<...snip...>
I have not been able to try this patch, it fails with this error :
There's a v2 : http://www.redhat.com/archives/libvir-list/2015-April/msg00503.html Although it appears that it too has a RNG issue according to what Jan just posted. John
error: internal error: Unable to parse RNG /test-libvirt/share/libvirt/schemas/domain.rng: Reference osexe has no matching definition Internal found no define for ref osexe
However, had some concerns purely by looking at this patch. This change is very x86-centric, it does not respect other architectures. I think the rationale for simplifying domaincommon.rng would have been to group all types that obey this pattern string:
<param name="pattern">[a-zA-Z0-9_\.\-]+</param>
However, this regex does not conform to machine types for _all_ architectures. As an example, see this : <define name="hvms390"> <group> <optional> <attribute name="arch"> <choice> <value>s390</value> <value>s390x</value> </choice> </attribute> </optional> <optional> <attribute name="machine"> <choice> <value>s390</value> <value>s390-virtio</value> <value>s390-ccw</value> <value>s390-ccw-virtio</value> </choice> </attribute> </optional> </group> </define>
The s390 arch only allows four machine names : "s390", "s390-virtio", "s390-ccw", "s390-ccw-virtio". With the patch you suggest, even a string such as "abcdefg" will become a legitimate machine type for s390x, which seems like an odd thing. Likewise, ppc64[le] architecture allows only strings such as pseries, pseries-2.1, pseries-2.2 .. This patch will allow any random machine name, which seems somewhat odd to me.
Regards,