On Wed, Apr 18, 2012 at 09:18:45AM -0600, Eric Blake wrote:
On 04/18/2012 05:14 AM, Hu Tao wrote:
--- examples/get_cpu_stats.ml | 2 +- libvirt/libvirt.ml | 2 +- libvirt/libvirt.mli | 2 +- libvirt/libvirt_c_oneoffs.c | 9 +++++---- 4 files changed, 8 insertions(+), 7 deletions(-)
+++ b/libvirt/libvirt_c_oneoffs.c @@ -532,20 +532,21 @@ extern int virDomainGetCPUStats (virDomainPtr domain, #endif
CAMLprim value -ocaml_libvirt_domain_get_cpu_stats (value domv, value nr_pcpusv) +ocaml_libvirt_domain_get_cpu_stats (value domv, value nr_pcpusv, value flagsv)
This adds flags support to the per-cpu half of the libvirt API, but you are still missing ocaml bindings for the portion of the libvirt API that accesses the domain total stats. Also, I never understood why the caller has to know how many cpus they are passing in advance - shouldn't the bindings be smart enough to return an appropriately sized array that covers all possible cpus without the caller having to pre-specify that sizing?
For comparison, in the python bindings, we expressed things as:
if (!PyArg_ParseTuple(args, (char *)"OOi:virDomainGetCPUStats", &pyobj_domain, &totalbool, &flags))
so that a user passes in the domain; a choice of whether they want total stats (pass true to get a 1-element array back, corresponding to the C code passing start_cpu of -1), or per-cpu stats (pass fals to get an n-element array back, with each element mapping to a cpu, and with the array sized according to all cpus available); and finally a flags parameter.
It's certainly a unusual, and I agree that we should go with the model used by the Python bindings. FWIW it's probably because 'nr_pcpus' happens to be available conveniently as a value in virt-top at the point when 'get_cpu_stats' is called. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org