Re: [Qemu-devel] [PATCH v2 000/124] VMState Simplification (Massive)

[Dr. David Alan Gilbert]

* Paolo Bonzini (address@hidden) wrote:
> Il 21/04/2014 13:34, Peter Maydell ha scritto:
> >Mostly just that I think that for vmstate definitions "this new field
> >was added in version X" is natural and normal, whereas other
> >test functions are odd and generally the exception. So a simple
> >way to indicate minimum version for fields seems useful to retain.
> >I don't mind if you want to unify the underlying implementation,
> >but I would prefer to retain the _V macros for vmstate definitions
> >to use (and it has the additional advantage of avoiding the need for
> >touching lots of devices...)
> 
> I agree.  In many cases, _TEST is a huge review warning sign that
> subsections should have been used instead.

I can see how the subsections should be used in some cases, but I've
come across at least one case where _TEST was used to avoid the need
for a version change.

Mst's 9e047b (hw/acpi/piix4.c) replaces an existing field, if a property
on the device is set, but if the property is as-before then the structure
stays exactly as it was.
I can see how that probably should have used a subsection for the new
version of the data, but I don't see how it could have otherwise kept
it's compatibility.

Dave
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK