gh-130567: fix strxfrm memory allocation

The posix specification does not define that wcsxfrm should return
needed buffer size, it just says:

If the value returned is n or more, the contents of the array pointed to
by ws1 are unspecified.

Therefore double the allocation when the original call has failed. It
might also make sense to increase the allocation repeatedly, but that
would make the code more complex.

Author: nijel

Lib/test/test_locale.py

@@ -371,6 +371,17 @@ def test_strcoll_with_diacritic(self):
     def test_strxfrm_with_diacritic(self):
         self.assertLess(locale.strxfrm('à'), locale.strxfrm('b'))
 
+    @unittest.skipIf(sys.platform.startswith('aix'),
+                     'bpo-29972: broken test on AIX')
+    @unittest.skipIf(
+        is_emscripten or is_wasi,
+        "musl libc issue on Emscripten/WASI, bpo-46390"
+    )
+    @unittest.skipIf(sys.platform.startswith("netbsd"),
+                     "gh-124108: NetBSD doesn't support UTF-8 for LC_COLLATE")
+    def test_strxfrm_non_latin_1(self):
+        self.assertLess(locale.strxfrm('s'), locale.strxfrm('š'))
+
 
 class NormalizeTest(unittest.TestCase):
     def check(self, localename, expected):

Modules/_localemodule.c

@@ -422,8 +422,14 @@ _locale_strxfrm_impl(PyObject *module, PyObject *str)
         goto exit;
     }
     if (n2 >= (size_t)n1) {
-        /* more space needed */
-        wchar_t * new_buf = PyMem_Realloc(buf, (n2+1)*sizeof(wchar_t));
+        /* more space needed, some implementations return needed size while
+           others just buffer length and it's up to the caller to figure
+           out needed buffer size */
+        size_t new_buf_size = n2 + 1;
+        if (n1 * 2 > new_buf_size) {
+            new_buf_size = n1 * 2;
+        }
+        wchar_t * new_buf = PyMem_Realloc(buf, new_buf_size * sizeof(wchar_t));
         if (!new_buf) {
             PyErr_NoMemory();
             goto exit;