Re: [PATCH 10/10] rust: bindings for MemoryRegionOps

[Paolo Bonzini <pbonzini@redhat.com>]

On 2/6/25 10:15, Zhao Liu wrote:
> Throughout the entire holiday, I couldn't think of a better way to
> express this. I find it particularly useful when wrapping multiple
> callbacks. In the future, I want to explore more use cases where this
> pattern can be applied.

Thanks very much.  Despite having written most of the bindings code, I 
don't want to give the impression that everything that I send to the 
mailing list is gold.  So it matters to me that you send serious 
reviews, and that you can learn something but also suggest 
clarifications and improvements.

> I find even if I drop the `inline` attribution, the test.rs can still be
> compiled (by `make check`), I think it's because test.rs hasn't involved
> memory related tests so that do_init_io isn't linked into test.rs.

Hmm, maybe it's only with some linkers.  I don't remember.

>> This is ugly because do_init_io exists *exactly* to extract the part
>> that is not generic. (See
>> https://users.rust-lang.org/t/soft-question-significantly-improve-rust-compile-time-via-minimizing-generics/103632/8
>> for an example of this; I think there's even a procedural macro crate
>> that does that for you, but I can't find it right now).
> 
> Thanks! I see. I agree to keep `inline` anyway.

In the meanwhile I found it, it's 
https://github.com/llogiq/momo/blob/master/wasm/src/lib.rs.  QEMU could 
add something like that macro to qemu_api_macros, so that

#[qemu_api_macros::upcast]
fn foo<T>(t: &T) where T: IsA<U> {
}

could be converted to

#[inline(always)]
fn foo<T>(&self) where T: IsA<U> {
     fn inner_foo(u: &U) {
     }
     inner_foo(self.upcast())
}

Paolo