[libvirt] [PATCH 0/8] Add XML validation to the APIs

[Daniel P. Berrange]

On Wed, Nov 19, 2014 at 08:23:22AM +0100, Martin Kletzander wrote:
> On Tue, Nov 18, 2014 at 05:59:47PM +0000, Daniel P. Berrange wrote:
> >This proof of concept patch extends the virDomainDefineXML
> >and virDomainCreateXML APIs so that they can validate
> >the user supplied XML document against the RNG schemas.
> >
> >The virsh command will enable validation by default, it
> >must be turned off with --skip-validation if desired.
> >
> >This series is not complete
> >
> >- The network, interface, storage pool, etc APIs are
> >  not wired up to support validation.
> >- Only the QEMU virt driver is wired up to validate
> >- The virsh edit command is not wired up to validate
> >
> >It is enough to demonstrate it working with 'virsh define'
> >and the QEMU driver though.
> >
> >The biggest problem I see is the really awful error
> >messages we get back from libxml2 when validation
> >fails :-( They are essentially useless :-(
> >
> 
> This is one of the things why I'm not convinced this work is worth
> it.  It may be nice if we tell the user their XML is invalid instead
> of silently losing information.  But error message similar to "invalid
> element in interleave" doesn't help much when you are adding 100-line
> XML.  There are some better validators, but requiring those would be
> too cumbersome.

At least when using 'virsh edit' you would know what element you
just changed / added. So if you got get a generic 'validation failed'
error you have a pretty good idea of where in teh document you made
the mistake. So I think it'd be useful in that scenario. The error
reporting is more of a problem for the apps where they're passing in
a big XML document to define the guest and basically anything could
be wrong.

> Other than that, with this series applied, I get some errors with make
> check, particularly these:
> 
> qemuxml2xmltest
> openvzutilstest
> cpuset
> define-dev-segfault
> read-non-seekable
> read-bufsiz
> 
> Most of those are due to the test driver not supporting the new flag:
> 
> error: unsupported flags (0x1) in function testDomainDefineXMLFlags

Opps, yes, I should have said this proof of concept also hasn't
fixed the tests yet :-)

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 :|