Re: [libvirt] [Qemu-devel] [RFC PATCH 2/4] qmp/hmp: Add getfd_file monitor command

On Tue, 22 May 2012 18:34:19 -0400 Corey Bryant<coreyb(a)linux.vnet.ibm.com> wrote: > On 05/22/2012 04:26 PM, Luiz Capitulino wrote: >> On Tue, 22 May 2012 16:02:19 -0400 >> Corey Bryant<coreyb(a)linux.vnet.ibm.com> wrote: >> >>>> But there's a small problem. Today getfd commands are closely tied to the >>>> Monitor. In Anthony's development tree, the getfd commands are tied to the >>>> new QMP server's session support. >>>> >>>> Asking you to integrate the new QMP server only to have the getfd command >>>> returning a simple integer would be too much, but at the same time I think >>>> you'll have to at least to break it from the monitor. This means moving its >>>> data structure away from the Monitor object and probably reworking the >>>> internal API used to get fds (ie. monitor_get_fd()). >>>> >>>> Shouldn't be hard, but you should be careful not to break external users. >>>> >>> >>> Just to verify, are you talking about moving the "fds" off the Monitor >>> struct? --> QLIST_HEAD(,mon_fd_t) fds; >> >> Yes. >> >>> Was this already moved away from the Monitor struct in Anthony's >>> development tree? If not do you have a recommendation on where to move it? >> >> Yes, iirc it moved inside the new QMP server session support in Anthony's tree. >> >>> I think this would make more sense to me if I took a look at the getfd >>> code in Anthony's development tree. Is this the correct tree? I had >>> some issues cloning it. https://github.com/aliguori/qemu-next.git >> >> The 'development' tree I'm referring to is the old glib branch in >> git://repo.or.cz/qemu/aliguori.git. >> > > Hmm, it looks like fds is still on the Monitor struct in that branch. Oh, you're right. That code is unfinished. It seems that I kept the finished version in my mind.

Well, I don't think that moving the fd array to another object will buy us much, so you can keep it this way. Note that you still have to convert do_getfd() to the QAPI as pointed out by Stefan and that the monitor object (*mon pointer) won't be passed to it. You'll have to use cur_mon in qmp_getfd() (as Anthony's version does).