On Fri, 2017-11-24 at 12:12 +0100, Jiri Denemark wrote:
> Any application that's been coded sensibly will deal with
the new
> values in a graceful manner, eg. displaying "unknown model" or
> something along the line rather than failing.
Showing an unknown device and failing is not very different in this
case. The application adds a *-serial device and gets something else
back. We would need to format the old names in the XML everytime and not
in migratable XML only. So this would just add new shorter aliases apps
could use when talking to new libvirt. But they wouldn't use them anyway
because they would lose compatibility with older libvirt.
Yeah, only parsing it and not formatting it would make no sense at
all, so it's either both or neither.
In other words, NACK to this patch.
Well, that's a bummer :(
Does it mean we need to drop patch 20/21 as well? With that patch
in we would have a similar behavior, eg. a <console> would be
created with target type "sclp" but it would be formatted back with
target type "serial", and a new <serial> element would be created
along with it. Or is it okay in that case because "serial" is a
known value?
--
Andrea Bolognani / Red Hat / Virtualization