On Thu, Nov 09, 2023 at 07:35:04PM +0100, Victor Toso wrote: > On Thu, Nov 09, 2023 at 09:29:28AM -0800, Andrea Bolognani wrote: > > Additionally, this would allow client code that *looks* at the > > union to keep working even if actual data is later added to the > > branch; client code that *creates* the union would need to be > > updated, of course, but that would be the case regardless. > > I think it is better to not have code that is working to keep > working in this case where Spice is implemented. > > Implementing Spice here would mean that a struct type > SetPasswordOptionsSpice was created but because the code handling > it before was using struct type Empty, it will not handle the new > struct, leading to possible runtime errors (e.g: not handling > username/password) > > A bool would be simpler, triggering compile time errors.
You've convinced me :) Let's leave it like this for now. Once we start seriously thinking about compatibility across versions then we might have to reconsider this, but it's going to be part of a much bigger, much more fun conversation ;) -- Andrea Bolognani / Red Hat / Virtualization
