DV> Hum, for normal code we usually rely on valgrind to find leaks in DV> the code, for example by instrumenting test program using the code DV> as part of soem regression tests, but with the way CIM providers DV> are used it looks harder to do this. Right, *much* harder. DV> Or would it be possible to design a minimal C(++) program loading DV> the /usr/lib*/cmpi/ shared libraries exercising them locally ? There is a *ton* of infrastructure that has to be in place to run the providers, which I'm not sure is worth implementing. DV> Maybe this exists already, an in general how can we design DV> regression tests for libvirt-cim sounds something worth spending DV> time on :-) For the most part, I think we've been quite good about this kind of thing. Keeping functions small and using CIMOM-managed memory where we can helps us avoid these types of issues. Maybe we could spend some time running a CIMOM under valgrind to see if we can get some suppression filters tweaked to make the output reasonable. -- Dan Smith IBM Linux Technology Center Open Hypervisor Team email: danms@us.ibm.com