reduce string formatting overhead for the very likely case of str values in Py2

Cython/Utility/StringTools.c

@@ -815,9 +815,14 @@ static CYTHON_INLINE int __Pyx_PyByteArray_Append(PyObject* bytearray, int value
 
 //////////////////// PyObjectFormatSimple.proto ////////////////////
 
-#if PY_MAJOR_VERSION < 3 || !CYTHON_COMPILING_IN_CPYTHON
+#if !CYTHON_COMPILING_IN_CPYTHON
     #define __Pyx_PyObject_FormatSimple(s, f) ( \
-        likely(PyUnicode_CheckExact(s)) ? (Py_INCREF(s), s) : \
+        likely(PyUnicode_CheckExact(s) ? (Py_INCREF(s), s) : \
+        PyObject_Format(s, f))
+#elif PY_MAJOR_VERSION < 3
+    // Py2 can handle both unicode and str in PyUnicode_Join() which we (normally) call afterwards
+    #define __Pyx_PyObject_FormatSimple(s, f) ( \
+        likely(PyUnicode_CheckExact(s) | PyString_CheckExact(s)) ? (Py_INCREF(s), s) : \
         PyObject_Format(s, f))
 #else
     // Py3 nicely returns unicode strings from str() which makes this quite efficient for builtin types