From 48609a8127692243904b10dd7c86a4e0d2effb3c Mon Sep 17 00:00:00 2001
From: Brian Sam-Bodden <bsb@redis.com>
Date: Tue, 28 May 2024 17:54:35 -0700
Subject: [PATCH] Fix vector type fields should not be encoded as strings
 (#2772)

---
 dev_requirements.txt            |  1 +
 redis/_parsers/encoders.py      |  6 ++++-
 redis/commands/search/result.py | 17 +++++---------
 tests/test_search.py            | 39 +++++++++++++++++++++++++++++++++
 4 files changed, 50 insertions(+), 13 deletions(-)

diff --git a/dev_requirements.txt b/dev_requirements.txt
index 48ec278d83..2357115934 100644
--- a/dev_requirements.txt
+++ b/dev_requirements.txt
@@ -16,3 +16,4 @@ ujson>=4.2.0
 wheel>=0.30.0
 urllib3<2
 uvloop
+numpy>=1.24.4
diff --git a/redis/_parsers/encoders.py b/redis/_parsers/encoders.py
index 6fdf0ad882..818d244d9d 100644
--- a/redis/_parsers/encoders.py
+++ b/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 as e:
+                    # Return the bytes unmodified
+                    return value
         return value
diff --git a/redis/commands/search/result.py b/redis/commands/search/result.py
index 5b19e6faa4..29d63e86f7 100644
--- a/redis/commands/search/result.py
+++ b/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:
diff --git a/tests/test_search.py b/tests/test_search.py
index f19c193891..d58d3b6a2d 100644
--- a/tests/test_search.py
+++ b/tests/test_search.py
@@ -4,6 +4,8 @@
 import time
 from io import TextIOWrapper
 
+import numpy as np
+
 import pytest
 import redis
 import redis.commands.search
@@ -2282,3 +2284,40 @@ 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)
\ No newline at end of file