Re: [PATCH v2 0/5] Introduce Owned type and Ownable trait (was: "rust: page: Add support for vmalloc_to_page")

[Danilo Krummrich <dakr@kernel.org>]

On Wed, Oct 23, 2024 at 11:51:47AM +0200, Miguel Ojeda wrote:
> On Wed, Oct 23, 2024 at 10:03 AM Danilo Krummrich <dakr@kernel.org> wrote:
> >
> > I wonder if it would make sense to make `CLIPPY=1` the default and only disable
> > it by explicitly passing `CLIPPY=0`.
> 
> That is what I wanted, but when I asked long ago to the Clippy
> maintainers if using `clippy-driver` was guaranteed to not affect
> codegen, the answer was no: for instance, optimization may be affected
> (at least back then), and the maintainers said the intention is that
> is not to be used for normal compiling. So I sent a PR to document
> that. See:
> 
>     https://github.com/rust-lang/rust-clippy/issues/8035
>     https://github.com/rust-lang/rust-clippy/pull/8037

That's pretty unfortunate, I didn't know.

I think for the long term it'd be good to find a way though. Once more and more
subsystems / people start adding Rust code, I could imagine patches to start
slipping through and fixing things up after it's been discovered in -next would
be painful.

> 
> Similarly, Christian proposed running `rustfmtcheck` unconditionally
> on build and offering a way to turn it off instead. I think that would
> be ideal too, but it could potentially lead to problems too, so I am
> not sure either; see e.g.:
> 
>     https://lore.kernel.org/rust-for-linux/CANiq72==AkkqCDaZMENQRg8cf4zdeHpTHwdWS3sZiFWm0vyJUA@mail.gmail.com/

Yeah, that's a tricky one; if not enabled by default I'd be a bit worried about
the same thing to happen as mentioned above.

Additionally, for development trees where things slipped through it'd be
annoying when `rustfmt` changes more stuff than expected.

> 
> So I wonder if we should instead go with a "dev mode" like `D=1` that
> enables Clippy, `rustfmtcheck`, `-Dwarnings` (even if `WERROR=n` and
> applying to everything, not just kernel objects,), potentially
> `rustdoc`-related warnings too, and whatever other tooling/checks in
> the future (e.g. klint), and not just for Rust but potentially for C
> and other bits too (e.g. `W=1`, some important subset of Coccinelle
> scripts...).

I think that'd be great for short / mid term, it'd make it much easier to ensure
that all relevant checks were executed and hence make it less likely for things
slip through.

For the long term, I think it'd be great to keep looking for ways to always
enable the clippy and format checks. Or at least the clippy ones if we're too
concerned about `rustfmt` to break.

> 
> That way, "normal builds" (i.e. those done by users) stay as
> fast/clean/warning-free/bug-free/optimized as possible even across
> compiler versions, potential bugs in the toolchain, etc. And I imagine
> it would be easier for newcomers, too.
> 
> Opinions welcome! I am happy to prepare an RFC, since it seems a few
> people would like something like that.
> 
> Cheers,
> Miguel
>