"Benno Lossin" <lossin@kernel.org> writes:
> Rewrite the proc-macros of pin-init by using the `syn` crate for Rust
> syntax parsing. This series has been a long way coming. At the very
> start of pin-init, I initially implemented everything using syn, since
> parsing is so much easier with it. Now after several years it is finally
> time to remove the dreaded 1600 lines of declarative macros required to
> parse and expand the initializer syntax.
>
> The move to syn is not only a blessing for the maintenance work, but
> also improves the implementation of new features. This series includes
> many such improvements.
>
> * Patch 1, 2, 3 and 4 prepare for the rewrite. The first removes the
> superfluous `try_` variants of the initializer macros from pin-init.
> Note that the kernel defines its own, so no code changes in the kernel
> are required. The second patch allows using `::pin_init` in the
> pin-init crate. The third adds the syn dependency and cleans up some
> old workarounds and new clippy warnings. The fourth adds better error
> handling on top of `syn`.
> * Patch 5, 6, 7, and 9 rewrite the derive macros for Zeroable,
> `#[pinned_drop]` attribute macro, `#[pin_data]` attribute macro, and
> the initializer macros respectively using `syn`.
> * Patch 8 ensures soundness in the future by fixing generic bounds in
> generated code by `#[pin_data]`.
> * Patch 10 adds the `#[default_error(type)]` attribute to initializer
> macros allowing them to specify a default error that is used when no
> error is manually specified.
> * Patch 11 uses `#[default_error(type)]` in the definition of the
> kernel's `try_` variants of the initializer macros (which defaults to
> the kernel's `Error` type).
> * Patch 12 allows putting attributes on fields in initializer macros
> (for example `cfg` attributes).
> * Patch 13 adds `#[disable_initialized_field_access]` to support packed
> structs.
> * Patch 14 simplifies the code generated by the initializer macros.
> * Patch 15 adds Gary as a maintainer.
>
> In addition to the new features, using syn results in much cleaner error
> messages and more accurate span information. The code is much easier to
> read as well and hopefully easier to understand as well.
>
> As always, tests that ensure the correctness of the macro output are
> included and updated in the upstream pin-init repository. Take a look at
> the pull request on GitHub for the diff in the test output:
>
> https://github.com/Rust-for-Linux/pin-init/pull/89
>
> I would greatly appreciate Tested-by's from people actively using
> pin-init weather in the kernel or outside.
>
> One thing I forgot to mention last time: the commit adding syn support
> to pin-init is also used by Gary's series to rewrite the `macros` crate
> using `syn`:
>
> https://lore.kernel.org/all/20260107161729.3855851-1-gary@kernel.org
>
> My series can be found at:
>
> https://github.com/BennoLossin/linux/commits/pin-init/sync
>
> The rust-dev branch in the RfL repository will be updated to contain
> both patch series for ease of use & testing.
>
> Link to v1:
>
> https://lore.kernel.org/all/20260108135127.3153925-1-lossin@kernel.org
>
> Cheers,
> Benno
Tested with downstream rust null block driver on v6.19-rc5 [1]. No issues encountered, and
conditional fields now work, yay!
#[pin_data]
struct NullBlkDevice {
...
#[cfg(CONFIG_BLK_DEV_ZONED)]
#[pin]
zoned: zoned::ZoneOptions,
}
try_pin_init!(Self {
...
#[cfg(CONFIG_BLK_DEV_ZONED)]
zoned <- zoned::ZoneOptions::new(
zoned,
device_capacity_mib,
*block_size_bytes,
zone_size_mib,
zone_capacity_mib,
zone_nr_conv,
zone_max_open,
zone_max_active,
zone_append_max_sectors,
)?,
}),
Tested-by: Andreas Hindborg <a.hindborg@kernel.org>
Best regards,
Andreas Hindborg
[1] https://github.com/metaspace/linux/tree/rnull-v6.19-rc5