diff -r dc721353b2e2 Doc/c-api/unicode.rst
--- a/Doc/c-api/unicode.rst	Sat Oct 06 23:55:33 2012 +0200
+++ b/Doc/c-api/unicode.rst	Sun Oct 07 23:01:43 2012 +0200
@@ -516,8 +516,7 @@ APIs:
    |                   |                     | :c:func:`PyObject_Repr`.       |
    +-------------------+---------------------+--------------------------------+
 
-   An unrecognized format character causes all the rest of the format string to be
-   copied as-is to the result string, and any extra arguments discarded.
+   Raise a :exc:`ValueError` if the format string is invalid.
 
    .. note::
 
@@ -530,6 +529,12 @@ APIs:
    .. versionchanged:: 3.3
       Support for ``"%li"``, ``"%lli"`` and ``"%zi"`` added.
 
+   .. versionchanged:: 3.4
+      Raise a :exc:`ValueError` if the format string is invalid, instead of
+      copying the rest of the format string as-is to the result string (and
+      discard any extra arguments).
+
+
 
 .. c:function:: PyObject* PyUnicode_FromFormatV(const char *format, va_list vargs)
 
diff -r dc721353b2e2 Lib/test/test_unicode.py
--- a/Lib/test/test_unicode.py	Sat Oct 06 23:55:33 2012 +0200
+++ b/Lib/test/test_unicode.py	Sun Oct 07 23:01:43 2012 +0200
@@ -1746,7 +1746,6 @@ class UnicodeTest(string_tests.CommonTes
         self.assertEqual(PyUnicode_FromFormat(b'%c', c_int(0x10ffff)), '\U0010ffff')
 
         # test "%"
-        self.assertEqual(PyUnicode_FromFormat(b'%'), '%')
         self.assertEqual(PyUnicode_FromFormat(b'%%'), '%')
         self.assertEqual(PyUnicode_FromFormat(b'%%s'), '%s')
         self.assertEqual(PyUnicode_FromFormat(b'[%%]'), '[%]')
@@ -1813,12 +1812,14 @@ class UnicodeTest(string_tests.CommonTes
         text = PyUnicode_FromFormat(b'repr=%V', None, b'abc\xff')
         self.assertEqual(text, 'repr=abc\ufffd')
 
-        # not supported: copy the raw format string. these tests are just here
-        # to check for crashs and should not be considered as specifications
-        self.assertEqual(PyUnicode_FromFormat(b'%1%s', b'abc'), '%s')
-        self.assertEqual(PyUnicode_FromFormat(b'%1abc'), '%1abc')
-        self.assertEqual(PyUnicode_FromFormat(b'%+i', c_int(10)), '%+i')
-        self.assertEqual(PyUnicode_FromFormat(b'%.%s', b'abc'), '%.%s')
+        # invalid format string
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%')
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%.s', b'abc')
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%.3', b'abc')
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%1%s', b'abc')
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%1abc')
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%+i', c_int(10))
+        self.assertRaises(ValueError, PyUnicode_FromFormat, b'%.%s', b'abc')
 
     # Test PyUnicode_AsWideChar()
     def test_aswidechar(self):
diff -r dc721353b2e2 Objects/unicodeobject.c
--- a/Objects/unicodeobject.c	Sat Oct 06 23:55:33 2012 +0200
+++ b/Objects/unicodeobject.c	Sun Oct 07 23:01:43 2012 +0200
@@ -2349,6 +2349,15 @@ unicode_fromformat_arg(_PyUnicodeWriter 
 
     p = f;
     f++;
+    if (*f == '%') {
+        if (_PyUnicodeWriter_Prepare(writer, 1, '%') == 1)
+            return NULL;
+        PyUnicode_WRITE(writer->kind, writer->data, writer->pos, '%');
+        writer->pos++;
+        f++;
+        return f;
+    }
+
     zeropad = 0;
     if (*f == '0') {
         zeropad = 1;
@@ -2369,6 +2378,8 @@ unicode_fromformat_arg(_PyUnicodeWriter 
     precision = 0;
     if (*f == '.') {
         f++;
+        if (!Py_ISDIGIT((unsigned)*f))
+            goto invalid_format;
         while (Py_ISDIGIT((unsigned)*f)) {
             if (precision > (INT_MAX - ((int)*f - '0')) / 10) {
                 PyErr_SetString(PyExc_ValueError,
@@ -2378,14 +2389,6 @@ unicode_fromformat_arg(_PyUnicodeWriter 
             precision = (precision*10) + (*f - '0');
             f++;
         }
-        if (*f == '%') {
-            /* "%.3%s" => f points to "3" */
-            f--;
-        }
-    }
-    if (*f == '\0') {
-        /* bogus format "%.123" => go backward, f points to "3" */
-        f--;
     }
 
     /* Handle %ld, %lu, %lld and %llu. */
@@ -2411,7 +2414,7 @@ unicode_fromformat_arg(_PyUnicodeWriter 
         ++f;
     }
 
-    if (f[1] == '\0')
+    if (f[0] != '\0' && f[1] == '\0')
         writer->overallocate = 0;
 
     switch (*f) {
@@ -2626,27 +2629,18 @@ unicode_fromformat_arg(_PyUnicodeWriter 
         break;
     }
 
-    case '%':
-        if (_PyUnicodeWriter_Prepare(writer, 1, '%') == 1)
-            return NULL;
-        PyUnicode_WRITE(writer->kind, writer->data, writer->pos, '%');
-        writer->pos++;
-        break;
-
     default:
-        /* if we stumble upon an unknown formatting code, copy the rest
-           of the format string to the output string. (we cannot just
-           skip the code, since there's no way to know what's in the
-           argument list) */
-        len = strlen(p);
-        if (_PyUnicodeWriter_WriteCstr(writer, p, len) == -1)
-            return NULL;
-        f = p+len;
-        return f;
+        goto invalid_format;
     }
 
     f++;
     return f;
+
+invalid_format:
+    PyErr_Format(PyExc_ValueError,
+        "PyUnicode_FromFormatV() got an invalid format string near \"%.10s\"",
+        p);
+    return NULL;
 }
 
 PyObject *