convert_nan_inf_to_null(x)
     try:
-        json.dumps(x, allow_nan=False)
+        json.dumps(x, allow_nan=True)
         return x
     except Exception:
-        return json.loads(json.dumps(x, allow_nan=False, default=_to_jsonable_python))
+        return json.loads(json.dumps(x, allow_nan=True, default=_to_jsonable_python))
 
 
 class LocalCollection:

tests/congruence_tests/test_updates.py

@@ -258,75 +258,6 @@ def test_upload_collection_dict_np_arrays(local_client, remote_client):
     compare_collections(local_client, remote_client, UPLOAD_NUM_VECTORS)
 
 
-def test_upload_payload_contain_nan_values():
-    # usual case when payload is extracted from pandas dataframe
-    from pydantic.version import VERSION
-
-    major, minor, patch = map(int, VERSION.split("."))
-    if major < 2 or (major == 2 and minor < 7):
-        pytest.skip("Test requires pydantic>=2.7.0")
-
-    def recreate_collections():
-        local_client.recreate_collection(
-            collection_name=nans_collection,
-            vectors_config=models.VectorParams(size=vector_size, distance=models.Distance.DOT),
-        )
-        remote_client.recreate_collection(
-            collection_name=nans_collection,
-            vectors_config=models.VectorParams(size=vector_size, distance=models.Distance.DOT),
-        )
-
-    local_client = init_local()
-    remote_client = init_remote()
-
-    vector_size = 2
-    nans_collection = "nans_collection"
-
-    points = generate_points(
-        num_points=UPLOAD_NUM_VECTORS,
-        vector_sizes=2,
-        with_payload=False,
-    )
-    ids, vectors, payload = [], [], []
-    for i in range(len(points)):
-        points[i].payload = {"surprise": math.nan}
-
-    for point in points:
-        ids.append(point.id)
-        vectors.append(point.vector)
-        payload.append(point.payload)
-
-    points_batch = models.Batch(
-        ids=ids,
-        vectors=vectors,
-        payloads=payload,
-    )
-
-    recreate_collections()
-    local_client.upload_collection(nans_collection, vectors, payload, ids)
-    remote_client.upload_collection(nans_collection, vectors, payload, ids)
-    compare_collections(
-        local_client, remote_client, UPLOAD_NUM_VECTORS, collection_name=nans_collection
-    )
-
-    recreate_collections()
-    local_client.upload_points(nans_collection, points)
-    remote_client.upload_points(nans_collection, points)
-    compare_collections(
-        local_client, remote_client, UPLOAD_NUM_VECTORS, collection_name=nans_collection
-    )
-
-    recreate_collections()
-    local_client.upsert(nans_collection, points=points_batch)
-    remote_client.upsert(nans_collection, points=points_batch)
-    compare_collections(
-        local_client, remote_client, UPLOAD_NUM_VECTORS, collection_name=nans_collection
-    )
-
-    local_client.delete_collection(nans_collection)
-    remote_client.delete_collection(nans_collection)
-
-
 def test_upload_wrong_vectors():
     local_client = init_local()
     remote_client = init_remote()