On Wed, Dec 24, 2025 at 02:47:20PM +0100, BALATON Zoltan wrote:
> On Wed, 24 Dec 2025, Akihiko Odaki wrote:
> > On 2025/12/24 6:49, BALATON Zoltan wrote:
> > > Our documentation says that memory regions are automatically freed
> > > when the owner dies and the reference counting to do this is also
> > > implemented. However this relies on the QOM free funtion that can only
> > > be set by creating objects with object_new but memory API only
> > > provides constructors that call object_initialize which clears the
> > > free function that prevents QOM to manage the memory region lifetime.
> > > Implement corresponding memory_region_new_* functions that do the same
> > > as the memory_region_init_* functions but create the memory region
> > > with object_new so the lifetime can be automatically managed by QOM as
> > > documented.
> > 
> > The documentation explains the existing functions so the discrepancy
> > between them you see should be fixed by updating them, not adding new
> > ones.
> 
> Do you mean replacing memory_region_init_* with these memory_region_new_*
> functions? The memory_region_init_* is still useful for embedded memory
> regions that are managed by some other way which is also mentioned in the
> documentation as an alternative so I think both of them are useful for
> different cases. If you mean we need to update docs to refer to
> memory_region_new instead of memory_region_init at some places then I think
> you're right, the docs may also need to be updated or clarified.

To me, it's less convincing to add new APIs without a solid user.

IMHO we can go either (1) leave patch 6 for later, making this series a
cleanup first, or, (2) add users for every new functions introduced, so at
least we know why each of the new functions are introduced and necessary.

Thanks,

-- 
Peter Xu


Reply via email to