Re: [Qemu-devel] [PATCH for-2.5 0/4] Expose ErrorClass through introspection

[Eric Blake]

On 11/12/2015 03:48 AM, Markus Armbruster wrote:
> Eric Blake <address@hidden> writes:
> 
>> I noticed that introspection was not documenting either
>> qmp_capabilities nor the ErrorClass enum.  I think this is worth
>> fixing for 2.5 when introspection is brand new, so that if we later
>> extend the ErrorClass enum or add future capability negotiation (and
>> in particular if such additions get backported in downstream builds),
>> a client will be able to use introspection to learn whether the new
>> features are supported, regardless of the qemu version.
>>
>> Note that this also adds qmp_capabilities to 'query-commands'.
>>
>> Yes, this is borderline, and you may decide that it doesn't deserve
>> to be called a bug and should wait for 2.6.

Or in other words, I posted this series precisely for this conversation.
 The mere fact that you have questions means that it is NOT appropriate
for 2.5 this late in the game.  But to answer you...

> 
> Two entirely separate issues: introspecting qmp_capabilities and
> introspecting error classes.  This reply is about the former.  I'll
> discuss the latter in a separate message.
> 
> Introspecting qmp_capabilities with query-qmp-schema is kind of
> academic, because by the time you can query-qmp-schema, you're already
> done using qmp_capabilities.  That's why docs/qmp-spec.txt nails down
> qmp_capabilities, unlike other commands.  Unfortunately, it's a bit
> screwy.  Let's have a closer look.
> 
> Right after connect, the server sends a greeting that includes a list of
> available capabilities (docs/qmp-spec.txt section 2.2 Server Greeting),
> currently none.
> 
> The only valid command then is qmp_capabilities (section 4. Capabilities
> Negotiation).  The command is meant to let clients enable capabilities
> from the greeting, but we've neglected to specify how.  Fortunately, QMP
> rejects all arguments, so we can still fix that by adding optional
> arguments, either a list of capabilities to enable, or a (boolean)
> argument for each capability.
> 
> My point is: unlike other commands, qmp_capabilities is baked into the
> protocol.  Introspection is useful to examine *variable* aspects.  With
> the hole in the protocol specification plugged, the only variable aspect
> of qmp_capabilities are the capabilities, and the practical way to
> introspect those is the server greeting.

Good point - if we ever add a capability, a client that knows how to
request the capability can easily learn whether the server will honor
the request by reading the server's advertisements.

> 
> Qapifying qmp_capabilities can make sense regardless.  But why does it
> need to be rushed into 2.5 then?

No need for the rush after all.  You are correct that qapi-fying
qmp_capabilities will NOT help the client learn anything it couldn't
have already learned from the server greeting.

And if we don't rush qmp_capabilities into 2.5 (patch 3), then we also
don't need to rush in a fix for explicitly empty types (patch 2) or a
way to detect empty types (patch 1).

As for patch 4 - we could still expose ErrorClass, via some means other
than qmp_capabilities, if desired; but that's a discussion for your
other mail.

-- 
Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature