numpy · charris · Feb 5, 2025 · Feb 4, 2025
diff --git a/numpy/_core/src/multiarray/stringdtype/dtype.c b/numpy/_core/src/multiarray/stringdtype/dtype.c
@@ -270,6 +270,15 @@ as_pystring(PyObject *scalar, int coerce)
                        "string coercion is disabled.");
        return NULL;
    }
+    else if (scalar_type == &PyBytes_Type) {
+        // assume UTF-8 encoding
+        char *buffer;
+        Py_ssize_t length;
+        if (PyBytes_AsStringAndSize(scalar, &buffer, &length) < 0) {
+            return NULL;
+        }
+        return PyUnicode_FromStringAndSize(buffer, length);
+    }
    else {
        // attempt to coerce to str
        scalar = PyObject_Str(scalar);

diff --git a/numpy/_core/tests/test_stringdtype.py b/numpy/_core/tests/test_stringdtype.py
@@ -190,10 +190,14 @@ def test_array_creation_utf8(dtype, data):
    ],
 )
 def test_scalars_string_conversion(data, dtype):
+    try:
+        str_vals = [str(d.decode('utf-8')) for d in data]
+    except AttributeError:
+        str_vals = [str(d) for d in data]
    if dtype.coerce:
        assert_array_equal(
            np.array(data, dtype=dtype),
-            np.array([str(d) for d in data], dtype=dtype),
+            np.array(str_vals, dtype=dtype),
        )
    else:
        with pytest.raises(ValueError):
@@ -284,6 +288,14 @@ def test_bytes_casts(self, dtype, strings):
            barr = np.array(utf8_bytes, dtype=bytes_dtype)
            assert_array_equal(barr, sarr.astype(bytes_dtype))
            assert_array_equal(barr.astype(dtype), sarr)
+            if dtype.coerce:
+                barr = np.array(utf8_bytes, dtype=dtype)
+                assert_array_equal(barr, sarr)
+                barr = np.array(utf8_bytes, dtype="O")
+                assert_array_equal(barr.astype(dtype), sarr)
+            else:
+                with pytest.raises(ValueError):
+                    np.array(utf8_bytes, dtype=dtype)
        except UnicodeEncodeError:
            with pytest.raises(UnicodeEncodeError):
                sarr.astype("S20")