[Libvirt-cim] [PATCH] Change cu_compare_ref() to be a little lighter-weight

[Heidi Eckhart]

Kaitlin Rupert wrote:
> Dan Smith wrote:
>> # HG changeset patch
>> # User Dan Smith <danms at us.ibm.com>
>> # Date 1194453648 28800
>> # Node ID 926009fc62a0080bcf93aaddf098b074ce71a096
>> # Parent  404c4803b1b542676c4d283226a7cc3b7f3ab58d
>> Change cu_compare_ref() to be a little lighter-weight
>> Now, we just check the keys that were provided by the client (or the 
>> CIMOM
>> if that ever starts to happen).  We will have already checked the 
>> required
>> keys to actually do the lookup in the provider, which serves the 
>> "required"
>> key purpose.  Changes to libvirt-cim to follow shortly...
>>
>> Signed-off-by: Dan Smith <danms at us.ibm.com>
>>   
> I think this is a good compromise - looks good.  +1
>
I agree to Kaitlin, that this is a good compromise.

I had discussions with some CIM specialists and we concluded that 
currently a strict definition - on how the key values have to be set and 
checked - can not be found in one of the DMTF Spec's. The only clear 
statement is, that the sum of all key properties form the unique 
instance identifier. An agreement was also, that the provider should 
only accept these object paths (in regard to the number of key 
properties and their values), that the provider itself delivers via an 
enumeration. The current implementation includes the - more academical - 
possibility that a client generates an object path by itself, which is 
definitely a controversial and not recommended approach. This is not 
problematic as long as the client takes care of all key properties. Only 
the case that key properties are not handed over by the client breaks 
interoperability. But this could be fixed by a slight switch in the 
cu_compare_ref() function. Instead of getting the key count out of the 
reference it could be retrieved out of the instance :

diff -r 6f09834081cf instance_util.c
--- a/instance_util.c   Mon Nov 12 11:03:04 2007 -0800
+++ b/instance_util.c   Tue Nov 13 12:36:24 2007 +0100
@@ -118,8 +118,13 @@ const char *cu_compare_ref(const CMPIObj
         CMPIStatus s;
         int count;
         const char *prop = NULL;
+        CMPIObjectPath *op;
 
-        count = CMGetKeyCount(ref, &s);
+        op = CMGetObjectPath(inst, &s);
+        if ((op == NULL) || (s.rc != CMPI_RC_OK))
+                return NULL;
+
+        count = CMGetKeyCount(op, &s);
         if (s.rc != CMPI_RC_OK) {
                 CU_DEBUG("Unable to get key count");
                 return NULL;

The reason why the CIMOMs do not check the key properties (and accept 
not set key properties ) might be, that no CIM spec does definitely 
clarify the responsibilities here. So the CIMOM hands it over to the 
provider. Which makes sense as the provider is the one with the 
implementation logic for the key generation and always needs a way to 
identify its real instances out of a given object path.

-- 
Regards

Heidi Eckhart
Software Engineer
Linux Technology Center - Open Hypervisor

heidieck at linux.vnet.ibm.com

**************************************************
IBM Deutschland Entwicklung GmbH
Vorsitzender des Aufsichtsrats: Martin Jetter
Geschaeftsfuehrung: Herbert Kircher
Sitz der Gesellschaft: Boeblingen
Registergericht: Amtsgericht Stuttgart, HRB 243294