On Wed, Sep 02, 2015 at 09:39:11AM +0800, Yang Hongyang wrote:
> On 09/01/2015 10:36 PM, Stefan Hajnoczi wrote:
> >On Tue, Sep 01, 2015 at 05:06:14PM +0800, Yang Hongyang wrote:
> >>diff --git a/qemu-options.hx b/qemu-options.hx
> >>index 77f5853..f1d42a1 100644
> >>--- a/qemu-options.hx
> >>+++ b/qemu-options.hx
> >>@@ -1991,6 +1991,27 @@ override the default configuration (@option{-net nic
> >>-net user}) which
> >> is activated if no @option{-net} options are provided.
> >> ETEXI
> >>
> >>+DEF("netfilter", HAS_ARG, QEMU_OPTION_netfilter,
> >>+ "-netfilter buffer,id=str,netdev=str[,chain=in|out|all,interval=t]\n"
> >>+ " buffer network packets on netdev. if interval
> >>provided, will release\n"
> >>+ " packets by interval. Interval scale: microsecond\n",
> >>QEMU_ARCH_ALL)
> >
> >Perhaps the -object option should be used:
> >
> > -object netfilter-buffer,id=str,netdev=str[,chain=in|out|all,interval=t]
> >
> >That is how IOThread and memory backends were recently added.
> >
> >They are QOM objects (see include/qom/object.h) and eliminate the need
> >to write boilerplate code that adds new command-line options and
> >instantiates objects.
>
> I thought -netfilter is more obvious for an object name, -object is kind of
> abstract name... but I'm not maintainer, if you think it's really need to
> change to -object, I can do that.
The advantage of QOM and -object is that it eliminates code for
command-line options, object instantiation, etc. They introduce a
single object model that all types within QEMU can use instead of
inventing their own.
Personally I'm not that involved in QAPI or command-line but I thought
-object was the new preferred way to do things.
This is where I guess Markus and Andreas might have opinions.
Stefan
