On 03/20/2014 11:58 AM, Cole Robinson wrote: > Eric pointed this out on the last patch, but I pushed it before noticing > his message. > --- > libvirt-override.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/libvirt-override.c b/libvirt-override.c > index 71e241e..2532400 100644 > --- a/libvirt-override.c > +++ b/libvirt-override.c > @@ -2350,7 +2350,7 @@ done: > > error: > Py_XDECREF(rv); > - rv = VIR_PY_NONE; > + rv = NULL; > goto done; ACK. We probably have other code not quite doing the right thing; maybe it's time to audit the code base. The rule of thumb (which took me some time to learn) is: If libvirt failed, we have a thread-local libvirt error set. Return a non-NULL sentinel value (-1 or None, depending on whether the interface normally returns an integer or a python struct) so that the wrapper code will then raise a libvirt exception class object. If python failed, we either encountered OOM, or there was some other problem with the user's input (perhaps the user passed in a python object that can't be converted to the type we expect), and we don't have any thread-local libvirt error set. Return NULL so that python will immediately raise a python exception class object without even getting to our wrapper. Returning NULL when there is no thread-local python exception ready to raise, or returning non-NULL when there is no thread-local libvirt error to convert to a libvirt exception object, or returning the wrong type of sentinel, results in broken code.