[Python.NET] Several Python.Runtime problems

[Alexey Borzenkov]

Hello,

While using Python.NET in one of my projects I found some problems:

1. PythonException uses Runtime.PyErr_Fetch before calling
AcquireLock, which is wrong. Here's the patch:

--- src/runtime/pythonexception.cs	(revision 138)
+++ src/runtime/pythonexception.cs	(working copy)
@@ -27,11 +27,11 @@

     public PythonException() : base()
     {
+        IntPtr gs = PythonEngine.AcquireLock();
         Runtime.PyErr_Fetch(ref _pyType, ref _pyValue, ref _pyTB);
         Runtime.Incref(_pyType);
         Runtime.Incref(_pyValue);
         Runtime.Incref(_pyTB);
-        IntPtr gs = PythonEngine.AcquireLock();
         if ((_pyType != IntPtr.Zero) && (_pyValue != IntPtr.Zero))
         {
             string type = new PyObject(_pyType).GetAttr("__name__").ToString();

2. Every time .NET code throws an exception Python.NET leaks memory,
because exception object is not decref'ed after calling
PyErr_SetObject. In my case there were lots of exceptions, so problem
was very noticable over time. Here's the patch:

-- src/runtime/exceptions.cs	(revision 138)
+++ src/runtime/exceptions.cs	(working copy)
@@ -473,6 +473,7 @@
             IntPtr etype = Runtime.PyObject_GetAttrString(op, "__class__");
             Runtime.PyErr_SetObject(etype, op);
             Runtime.Decref(etype);
+            Runtime.Decref(op);
         }

         /// <summary>

Hope this helps others in their fights for thread safety and against
memory leaks. :)

Thanks,
Alexey.