On Fri, 2018-10-05 at 11:56 +0200, Pavel Hrdina wrote:
On Fri, Oct 05, 2018 at 11:38:14AM +0200, Andrea Bolognani wrote:
> Warnings printed on stderr -> users and developers will actually
> see them, be annoyed by them, eventually cave in and act upon them.
>
> Warnings written to a log -> nobody will notice them, until one day
> things suddenly stop working apparently out of the blue.
>
> We might pretend that's not the case, but really, it is.
You cannot expect that libvirt will duplicate all the deprecation
warnings from all hypervisors for all models or machine types or other
values that can be set to a lot of different attributes in our XML.
Once we do that for machine type there will be requests to do it for
other attributes as well. I think that this is a big NO and way out
of scope of libvirt.
Of course not, that would become very silly very fast - although
in many cases we already go out of our way to report a libvirt
error for configurations that would cause QEMU to ultimately bail
out anyway, so it wouldn't be entirely unprecedented either.
What I meant to say is that we should report deprecations of
*libvirt* features to the user directly instead of stuffing them
into a log file where they will possibly never be noticed until
it's way too late.
--
Andrea Bolognani / Red Hat / Virtualization