diff --git a/stringdtype/pyproject.toml b/stringdtype/pyproject.toml
index 3ad86297..1fcf0803 100644
--- a/stringdtype/pyproject.toml
+++ b/stringdtype/pyproject.toml
@@ -31,6 +31,6 @@ per-file-ignores = {"__init__.py" = ["F401"]}
 
 [tool.meson-python.args]
 dist = []
-setup = ["-Ddebug=true", "-Doptimization=0"]
+setup = ["-Ddebug=true", "-Doptimization=2"]
 compile = []
 install = []
diff --git a/stringdtype/stringdtype/scalar.py b/stringdtype/stringdtype/scalar.py
index e95aac35..db82debd 100644
--- a/stringdtype/stringdtype/scalar.py
+++ b/stringdtype/stringdtype/scalar.py
@@ -2,10 +2,4 @@
 
 
 class StringScalar(str):
-    def partition(self, sep):
-        ret = super().partition(sep)
-        return (str(ret[0]), str(ret[1]), str(ret[2]))
-
-    def rpartition(self, sep):
-        ret = super().rpartition(sep)
-        return (str(ret[0]), str(ret[1]), str(ret[2]))
+    pass
diff --git a/stringdtype/stringdtype/src/dtype.c b/stringdtype/stringdtype/src/dtype.c
index 37915133..c188d774 100644
--- a/stringdtype/stringdtype/src/dtype.c
+++ b/stringdtype/stringdtype/src/dtype.c
@@ -146,26 +146,35 @@ static PyObject *
 stringdtype_getitem(StringDTypeObject *NPY_UNUSED(descr), char **dataptr)
 {
     char *data;
+    size_t len;
 
     if (*dataptr == NULL) {
         data = "\0";
+        len = 0;
     }
     else {
         data = ((ss *)dataptr)->buf;
+        len = ((ss *)dataptr)->len;
     }
 
-    PyObject *val_obj = PyUnicode_FromString(data);
+    PyObject *val_obj = PyUnicode_FromStringAndSize(data, len);
 
     if (val_obj == NULL) {
         return NULL;
     }
 
-    PyObject *res = PyObject_CallFunctionObjArgs((PyObject *)StringScalar_Type,
-                                                 val_obj, NULL);
-
-    Py_DECREF(val_obj);
-
-    return res;
+    /*
+     * In principle we should return a StringScalar instance here, but
+     * creating a StringScalar via PyObject_CallFunctionObjArgs has
+     * approximately 4 times as much overhead than just returning a str
+     * here. This is due to Python overhead as well as copying the string
+     * buffer from val_obj to the StringScalar we'd like to return. In
+     * principle we could avoid this by making a C function like
+     * PyUnicode_FromStringAndSize that fills a StringScalar instead of a
+     * str. For now (4-11-23) we are punting on that with the expectation that
+     * eventually the scalar type for this dtype will be str.
+     */
+    return val_obj;
 }
 
 // PyArray_NonzeroFunc
diff --git a/stringdtype/stringdtype/src/main.c b/stringdtype/stringdtype/src/main.c
index 64797492..00f4d0e0 100644
--- a/stringdtype/stringdtype/src/main.c
+++ b/stringdtype/stringdtype/src/main.c
@@ -104,6 +104,7 @@ PyInit__main(void)
     if (mod == NULL) {
         goto error;
     }
+
     StringScalar_Type =
             (PyTypeObject *)PyObject_GetAttrString(mod, "StringScalar");
     Py_DECREF(mod);
@@ -116,7 +117,9 @@ PyInit__main(void)
         goto error;
     }
 
+    Py_INCREF((PyObject *)&StringDType);
     if (PyModule_AddObject(m, "StringDType", (PyObject *)&StringDType) < 0) {
+        Py_DECREF((PyObject *)&StringDType);
         goto error;
     }