Re: [RFC PATCH v1 13/17] vmstate: Drop VMS_ARRAY_OF_POINTER_AUTO_ALLOC

[Peter Xu <peterx@redhat.com>]

On Wed, Mar 25, 2026 at 03:29:01PM -0400, Peter Xu wrote:
> On Tue, Mar 24, 2026 at 04:43:28PM -0300, Fabiano Rosas wrote:
> > The semantics of the VMS_ARRAY_OF_POINTER_AUTO_ALLOC of allocating
> > memory dynamically for an array of pointers could be thought to be the
> > same as the currently non-existent combination of VMS_ARRAY_OF_POINTER
> > | VMS_ALLOC. The code is now able to handle such invalid combination,
> > so drop the new flag and enable the VMS_ARRAY_OF_POINTER | VMS_ALLOC
> > combo.
> 
> I was intentionally leaving VMS_ALLOC only for the top layer vmsd field,
> but then.. I think you're right, there's no way the driver code can offload
> the allocation of top layer vmsd field, while without offloading together
> the pointer allocations inside the array altogether when it's
> VMS_ARRAY_OF_POINTERS.
> 
> Let me see how I can incorporate this change, unless extremely awkward to
> rearrange, otherwise I'll make it not introduce this flag when repost.

Hmm, I forgot one thing, which is when the array doesn't need allocation
but the field does.  In that case VMS_ALLOC mustn't be set, but we still
need another flag to say "we need allocation for the elements in the array".

So maybe we still need a flag.

-- 
Peter Xu