On Sat, Apr 17, 2021 at 01:46:35AM +0200, Miguel Ojeda wrote:
> On Sat, Apr 17, 2021 at 12:04 AM Willy Tarreau <w@xxxxxx> wrote:
> >
> > But my point remains that the point of extreme care is at the interface
> > with the rest of the kernel because there is a change of semantics
> > there.
> >
> > Sure but as I said most often (due to API or ABI inheritance), both
> > are already exclusive and stored as ranges. Returning 1..4095 for
> > errno or a pointer including NULL for a success doesn't shock me at
> > all.
>
> At the point of the interface we definitely need to take care of
> converting properly, but for Rust-to-Rust code (i.e. the ones using
> `Result` etc.), that would not be a concern.
Sure.
> Just to ensure I understood your concern, for instance, in this case
> you mentioned:
>
> result.status = foo_alloc();
> if (!result.status) {
> result.error = -ENOMEM;
> return result;
> }
Yes I mentioned this when it was my understanding that the composite
result returned was made both of a pointer and an error code, but Connor
explained that it was in fact more of a selector and a union.
> Is your concern is that the caller would mix up the `status` with the
> `error`, basically bubbling up the `status` as an `int` and forgetting
> about the `error`, and then someone else later understanding that
> `int` as a non-error because it is non-negative?
My concern was to know what field to look at to reliably detect an error
from the C side after a sequence doing C -> Rust -> C when the inner C
code uses NULL to mark an error and the upper C code uses NULL as a valid
value and needs to look at an error code instead to rebuild a result. But
if it's more:
if (result.ok)
return result.pointer;
else
return (void *)-result.error;
then it shouldn't be an issue.
Willy
