Re: [RFC PATCH 00/12] QOM/QAPI integration part 1

[Kevin Wolf]

Am 14.12.2021 um 11:40 hat Peter Maydell geschrieben:
> On Tue, 14 Dec 2021 at 10:26, Kevin Wolf <kwolf@redhat.com> wrote:
> > Configuration is for creating objects, properties are for runtime after
> > the creation.
> 
> Well, yes and no. In a few places we have some properties which
> are morally speaking configuration stuff, but which are runtime
> settable. This happens because the property needs to be set after
> the device is realized but before the machine is run, and we
> don't have a concept of "settable only during the machine creation
> phase", only of "settable only before realize". (I can't find an
> example of this in the codebase offhand, but I definitely have one
> in a patchset I'm working on -- the code which needs to
> configure the property of the configured object is far removed
> in both location in the codebase and point at which it runs
> from the code which is doing the initialize-and-realize.)

Yes, that's fair, but for the infrastructure it doesn't matter much what
something is "morally speaking". These things use properties at
"runtime" (i.e. after realize) today and will keep using them until we
find a different solution. I have no intention to change anything about
it in the context of QAPIfication. The only liberty I'm taking is
removing setters that can't work after realize.

Kevin