[libvirt] [TCK] nwfilter tests and libvirt commit 4f209434

[Laine Stump]

On 03/27/2014 12:22 AM, Mike Latimer wrote:
> Hi,
>
> As I've been looking through libvirt-tck tests, I found that commit 4f209434 
> (in libvirt) changes a condition that the nwfilter/050-apply-verify-host.t 
> relies on.
>
> Specifically, the 050-apply-verify-host.t test creates a number of filters with 
> invalid values (such as dscp='64', and protocol='256'). Without commit 
> 4f209434, these invalid values are silently dropped off the end of the rule, as 
> in the following example:
>
> #virsh nwfilter-define  <xml with illegal value:>
> <filter name='tck-testcase'>
>   <uuid>5c6d49af-b071-6127-b4ec-6f8ed4b55335</uuid>
>   <rule action='accept' direction='in'>
>      <ip  srcipaddr='10.1.2.3' srcipmask='255.255.255.254'
>           dstipaddr='10.1.2.3' dstipmask='255.255.255.128'
>           protocol='255' dscp='64'
>       />
>   </rule>
>
> #virsh nwfilter-dumpxml tck-testcase
>   <rule action='accept' direction='in' priority='500'>
>     <ip srcipaddr='10.1.2.3' srcipmask='31' dstipaddr='10.1.2.3' 
> dstipmask='25' protocol='255'/>
>   </rule>
>
> With commit 4f209434 in place, the entire filter is rejected with the following 
> error:  'internal error: dscp has illegal value 64'. As the filter is not 
> created, the testing of that filter by the 050-apply-verify-host test fails.
>
> I agree that the change is the right thing to do, but I'm wondering how best 
> to handle the now failing tests. This test (050-apply-verify-host.t) runs 
> ~1200 subtests, and about 10% of those tests fail due to the change. I'm 
> thinking that the illegal options (and only those illegal options) should be 
> removed from the now broken tests, and a few new tests added to specifically 
> test for the failure to add the entire rule when illegal data is passed. 
> Thoughts?


I agree.


>
> BTW - Is anyone else running the full libvirt-tck suite against recent 
> versions of libvirt?


I had thought someone had setup a continuous run of libvirt-tck against
current upstream, but maybe that's only being done for autotest.


> There are quite a few issues (such as this one) which 
> should be easily seen...
>