[libvirt] [PATCH 0/8] Normalization API

[Cole Robinson]

On 09/18/2013 12:02 PM, Daniel P. Berrange wrote:
> On Wed, Sep 18, 2013 at 11:55:39AM -0400, Cole Robinson wrote:
>> On 09/18/2013 11:43 AM, Daniel P. Berrange wrote:
>>> On Wed, Sep 18, 2013 at 09:38:23AM -0600, Eric Blake wrote:
>>>> On 09/18/2013 09:32 AM, Cole Robinson wrote:
>>>>>
>>>>> Seems overkill IMO. Do we really need to add an API to facilitate the crappy
>>>>> interface that is virsh detach-device? Will any other application ever want to
>>>>> use this API?
>>>>
>>>> Yes, I can envision other use cases.  In fact, virt-manager is one of them:
>>>>
>>>> Oftentimes, we add features to XML, but you don't have an easy way to
>>>> probe if the feature is supported.  Rather than complicate the (already
>>>> large) capabilities xml to call out yet more features, it is rather
>>>> simple to write up XML that tries to use the feature, then run it
>>>> through the normalizing API, then look at the result.  If the feature is
>>>> still present in the output, then libvirt understands the feature (and
>>>> you can safely use it); if the feature got stripped as unrecognized,
>>>> then you can issue a much nicer error message to the user stating that
>>>> the libvirtd on the other end of the connection can't honor the user's
>>>> request.
>>>
>>> IMHO a better way to deal with that usage scenario is to add ability
>>> to validate XML via the RNG schema to the API.
>>>
>>> eg we should add a "validate" flag that can be passed to virDomainDefineXML
>>> / virDomainCreateXML that would enable RNG validation during parsing of
>>> those APis. And possibly even have a dedicated API to request validation
>>> of an XML doc in general (virt-xml-validate, but part of the API instead
>>> of local CLI tool)
>>>
>>
>> Yeah that could be useful. Though so would an API to just return the libvirt
>> domain rng and let apps cache it, or parse it and use that as a poor man's XML
>> capabilities and make virt-install --disk cache=? list all available cache
>> modes without having to hardcode them. Granted that won't be accurate across
>> hypervisors but its a start.
> 
> I would not want apps to be in the business of parsing RNG docs to
> extract data. We make no guarantees about the structure of the RNG
> documents being stable at all. We can arbitrarily re-structure them
> at any time to suit our current needs. The only thing RNG should be
> used for is validating a document.  This is the reason why I also
> think any tools which generate APIs based on RNG schemas are a bad
> idea.
> 

I was going on the assumption that there's a standard way using libxml to say
'parse this rng' and 'give me listed values for
/domain/devices/disk/driver at cache' but maybe that's not true.

- Cole