On Mon, Aug 22, 2011 at 01:52:07PM -0400, Laine Stump wrote: > On 08/22/2011 09:10 AM, Eric Blake wrote: > >On 08/22/2011 01:36 AM, Laine Stump wrote: > >>The problem I was *really* trying to point out is that of a keymap > >>attribute being given *at all* when type is something other than "spice" > >>or "vnc". The problem is that keymap is stored in a union, and the parts > >>of the union that are active when type != "(spice|vnc)" don't have any > >>keymap members. So even if keymap has something that would have been > >>valid when type="(spice|vnc)" (but not otherwise), it isn't flagged in > >>the parser (because the parser ignores keymap when it's not appropriate, > >>rather than logging an error) and the driver *can't* do anything about > >>it (because it has no way of knowing that a keymap was given - the > >>parser can't put keymap into the config object because there's no place > >>to put it). > > > >To some degree, rng validation can map unions, if the > >distinguishing choice between unions is an attribute also exposed > >in the xml. It is done via <choice> and <group>, but it does make > >the rng larger; it also helps to have more <define>s in order to > >avoid some repetition of subelements that are common between more > >than one choice. > > > >> > >>In particular, to solve > >>https://bugzilla.redhat.com/show_bug.cgi?id=638633 according to your > >>recommendation, I think what I need to do is: > >> > >>1) Remove the script attribute from the bridge & ethernet unions in > >>virDomainNetDef and place it in the common part of the struct. > >> > >>2) Change the parser to always populate script, no matter what is the > >>type of the virDomainNetDef. > > > >Not necessarily. We can instead rewrite the rng to forbid script > >attributes from all but bridge and ethernet attributes. However, > >there's still the issue that some, but not all, hypervisors can > >support scripts with a bridge, so there's still some validation > >that will have to be done in drivers to reject unsupported items > >even though the rng allows it. > > > That's all great, but the fact remains that we don't use the RNG > anywhere in the libvirt API, so it does us no good, for example, > when running "virsh edit mydomain". > > Until/unless we do that (validate input XML to the relevant RNG), > having RNGs that properly mirror unions in the config objects is > just preventing us from detecting improper usage. (and I'm doubtful > this validation is a good candidate for backporting) As I mentioned earlier in this thread, we can easily add a new flag to the XML related APIs to request validation against the RNG. virsh edit would set this flag and if it fails, would prompt the user whether they want to correct the XML, or force the change ignoring validation errors. Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :| -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list