Re: [RFC 07/18] rust: move Cell vmstate impl

[Paolo Bonzini <pbonzini@redhat.com>]

On 8/26/25 21:06, Marc-André Lureau wrote:
> Hi
> 
> On Tue, Aug 26, 2025 at 10:28 PM Paolo Bonzini <pbonzini@redhat.com 
> <mailto:pbonzini@redhat.com>> wrote:
> 
>     On 8/26/25 16:04, marcandre.lureau@redhat.com
>     <mailto:marcandre.lureau@redhat.com> wrote:
>      > From: Marc-André Lureau <marcandre.lureau@redhat.com
>     <mailto:marcandre.lureau@redhat.com>>
>      >
>      > This will allow to split vmstate to a standalone crate next.
> 
>     Can you explain why this is needed?  Could "migration" depend on "bql"
>     (or even, "bql" could stay in util), and keep the implementation of
>     VMState for cells, just like you do for Opaque?
> 
> 
> vmstate doesn't require bql. Why should we enforce it at rust level?

In some sense it does: pre_load/post_load/pre_save/post_save run inside 
the BQL.  It doesn't require it at the language level, but it does at 
the conceptual level.

Anyhow the code is fine, I just wanted to understand if there was 
something else.

> If we merge bql in util, then sure we can make vmstate keep the VMState 
> impl for BqlCells. But why should we do that? To save one crate? I think 
> it will more future proof if we have a lower-level util crate without 
> bql, and higher-level crates that rely on it. Perhaps "bql" should be 
> renamed though (qemu-loop/iothread or something?)

I'm okay with keeping bql separate, no problem. event_loop is also okay 
as the name.

Paolo