Paolo Bonzini <pbonz...@redhat.com> writes: > On 20/07/21 03:00, Eric Blake wrote: >> Deceptively simple; all the work was in the previous patch writing up >> the forwarding visitor. I still wonder if Kevin's QAPI aliases will >> do this more gracefully, but if we're trying to justify this as a bug >> fix worthy of 6.1, this is certainly a smaller approach than Kevin's. >> Reviewed-by: Eric Blake<ebl...@redhat.com> > > As discussed on IRC, this is unrelated to QAPI aliases; QOM alias > properties typically target a property *on a different object*.
Yes, these are different beasts. A QAPI alias provides an alternate name for a member. The member may be nested. It's still within the same QAPI object. Can be useful for maintaining backwards compatibility, in particular for replacing (flat) QemuOpts by QAPI-based dotted keys. A QOM alias property is a proxy for a property of an *arbitrary* QOM object. Not limited to the alias's object and its sub-objects. Strictly more powerful. QOM alias properties are created at run time: creation requires the target object. QAPI aliases are completely defined at compile-time. > This is a regression, so it certainly has to be fixed in 6.1 one way > or the other. Understood.