From f90c36a5dc6b09e48e2b2ca3a6326159496716d5 Mon Sep 17 00:00:00 2001
From: AryanRG <aryanrganjoo@gmail.com>
Date: Sat, 31 Oct 2026 00:25:39 +0530
Subject: [PATCH] Handle empty text arrays in normalize_bytes2str

---
 etils/enp/numpy_utils.py              |  3 +
 etils/enp/numpy_utils_strings_test.py | 81 +++++++++++++++++++++++++++
 2 files changed, 84 insertions(+)
 create mode 100644 etils/enp/numpy_utils_strings_test.py

diff --git a/etils/enp/numpy_utils.py b/etils/enp/numpy_utils.py
index ec2f6cab..94e48c7b 100644
--- a/etils/enp/numpy_utils.py
+++ b/etils/enp/numpy_utils.py
@@ -382,6 +382,9 @@ def normalize_bytes2str(x):
   if isinstance(x, bytes):
     return x.decode('utf8')
   elif is_array_str(x):
+    # `np.vectorize` cannot infer the output dtype without any elements.
+    if not x.size:
+      return x.astype(str)
     # Note: `np.char.decode` is likely faster but don't work on `object` nor
     # bytes arrays.
     return _to_str_array(x)
diff --git a/etils/enp/numpy_utils_strings_test.py b/etils/enp/numpy_utils_strings_test.py
new file mode 100644
index 00000000..1e764a4f
--- /dev/null
+++ b/etils/enp/numpy_utils_strings_test.py
@@ -0,0 +1,81 @@
+# Copyright 2026 The etils Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""String normalization tests without optional array frameworks."""
+
+from etils import enp
+import numpy as np
+import pytest
+
+
+@pytest.mark.parametrize('dtype', ['S8', 'U8', object])
+@pytest.mark.parametrize('shape', [(0,), (0, 3), (2, 0, 3)])
+def test_normalize_bytes2str_empty(dtype, shape):
+  array = np.empty(shape, dtype=dtype)
+
+  normalized = enp.normalize_bytes2str(array)
+
+  assert normalized.shape == shape
+  assert normalized.dtype.kind == 'U'
+  assert normalized.size == 0
+  assert array.dtype == np.dtype(dtype)
+
+
+def test_normalize_bytes2str_filtered_batch():
+  features = np.array([b'train', b'validation'], dtype=object)
+  keep = np.array([False, False])
+
+  normalized = enp.normalize_bytes2str(features[keep])
+
+  assert normalized.shape == (0,)
+  assert normalized.dtype.kind == 'U'
+  np.testing.assert_array_equal(features, [b'train', b'validation'])
+
+
+@pytest.mark.parametrize('dtype', ['S16', 'U16', object])
+def test_normalize_bytes2str_unicode(dtype):
+  text = [['a', 'caf\u00e9'], ['', '\u65e5\u672c\u8a9e']]
+  values = (
+      text
+      if dtype == 'U16'
+      else [[value.encode('utf8') for value in row] for row in text]
+  )
+  array = np.array(values, dtype=dtype)
+
+  normalized = enp.normalize_bytes2str(array)
+
+  assert normalized.shape == (2, 2)
+  assert normalized.dtype.kind == 'U'
+  np.testing.assert_array_equal(normalized, text)
+
+
+@pytest.mark.parametrize('dtype', ['S8', 'U8', object])
+def test_normalize_bytes2str_scalar_array(dtype):
+  value = 'caf\u00e9' if dtype == 'U8' else 'caf\u00e9'.encode('utf8')
+
+  normalized = enp.normalize_bytes2str(np.array(value, dtype=dtype))
+
+  assert normalized.shape == ()
+  assert normalized.item() == 'caf\u00e9'
+
+
+def test_normalize_bytes2str_numeric_empty():
+  array = np.empty((2, 0), dtype=np.float32)
+
+  assert enp.normalize_bytes2str(array) is array
+
+
+def test_normalize_bytes2str_invalid_utf8():
+  with pytest.raises(UnicodeDecodeError):
+    enp.normalize_bytes2str(np.array([b'\xff']))