Fix vector type fields should not be encoded as strings (redis#2772)

Author: bsbodden

CHANGES

@@ -1,3 +1,4 @@
+    * Fix #2772, Fix vector type fields should not be encoded as strings
     * Update `ResponseT` type hint
     * Allow to control the minimum SSL version
     * Add an optional lock_name attribute to LockError.
@@ -59,7 +60,7 @@
     * Fix Sentinel.execute_command doesn't execute across the entire sentinel cluster bug (#2458)
     * Added a replacement for the default cluster node in the event of failure (#2463)
     * Fix for Unhandled exception related to self.host with unix socket (#2496)
-    * Improve error output for master discovery 
+    * Improve error output for master discovery
     * Make `ClusterCommandsProtocol` an actual Protocol
     * Add `sum` to DUPLICATE_POLICY documentation of `TS.CREATE`, `TS.ADD` and `TS.ALTER`
     * Prevent async ClusterPipeline instances from becoming "false-y" in case of empty command stack (#3061)

dev_requirements.txt

@@ -16,3 +16,4 @@ ujson>=4.2.0
 wheel>=0.30.0
 urllib3<2
 uvloop
+numpy>=1.24.4

redis/_parsers/encoders.py

@@ -40,5 +40,9 @@ def decode(self, value, force=False):
             if isinstance(value, memoryview):
                 value = value.tobytes()
             if isinstance(value, bytes):
-                value = value.decode(self.encoding, self.encoding_errors)
+                try:
+                    value = value.decode(self.encoding, self.encoding_errors)
+                except UnicodeDecodeError:
+                    # Return the bytes unmodified
+                    return value
         return value

redis/commands/search/result.py

@@ -39,18 +39,11 @@ def __init__(
 
             fields = {}
             if hascontent and res[i + fields_offset] is not None:
-                fields = (
-                    dict(
-                        dict(
-                            zip(
-                                map(to_string, res[i + fields_offset][::2]),
-                                map(to_string, res[i + fields_offset][1::2]),
-                            )
-                        )
-                    )
-                    if hascontent
-                    else {}
-                )
+                for j in range(0, len(res[i + fields_offset]), 2):
+                    key = to_string(res[i + fields_offset][j])
+                    value = res[i + fields_offset][j + 1]
+                    fields[key] = value
+
             try:
                 del fields["id"]
             except KeyError:

tests/test_search.py

@@ -4,6 +4,7 @@
 import time
 from io import TextIOWrapper
 
+import numpy as np
 import pytest
 import redis
 import redis.commands.search
@@ -2282,3 +2283,47 @@ def test_geoshape(client: redis.Redis):
     assert result.docs[0]["id"] == "small"
     result = client.ft().search(q2, query_params=qp2)
     assert len(result.docs) == 2
+
+
+@pytest.mark.redismod
+def test_vector_storage_and_retrieval(client):
+    # Constants
+    INDEX_NAME = "vector_index"
+    DOC_PREFIX = "doc:"
+    VECTOR_DIMENSIONS = 4
+    VECTOR_FIELD_NAME = "my_vector"
+
+    # Create index
+    client.ft(INDEX_NAME).create_index(
+        (
+            VectorField(
+                VECTOR_FIELD_NAME,
+                "FLAT",
+                {
+                    "TYPE": "FLOAT32",
+                    "DIM": VECTOR_DIMENSIONS,
+                    "DISTANCE_METRIC": "COSINE",
+                },
+            ),
+        ),
+        definition=IndexDefinition(prefix=[DOC_PREFIX], index_type=IndexType.HASH),
+    )
+
+    # Add a document with a vector value
+    vector_data = [0.1, 0.2, 0.3, 0.4]
+    client.hset(
+        f"{DOC_PREFIX}1",
+        mapping={VECTOR_FIELD_NAME: np.array(vector_data, dtype=np.float32).tobytes()},
+    )
+
+    # Perform a search to retrieve the document
+    query = Query("*").return_fields(VECTOR_FIELD_NAME).dialect(2)
+    res = client.ft(INDEX_NAME).search(query)
+
+    # Assert that the document is retrieved and the vector matches the original data
+    assert res.total == 1
+    assert res.docs[0].id == f"{DOC_PREFIX}1"
+    retrieved_vector_data = np.frombuffer(
+        res.docs[0].__dict__[VECTOR_FIELD_NAME], dtype=np.float32
+    )
+    assert np.allclose(retrieved_vector_data, vector_data)