Re: [PATCH v3 1/2] scripts: generate_rust_analyzer.py: add versioning infrastructure

[Tamir Duberstein]

On 2026-03-15 16:01 +0900, Jesung Yang wrote:
> On Tue Mar 10, 2026 at 3:34 AM KST, Tamir Duberstein wrote:
> > On Sun, 08 Mar 2026 08:30:34 +0900, Jesung Yang <y.j3ms.n@xxxxxxxxx> wrote:
> [...]
> >> @@ -335,12 +390,90 @@ def generate_crates(
> >>              append_crate(
> >>                  name,
> >>                  path,
> >> -                [core, kernel, pin_init],
> >> +                sysroot_deps(core) + [kernel, pin_init],
> >>                  cfg=generated_cfg,
> >> +                crate_attrs=["no_std"],
> >>              )
> >
> > Can you help me understand this addition? It's not mentioned except in the
> > cover letter as a diff from v2.
> 
> Assuming you're referring to `crate_attrs=["no_std"]`, this makes
> rust-analyzer treat crates in `driver/` and `samples/` as if
> `#![no_std]` were specified in their crate roots (they don't contain
> `#![no_std]` themselves).

Yes, that's what I was referring to. Still, it's not clear to me why
that is part of this patch. Is it intentional, or incidental?

> 
> (Actually, I'm uncertain if this is the part you wanted me to elaborate
> on. Please let me know if you need more context.)

Yes, sorry for the ambiguity.

> 
> >> +def generate_rust_project(
> >> +    version_info: RaVersionInfo,
> >> +    srctree: pathlib.Path,
> >> +    objtree: pathlib.Path,
> >> +    sysroot: pathlib.Path,
> >> +    sysroot_src: pathlib.Path,
> >> +    external_src: Optional[pathlib.Path],
> >> +    cfgs: List[str],
> >> +    core_edition: str,
> >> +) -> RustProject:
> >> +    assert len(BASELINES) == 1, "Exhaustiveness check: update if branches!"
> >> +
> >> +    ctx: RaVersionCtx
> >> +
> >
> > Could we make RaVersionInfo an enum? That would allow mypy to do this check
> > (using `match`) rather than relying on this.
> 
> I think you want something like the following?
> 
> @enum.unique
> class RaVersionInfo(enum.Enum):
>     V20240311 = (
>         datetime.strptime("2024-03-11", "%Y-%m-%d"),
>         ra_version=(0, 3, 1877),
>         rust_version=(1, 78, 0),
>     )
>     V20251222 = ( ... )
> 
>     def __init__(
>         self,
>         release_date: date,
>         ra_version: Version,
>         rust_version: Version,
>     ) -> "RaVersionInfo":
>         self.release_date = release_date
>         self.ra_version = ra_version
>         self.rust_version = rust_version
> 
>     @staticmethod
>     def default() -> "RaVersionInfo":
>         return RaVersionInfo.V20240311
> 
> At the call site, we can access each field via the dot operator. This
> not only removes `BASELINES` and `DEFAULT_BASELINE` but also ensures
> that we only match against valid variants, which helps reduce the chance
> of human error. But unfortunately, `match` was introduced in Python
> 3.10, whereas the kernel only requires Python 3.9 [1]. This means we
> cannot leverage mypy's exhaushtiveness check yet. Perhaps I could leave
> a TODO to remove the assertion and switch to `match` once Python 3.10 is
> adopted.

That sounds reasonable to me. You can also name the variants in a more
semantically meaningful way like DEFAULT, SUPPORTS_XXX, etc.

Finally, `__init__` should return None, not Self.

> 
> The rest of the feedback sounds reasonable to me.
> 
> [1] https://docs.kernel.org/process/changes.html#kernel-documentation
> 
> Best regards,
> Jesung
> 

Thanks for working on this!