On 04/17/2012 11:49 AM, Eric Blake wrote:
On 04/17/2012 11:45 AM, Guannan Ren wrote:
The result is indeterminate for NULL argument to python functions as follows. It's better to return negative value in these situations.
PyObject_IsTrue will segfault if the argument is NULL PyFloat_AsDouble(NULL) is -1.000000 PyLong_AsUnsignedLongLong(NULL) is 0.000000 --- python/typewrappers.c | 26 +++++++++++++++++++++++--- 1 files changed, 23 insertions(+), 3 deletions(-)
ACK.
I spoke too soon. On looking at it more, I'm wondering if all callers do the right thing in this case. It would probably be better if we explicitly raised a python exception before returning -1, so that callers can blindly assume that a return of -1 means that the correct python error is already present. That is, instead of: if (!obj) return -1; should we instead be doing something like: if (!obj) { PyErr_SetString(PyExc_TypeError, "unexpected type"); return -1; } Or are we guaranteed that the only way obj is NULL is if the caller already encountered an error, so there is already a python error set? -- Eric Blake eblake@redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org