vtables and procedural macros (was Re: [PATCH] rust: pl011: convert pl011_create to safe Rust)

[Paolo Bonzini]

On 2/10/25 10:59, Zhao Liu wrote:
> On Thu, Feb 06, 2025 at 12:15:14PM +0100, Paolo Bonzini wrote:
> > Date: Thu,  6 Feb 2025 12:15:14 +0100
> > From: Paolo Bonzini <pbonzini@redhat.com>
> > Subject: [PATCH] rust: pl011: convert pl011_create to safe Rust
> > X-Mailer: git-send-email 2.48.1
> >
> > Not a major change but, as a small but significant step in creating
> > qdev bindings, show how pl011_create can be written without "unsafe"
> > calls (apart from converting pointers to references).
> >
> > This also provides a starting point for creating Error** bindings.
> >
> > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
> > ---
> >   rust/hw/char/pl011/src/device.rs | 39 ++++++++++++++++----------------
> >   rust/qemu-api/src/sysbus.rs      | 34 +++++++++++++++++++++++++---
> >   2 files changed, 50 insertions(+), 23 deletions(-)
>
> ...
>
> > +    fn realize(&self) {
>
> What about renaming this as "realize_with_sysbus"?
>
> Elsewhere, the device's own realize method is often used to set
> DeviceImpl::REALIZE.

I *think* this is not a problem in practice because trait methods are 
public (if the trait is in scope) whereas the device's realize method if 
private.

I agree that the naming conflict is unfortunate though, if only because 
it can cause confusion.  I don't know if this can be solved by 
procedural macros; for example a #[vtable] attribute that changes

    #[qemu_api_macros::vtable]
    fn realize(...) {
    }

into

    const fn REALIZE: ... = Some({
        fn realize(...) {
        }
        realize
    }

This way, the REALIZE item would be included in the "impl DeviceImpl for 
PL011State" block instead of "impl PL011State".  It's always a fine line 
between procedural macros cleaning vs. messing things up, which is why 
until now I wanted to see what things look like with pure Rust code; but 
I guess now it's the time to evaluate this kind of thing.

Adding Junjie since he contributed the initial proc macro infrastructure 
and may have opinions on this.

Paolo