Fix seg fault using Python 2 invalid utf-8 strings and wstring

Fixes seg fault when passing a Python string, containing invalid utf-8 content,
to a wstring or wchar * parameter.  A TypeError is thrown instead, eg:

  %include <std_wstring.i>
  void instring(const std::wstring& s);

  instring(b"h\xe9llooo") # Python

CHANGES.current

@@ -7,6 +7,15 @@ the issue number to the end of the URL: https://github.com/swig/swig/issues/
 Version 4.0.0 (in progress)
 ===========================
 
+2018-06-15: wsfulton
+            [Python] Fix seg fault using Python 2 when passing a Python string, containing
+            invalid utf-8 content, to a wstring or wchar * parameter. A TypeError is thrown instead, eg:
+
+              %include <std_wstring.i>
+              void instring(const std::wstring& s);
+
+              instring(b"h\xe9llooo") # Python
+
 2018-06-12: wsfulton
             [Python] The %pythonabc feature in pyabc.i now uses base classes
               collections.abc.MutableSequence

Doc/Manual/Python.html

@@ -6501,6 +6501,7 @@ void instring(const char *s) {
 </pre></div>
 
 <p>
+Note that "\xe9" is an invalid UTF-8 encoding, but "\xc3\xb6" is valid.
 When this method is called from Python 3, the return value is the following
 text string:
 </p>

Examples/test-suite/li_std_wstring.i

@@ -94,6 +94,14 @@ void test_throw() TESTCASE_THROW1(std::wstring){
   throw x;
 }
 
+const char * non_utf8_c_str() {
+  return "h\xe9llo";
+}
+
+size_t size_wstring(const std::wstring& s) {
+  return s.size();
+}
+
 #ifdef SWIGPYTHON_BUILTIN
 bool is_python_builtin() { return true; }
 #else

Examples/test-suite/python/li_std_wstring_runme.py

@@ -1,4 +1,5 @@
 import li_std_wstring
+import sys
 
 x = u"h"
 
@@ -81,3 +82,25 @@ if b.name != "hello":
 b.a = li_std_wstring.A("hello")
 if b.a != u"hello":
     raise RuntimeError("bad string mapping")
+
+# Byte strings only converted in Python 2
+if sys.version_info[0:2] < (3, 0):
+    x = b"hello there"
+    if li_std_wstring.test_value(x) != x:
+        raise RuntimeError("bad string mapping")
+
+# Invalid utf-8 in a byte string fails in all versions
+x = b"h\xe9llo"
+try:
+    li_std_wstring.test_value(x)
+    raise RuntimeError("TypeError not thrown")
+except TypeError:
+    pass
+
+# Check surrogateescape
+if sys.version_info[0:2] > (3, 1):
+    x = u"h\udce9llo"  # surrogate escaped representation of C char*: "h\xe9llo"
+    if li_std_wstring.non_utf8_c_str() != x:
+        raise RuntimeError("surrogateescape not working")
+    if li_std_wstring.size_wstring(x) != 5 and len(x) != 5:
+        raise RuntimeError("Unexpected length")

Lib/python/pywstrings.swg

@@ -18,16 +18,28 @@ SWIG_AsWCharPtrAndSize(PyObject *obj, wchar_t **cptr, size_t *psize, int *alloc)
   int isunicode = PyUnicode_Check(obj);
 %#if PY_VERSION_HEX < 0x03000000 && !defined(SWIG_PYTHON_STRICT_UNICODE_WCHAR)
   if (!isunicode && PyString_Check(obj)) {
-    obj = tmp = PyUnicode_FromObject(obj);
+    tmp = PyUnicode_FromObject(obj);
-    isunicode = 1;
+    if (tmp) {
+      isunicode = 1;
+      obj = tmp;
+    } else {
+      PyErr_Clear();
+      return SWIG_TypeError;
+    }
   }
 %#endif
   if (isunicode) {
     Py_ssize_t len = PyUnicode_GetSize(obj);
     if (cptr) {
+      Py_ssize_t length;
       *cptr = %new_array(len + 1, wchar_t);
-      PyUnicode_AsWideChar(SWIGPY_UNICODE_ARG(obj), *cptr, len);
+      length = PyUnicode_AsWideChar(SWIGPY_UNICODE_ARG(obj), *cptr, len);
-      (*cptr)[len] = 0;
+      if (length == -1) {
+        PyErr_Clear();
+        Py_XDECREF(tmp);
+        return SWIG_TypeError;
+      }
+      (*cptr)[length] = 0;
     }
     if (psize) *psize = (size_t) len + 1;
     if (alloc) *alloc = cptr ? SWIG_NEWOBJ : 0;