connect: Do not deref encoding parameter (#1343)

The `encoding` parameter was being put into an Object wrapper which decremented its refcount
causing a segfault if used.  The value is not kept so the easiest solution is to simply borrow
the original pointer and do not wrap it.  I'm not sure why I did.

src/connection.cpp

@@ -59,7 +59,7 @@ static char* StrDup(const char* text) {
 }
 
 
-static bool Connect(PyObject* pConnectString, HDBC hdbc, long timeout, Object& encoding)
+static bool Connect(PyObject* pConnectString, HDBC hdbc, long timeout, PyObject* encoding)
 {
     assert(PyUnicode_Check(pConnectString));
 
@@ -170,7 +170,7 @@ static bool ApplyPreconnAttrs(HDBC hdbc, SQLINTEGER ikey, PyObject *value, char
 }
 
 PyObject* Connection_New(PyObject* pConnectString, bool fAutoCommit, long timeout, bool fReadOnly,
-                         PyObject* attrs_before, Object& encoding)
+                         PyObject* attrs_before, PyObject* encoding)
 {
     //
     // Allocate HDBC and connect
@@ -196,7 +196,7 @@ PyObject* Connection_New(PyObject* pConnectString, bool fAutoCommit, long timeou
         PyObject* value = 0;
 
         Object encodingholder;
-        char *strencoding = encoding.Get() ?
+        char *strencoding = (encoding != 0) ?
             (PyUnicode_Check(encoding) ? PyBytes_AsString(encodingholder = PyCodec_Encode(encoding, "utf-8", "strict")) :
              PyBytes_Check(encoding) ? PyBytes_AsString(encoding) : 0) : 0;
 

src/connection.h

@@ -104,7 +104,7 @@ struct Connection
  * exception is set and zero is returned.
  */
 PyObject* Connection_New(PyObject* pConnectString, bool fAutoCommit, long timeout, bool fReadOnly,
-                         PyObject* attrs_before, Object& encoding);
+                         PyObject* attrs_before, PyObject* encoding);
 
 /*
  * Used by the Cursor to implement commit and rollback.

src/pyodbcmodule.cpp

@@ -399,7 +399,7 @@ static PyObject* mod_connect(PyObject* self, PyObject* args, PyObject* kwargs)
     int fAutoCommit = 0;
     int fReadOnly = 0;
     long timeout = 0;
-    Object encoding;
+    PyObject* encoding = 0;
 
     Object attrs_before; // Optional connect attrs set before connecting
 

tests/postgresql_test.py

@@ -593,3 +593,30 @@ def convert(value):
     cursor.connection.add_output_converter(pyodbc.SQL_WVARCHAR, None)
     value = cursor.execute("select v from t1").fetchone()[0]
     assert value == '123.45'
+
+
+def test_refcount_encoding():
+    """
+    Ensure we handle the reference count to `encoding` properly.  In the past we freed a
+    borrowed reference.  This would cause a segfault.
+    """
+    # https://github.com/mkleehammer/pyodbc/issues/1343
+    import sys
+    encoding = 'utf-16le'
+    count_before = sys.getrefcount(encoding)
+
+    def _test():
+        # I've moved this into a function so the exception's stack trace will be freed under
+        # the covers when we leave the function.  Otherwise we'd have a 2nd reference to
+        # `encoding` in the stack trace of the exception.
+        try:
+            cnxn = pyodbc.connect(CNXNSTR, encoding=encoding)
+        except:
+            pass
+
+    for i in range(10):
+        _test()
+
+    count_after = sys.getrefcount(encoding)
+
+    assert count_after == count_before