Re: [PATCH v3 03/19] docs/devel: document expectations for QAPI data modelling for QMP

[Daniel P . Berrangé]

On Mon, Oct 04, 2021 at 01:13:00PM +0100, Dr. David Alan Gilbert wrote:
> * Daniel P. Berrangé (berrange@redhat.com) wrote:
> > Traditionally we have required that newly added QMP commands will model
> > any returned data using fine grained QAPI types. This is good for
> > commands that are intended to be consumed by machines, where clear data
> > representation is very important. Commands that don't satisfy this have
> > generally been added to HMP only.
> > 
> > In effect the decision of whether to add a new command to QMP vs HMP has
> > been used as a proxy for the decision of whether the cost of designing a
> > fine grained QAPI type is justified by the potential benefits.
> > 
> > As a result the commands present in QMP and HMP are non-overlapping
> > sets, although HMP comamnds can be accessed indirectly via the QMP
> > command 'human-monitor-command'.
> > 
> > One of the downsides of 'human-monitor-command' is that the QEMU monitor
> > APIs remain tied into various internal parts of the QEMU code. For
> > example any exclusively HMP command will need to use 'monitor_printf'
> > to get data out. It would be desirable to be able to fully isolate the
> > monitor implementation from QEMU internals, however, this is only
> > possible if all commands are exclusively based on QAPI with direct
> > QMP exposure.
> > 
> > The way to achieve this desired end goal is to finese the requirements
> > for QMP command design. For cases where the output of a command is only
> > intended for human consumption, it is reasonable to want to simplify
> > the implementation by returning a plain string containing formatted
> > data instead of designing a fine grained QAPI data type. This can be
> > permitted if-and-only-if the command is exposed under the 'x-' name
> > prefix. This indicates that the command data format is liable to
> > future change and that it is not following QAPI design best practice.
> > 
> > The poster child example for this would be the 'info registers' HMP
> > command which returns printf formatted data representing CPU state.
> > This information varies enourmously across target architectures and
> > changes relatively frequently as new CPU features are implemented.
> > It is there as debugging data for human operators, and any machine
> > usage would treat it as an opaque blob. It is thus reasonable to
> > expose this in QMP as 'x-query-registers' returning a 'str' field.
> > 
> > Reviewed-by: Markus Armbruster <armbru@redhat.com>
> > Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
> > ---
> >  docs/devel/writing-monitor-commands.rst | 27 +++++++++++++++++++++++++
> >  1 file changed, 27 insertions(+)
> > 
> > diff --git a/docs/devel/writing-monitor-commands.rst 
> > b/docs/devel/writing-monitor-commands.rst
> > index a973c48f66..0f3b751dab 100644
> > --- a/docs/devel/writing-monitor-commands.rst
> > +++ b/docs/devel/writing-monitor-commands.rst
> > @@ -350,6 +350,33 @@ In this section we will focus on user defined types. 
> > Please, check the QAPI
> >  documentation for information about the other types.
> >  
> >  
> > +Modelling data in QAPI
> > +~~~~~~~~~~~~~~~~~~~~~~
> > +
> > +For a QMP command that to be considered stable and supported long term,
> > +there is a requirement returned data should be explicitly modelled
> > +using fine-grained QAPI types. As a general guide, a caller of the QMP
> > +command should never need to parse individual returned data fields. If
> > +a field appears to need parsing, then it should be split into separate
> > +fields corresponding to each distinct data item. This should be the
> > +common case for any new QMP command that is intended to be used by
> > +machines, as opposed to exclusively human operators.
> > +
> > +Some QMP commands, however, are only intended as ad hoc debugging aids
> > +for human operators. While they may return large amounts of formatted
> > +data, it is not expected that machines will need to parse the result.
> > +The overhead of defining a fine grained QAPI type for the data may not
> > +be justified by the potential benefit. In such cases, it is permitted
> > +to have a command return a simple string that contains formatted data,
> > +however, it is mandatory for the command to use the 'x-' name prefix.
> > +This indicates that the command is not guaranteed to be long term
> > +stable / liable to change in future and is not following QAPI design
> > +best practices. An example where this approach is taken is the QMP
> > +command "x-query-registers". This returns a formatted dump of the
> > +architecture specific CPU state. The way the data is formatted varies
> > +across QEMU targets, is liable to change over time, and is only
> > +intended to be consumed as an opaque string by machines.
> > +
> 
> Are they required to do it this way - or are they allowed to define x-
> qapi types and do the formatting in the HMP code?
> For example, a lot of the info commands produce lists of data,
> you can imagine some of them could add x- types for each list entry.

If it makes sense for the corresponding QMP command to return more
explicitly structured data than HumanReadableText, but still using an
x- prefixed data type, then I'd say that is fine. That is an
incrementally better step towards the ideal of a fully stable QMP
command + data type. It'll be a judgement call by whomever designs
it as to how much time to invest in the QAPI side of things, depending
on how likely it is that machines want to interpret the strucutred
data.

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|