gh-158213: Don't return a str subclass from PyUnicodeWriter_Finish() (#158214)

Since GH-157861, every writer uses the read-only optimization of
_PyUnicodeWriter_WriteStr(), so the first write of a str subclass
instance into an empty writer kept that object as the buffer, and
PyUnicodeWriter_Finish() returned it.  io.StringIO.getvalue() then
returned the written object itself, and the next write re-read it
through its __str__() method, which changed the contents and could make
read() read past the end of the buffer.

Only use the read-only optimization for exact str objects.  A subclass
is copied into a new buffer, as before GH-157861.

Author: fedonman

Lib/test/test_capi/test_unicode.py

@@ -1910,6 +1910,18 @@ def test_create(self):
         self.assertGreater(writer.get_buffer()[0], len(s))
         self.assertEqual(writer.finish(), s)
 
+    def test_str_subclass(self):
+        # The read-only optimization must not return a str subclass
+        class MyStr(str):
+            def __str__(self):
+                return self
+
+        writer = self.create_writer(0)
+        writer.write_str(MyStr('abc'))
+        result = writer.finish()
+        self.assertEqual(result, 'abc')
+        self.assertIs(type(result), str)
+
     def test_repr_null(self):
         writer = self.create_writer(0)
         writer.write_utf8(b'var=', -1)

Lib/test/test_io/test_memoryio.py

@@ -1118,7 +1118,11 @@ def __str__(self):
         s = MyStr("correct")
         memio = self.ioclass()
         memio.write(s)
-        self.assertEqual(memio.getvalue(), "correct")
+        value = memio.getvalue()
+        self.assertEqual(value, "correct")
+        self.assertIs(type(value), str)
+        memio.write("!")
+        self.assertEqual(memio.getvalue(), "correct!")
 
         # Also test the fast path where pos == string_size (STATE_ACCUMULATING)
         memio2 = self.ioclass()

Objects/unicode_writer.c

@@ -313,7 +313,7 @@ _PyUnicodeWriter_WriteStr(_PyUnicodeWriter *writer, PyObject *str)
     Py_UCS4 maxchar = PyUnicode_MAX_CHAR_VALUE(str);
 
     if (maxchar > writer->maxchar || len > writer->size - writer->pos) {
-        if (writer->buffer == NULL) {
+        if (writer->buffer == NULL && PyUnicode_CheckExact(str)) {
             assert(_PyUnicode_CheckConsistency(str, 1));
             writer->readonly = 1;
             writer->buffer = Py_NewRef(str);