[PATCH v13 2/5] rust: support formatting of foreign types
Tamir Duberstein
tamird at gmail.com
Thu Jul 3 13:55:52 UTC 2025
On Thu, Jul 3, 2025 at 5:32 AM Benno Lossin <lossin at kernel.org> wrote:
>
> On Tue Jul 1, 2025 at 6:49 PM CEST, Tamir Duberstein wrote:
> > Introduce a `fmt!` macro which wraps all arguments in
> > `kernel::fmt::Adapter` and a `kernel::fmt::Display` trait. This enables
> > formatting of foreign types (like `core::ffi::CStr`) that do not
> > implement `core::fmt::Display` due to concerns around lossy conversions which
> > do not apply in the kernel.
> >
> > Replace all direct calls to `format_args!` with `fmt!`.
> >
> > Replace all implementations of `core::fmt::Display` with implementations
> > of `kernel::fmt::Display`.
> >
> > Suggested-by: Alice Ryhl <aliceryhl at google.com>
> > Link: https://rust-for-linux.zulipchat.com/#narrow/channel/288089-General/topic/Custom.20formatting/with/516476467
> > Acked-by: Greg Kroah-Hartman <gregkh at linuxfoundation.org>
> > Reviewed-by: Alice Ryhl <aliceryhl at google.com>
> > Signed-off-by: Tamir Duberstein <tamird at gmail.com>
> > ---
> > drivers/block/rnull.rs | 2 +-
> > drivers/gpu/nova-core/gpu.rs | 4 +-
> > rust/kernel/block/mq.rs | 2 +-
> > rust/kernel/device.rs | 2 +-
> > rust/kernel/fmt.rs | 89 +++++++++++++++++++++++++++++++++++++++
> > rust/kernel/kunit.rs | 6 +--
> > rust/kernel/lib.rs | 1 +
> > rust/kernel/prelude.rs | 3 +-
> > rust/kernel/print.rs | 4 +-
> > rust/kernel/seq_file.rs | 2 +-
> > rust/kernel/str.rs | 22 ++++------
> > rust/macros/fmt.rs | 99 ++++++++++++++++++++++++++++++++++++++++++++
> > rust/macros/lib.rs | 19 +++++++++
> > rust/macros/quote.rs | 7 ++++
> > scripts/rustdoc_test_gen.rs | 2 +-
> > 15 files changed, 236 insertions(+), 28 deletions(-)
>
> This would be a lot easier to review if he proc-macro and the call
> replacement were different patches.
>
> Also the `kernel/fmt.rs` file should be a different commit.
Can you help me understand why? The changes you ask to be separated
would all be in different files, so why would separate commits make it
easier to review?
I prefer to keep things in one commit because the changes are highly
interdependent. The proc macro doesn't make sense without
kernel/fmt.rs and kernel/fmt.rs is useless without the proc macro.
>
> > diff --git a/rust/kernel/fmt.rs b/rust/kernel/fmt.rs
> > new file mode 100644
> > index 000000000000..348d16987de6
> > --- /dev/null
> > +++ b/rust/kernel/fmt.rs
> > @@ -0,0 +1,89 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +
> > +//! Formatting utilities.
> > +
> > +use core::fmt;
>
> I think we should pub export all types that we are still using from
> `core::fmt`. For example `Result`, `Formatter`, `Debug` etc.
>
> That way I can still use the same pattern of importing `fmt` and then
> writing
>
> impl fmt::Display for MyType {
> fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {}
> }
Great idea, done for the next spin. It would be nice to be able to
lint against references to `core::fmt` outside of kernel/fmt.rs.
> > +
> > +/// Internal adapter used to route allow implementations of formatting traits for foreign types.
> > +///
> > +/// It is inserted automatically by the [`fmt!`] macro and is not meant to be used directly.
> > +///
> > +/// [`fmt!`]: crate::prelude::fmt!
> > +#[doc(hidden)]
> > +pub struct Adapter<T>(pub T);
> > +
> > +macro_rules! impl_fmt_adapter_forward {
> > + ($($trait:ident),* $(,)?) => {
> > + $(
> > + impl<T: fmt::$trait> fmt::$trait for Adapter<T> {
> > + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
> > + let Self(t) = self;
> > + fmt::$trait::fmt(t, f)
> > + }
> > + }
> > + )*
> > + };
> > +}
> > +
> > +impl_fmt_adapter_forward!(Debug, LowerHex, UpperHex, Octal, Binary, Pointer, LowerExp, UpperExp);
> > +
> > +/// A copy of [`fmt::Display`] that allows us to implement it for foreign types.
> > +///
> > +/// Types should implement this trait rather than [`fmt::Display`]. Together with the [`Adapter`]
> > +/// type and [`fmt!`] macro, it allows for formatting foreign types (e.g. types from core) which do
> > +/// not implement [`fmt::Display`] directly.
> > +///
> > +/// [`fmt!`]: crate::prelude::fmt!
> > +pub trait Display {
> > + /// Same as [`fmt::Display::fmt`].
> > + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result;
> > +}
> > +
> > +impl<T: ?Sized + Display> Display for &T {
> > + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
> > + Display::fmt(*self, f)
> > + }
> > +}
> > +
> > +impl<T: ?Sized + Display> fmt::Display for Adapter<&T> {
> > + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
> > + let Self(t) = self;
> > + Display::fmt(t, f)
>
> Why not `Display::fmt(&self.0, f)`?
I like destructuring because it shows me that there's only one field.
With `self.0` I don't see that.
> > + }
> > +}
> > +
> > +macro_rules! impl_display_forward {
> > + ($(
> > + $( { $($generics:tt)* } )? $ty:ty $( { where $($where:tt)* } )?
> > + ),* $(,)?) => {
> > + $(
> > + impl$($($generics)*)? Display for $ty $(where $($where)*)? {
> > + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
> > + fmt::Display::fmt(self, f)
> > + }
> > + }
> > + )*
> > + };
> > +}
> > +
> > +impl_display_forward!(
> > + bool,
> > + char,
> > + core::panic::PanicInfo<'_>,
> > + fmt::Arguments<'_>,
> > + i128,
> > + i16,
> > + i32,
> > + i64,
> > + i8,
> > + isize,
> > + str,
> > + u128,
> > + u16,
> > + u32,
> > + u64,
> > + u8,
> > + usize,
> > + {<T: ?Sized>} crate::sync::Arc<T> {where crate::sync::Arc<T>: fmt::Display},
> > + {<T: ?Sized>} crate::sync::UniqueArc<T> {where crate::sync::UniqueArc<T>: fmt::Display},
> > +);
>
> > diff --git a/rust/macros/fmt.rs b/rust/macros/fmt.rs
> > new file mode 100644
> > index 000000000000..edc37c220a89
> > --- /dev/null
> > +++ b/rust/macros/fmt.rs
> > @@ -0,0 +1,99 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +
> > +use proc_macro::{Ident, TokenStream, TokenTree};
> > +use std::collections::BTreeSet;
> > +
> > +/// Please see [`crate::fmt`] for documentation.
> > +pub(crate) fn fmt(input: TokenStream) -> TokenStream {
> > + let mut input = input.into_iter();
> > +
> > + let first_opt = input.next();
> > + let first_owned_str;
> > + let mut names = BTreeSet::new();
> > + let first_lit = {
> > + let Some((mut first_str, first_lit)) = (match first_opt.as_ref() {
> > + Some(TokenTree::Literal(first_lit)) => {
> > + first_owned_str = first_lit.to_string();
> > + Some(first_owned_str.as_str()).and_then(|first| {
> > + let first = first.strip_prefix('"')?;
> > + let first = first.strip_suffix('"')?;
> > + Some((first, first_lit))
>
> You're only using first_lit to get the span later, so why not just get
> the span directly here?
Good point. I was probably using it for more stuff in an earlier iteration.
>
> > + })
> > + }
> > + _ => None,
> > + }) else {
> > + return first_opt.into_iter().chain(input).collect();
> > + };
> > + while let Some((_, rest)) = first_str.split_once('{') {
>
> Let's put a comment above this loop mentioning [1] and saying that it
> parses the identifiers from the format arguments.
>
> [1]: https://doc.rust-lang.org/std/fmt/index.html#syntax
👍
>
> > + first_str = rest;
> > + if let Some(rest) = first_str.strip_prefix('{') {
> > + first_str = rest;
> > + continue;
> > + }
> > + if let Some((name, rest)) = first_str.split_once('}') {
> > + first_str = rest;
> > + let name = name.split_once(':').map_or(name, |(name, _)| name);
> > + if !name.is_empty() && !name.chars().all(|c| c.is_ascii_digit()) {
> > + names.insert(name);
> > + }
> > + }
> > + }
> > + first_lit
> > + };
> > +
> > + let first_span = first_lit.span();
> > + let adapter = quote_spanned! {
> > + first_span => ::kernel::fmt::Adapter
> > + };
>
> I think we should follow the formatting convention from the quote crate:
>
> let adapter = quote_spanned!(first_span=> ::kernel::fmt::Adapter);
Sure.
>
> > +
> > + let mut args = TokenStream::from_iter(first_opt);
> > + {
> > + let mut flush = |args: &mut TokenStream, current: &mut TokenStream| {
>
> You don't need to pass `args` as a closure argument, since you always
> call it with `&mut args`.
This doesn't work because of the borrow checker. If I wrote what you
suggest, then `args` is mutably borrowed by the closure, which
prohibits the mutable borrow needed for the .extend() call here:
for tt in input {
match &tt {
TokenTree::Punct(p) if p.as_char() == ',' => {
flush(&mut args, &mut current);
&mut args
}
_ => &mut current,
}
.extend([tt]);
}
>
> > + let current = std::mem::take(current);
> > + if !current.is_empty() {
> > + let (lhs, rhs) = (|| {
> > + let mut current = current.into_iter();
> > + let mut acc = TokenStream::new();
> > + while let Some(tt) = current.next() {
> > + // Split on `=` only once to handle cases like `a = b = c`.
> > + if matches!(&tt, TokenTree::Punct(p) if p.as_char() == '=') {
> > + names.remove(acc.to_string().as_str());
> > + // Include the `=` itself to keep the handling below uniform.
> > + acc.extend([tt]);
> > + return (Some(acc), current.collect::<TokenStream>());
> > + }
> > + acc.extend([tt]);
> > + }
> > + (None, acc)
> > + })();
> > + args.extend(quote_spanned! {
> > + first_span => #lhs #adapter(&#rhs)
> > + });
> > + }
> > + };
> > +
> > + let mut current = TokenStream::new();
>
> Define this before the closure, then you don't need to pass it as an
> argument.
Same reason as above. Borrow checker says no.
>
> ---
> Cheers,
> Benno
>
> > + for tt in input {
> > + match &tt {
> > + TokenTree::Punct(p) if p.as_char() == ',' => {
> > + flush(&mut args, &mut current);
> > + &mut args
> > + }
> > + _ => &mut current,
> > + }
> > + .extend([tt]);
> > + }
> > + flush(&mut args, &mut current);
> > + }
> > +
> > + for name in names {
> > + let name = Ident::new(name, first_span);
> > + args.extend(quote_spanned! {
> > + first_span => , #name = #adapter(&#name)
> > + });
> > + }
> > +
> > + quote_spanned! {
> > + first_span => ::core::format_args!(#args)
> > + }
> > +}
More information about the dri-devel
mailing list