Re: [PATCH 5/5] rust: error: Add from_kernel_result!() macro

[Boqun Feng]

On Sun, Feb 26, 2023 at 01:36:06PM +0000, Gary Guo wrote:
> On Sat, 25 Feb 2023 18:22:11 -0800
> Boqun Feng <boqun.feng@xxxxxxxxx> wrote:
> 
> > On Sat, Feb 25, 2023 at 10:23:40PM +0000, Gary Guo wrote:
> > > On Fri, 24 Feb 2023 15:56:05 -0800
> > > Boqun Feng <boqun.feng@xxxxxxxxx> wrote:
> > >   
> > > > On Fri, Feb 24, 2023 at 05:50:23PM +0900, Asahi Lina wrote:  
> > > > > From: Wedson Almeida Filho <wedsonaf@xxxxxxxxx>
> > > > > 
> > > > > Add a helper macro to easily return C result codes from a Rust function
> > > > > that calls functions which return a Result<T>.
> > > > > 
> > > > > Lina: Imported from rust-for-linux/rust, originally developed by Wedson
> > > > > as part of file_operations.rs. Added the allow() flags since there is no
> > > > > user in the kernel crate yet and fixed a typo in a comment.
> > > > > 
> > > > > Co-developed-by: Fox Chen <foxhlchen@xxxxxxxxx>
> > > > > Signed-off-by: Fox Chen <foxhlchen@xxxxxxxxx>
> > > > > Co-developed-by: Miguel Ojeda <ojeda@xxxxxxxxxx>
> > > > > Signed-off-by: Miguel Ojeda <ojeda@xxxxxxxxxx>
> > > > > Signed-off-by: Wedson Almeida Filho <wedsonaf@xxxxxxxxx>
> > > > > Signed-off-by: Asahi Lina <lina@xxxxxxxxxxxxx>
> > > > > ---
> > > > >  rust/kernel/error.rs | 52 ++++++++++++++++++++++++++++++++++++++++++++++++++++
> > > > >  1 file changed, 52 insertions(+)
> > > > > 
> > > > > diff --git a/rust/kernel/error.rs b/rust/kernel/error.rs
> > > > > index cf3d089477d2..8a9222595cd1 100644
> > > > > --- a/rust/kernel/error.rs
> > > > > +++ b/rust/kernel/error.rs
> > > > > @@ -226,3 +226,55 @@ pub(crate) fn from_kernel_err_ptr<T>(ptr: *mut T) -> Result<*mut T> {
> > > > >      }
> > > > >      Ok(ptr)
> > > > >  }
> > > > > +
> > > > > +// TODO: Remove `dead_code` marker once an in-kernel client is available.
> > > > > +#[allow(dead_code)]
> > > > > +pub(crate) fn from_kernel_result_helper<T>(r: Result<T>) -> T
> > > > > +where
> > > > > +    T: From<i16>,
> > > > > +{
> > > > > +    match r {
> > > > > +        Ok(v) => v,
> > > > > +        // NO-OVERFLOW: negative `errno`s are no smaller than `-bindings::MAX_ERRNO`,
> > > > > +        // `-bindings::MAX_ERRNO` fits in an `i16` as per invariant above,
> > > > > +        // therefore a negative `errno` always fits in an `i16` and will not overflow.
> > > > > +        Err(e) => T::from(e.to_kernel_errno() as i16),
> > > > > +    }
> > > > > +}
> > > > > +
> > > > > +/// Transforms a [`crate::error::Result<T>`] to a kernel C integer result.
> > > > > +///
> > > > > +/// This is useful when calling Rust functions that return [`crate::error::Result<T>`]
> > > > > +/// from inside `extern "C"` functions that need to return an integer
> > > > > +/// error result.
> > > > > +///
> > > > > +/// `T` should be convertible from an `i16` via `From<i16>`.
> > > > > +///
> > > > > +/// # Examples
> > > > > +///
> > > > > +/// ```ignore
> > > > > +/// # use kernel::from_kernel_result;
> > > > > +/// # use kernel::bindings;
> > > > > +/// unsafe extern "C" fn probe_callback(
> > > > > +///     pdev: *mut bindings::platform_device,
> > > > > +/// ) -> core::ffi::c_int {
> > > > > +///     from_kernel_result! {
> > > > > +///         let ptr = devm_alloc(pdev)?;
> > > > > +///         bindings::platform_set_drvdata(pdev, ptr);
> > > > > +///         Ok(0)
> > > > > +///     }
> > > > > +/// }
> > > > > +/// ```
> > > > > +// TODO: Remove `unused_macros` marker once an in-kernel client is available.
> > > > > +#[allow(unused_macros)]
> > > > > +macro_rules! from_kernel_result {    
> > > > 
> > > > This actually doesn't need to be a macro, right? The following function
> > > > version:
> > > > 
> > > > 	pub fn from_kernel_result<T, F>(f: F) -> T
> > > > 	where
> > > > 	    T: From<i16>,
> > > > 	    F: FnOnce() -> Result<T>;
> > > > 
> > > > is not bad, the above case then can be written as:
> > > > 
> > > > 	unsafe extern "C" fn probe_callback(
> > > > 	    pdev: *mut bindings::platform_device,
> > > > 	) -> core::ffi::c_int {
> > > > 	    from_kernel_result(|| {
> > > > 		let ptr = devm_alloc(pdev)?;
> > > > 		bindings::platform_set_drvdata(pdev, ptr);
> > > > 		Ok(0)
> > > > 	    })
> > > > 	}
> > > > 
> > > > less magical, but the control flow is more clear.
> > > > 
> > > > Thoughts?  
> > > 
> > > I don't think even the closure is necessary? Why not just
> > > 
> > >  	pub fn from_kernel_result<T: From<i16>>(r: Result<T>) -> T
> > > 
> > > and
> > > 
> > >  	unsafe extern "C" fn probe_callback(
> > >  	    pdev: *mut bindings::platform_device,
> > >  	) -> core::ffi::c_int {
> > >  	    from_kernel_result({
> > >  		let ptr = devm_alloc(pdev)?;  
> > 
> > Hmm.. looks like the `?` only "propagating them (the errors) to the
> > calling *function*":
> > 
> > 	https://doc.rust-lang.org/reference/expressions/operator-expr.html#the-question-mark-operator
> > 
> > , so this doesn't work as we expect:
> > 
> > 	https://play.rust-lang.org/?version=stable&mode=debug&edition=2021&gist=64c9039e1da2c436f9f4f5ea46e97e90
> 
> Oh, you're absolutely right, I guess I shouldn't be doing code review
> in the middle of the night...
> 

;-)

> However, if we have the try blocks then we should be able to just
> rewrite it as
> 
> 	from_kernel_result(try {
> 	    ...
> 	})
> 

Thank you and Miguel for the references on try blocks, I vaguely
remember I heard of them, but never take a close look.

> I guess in this sense it might worth keeping this as a macro so we can
> tweak the implementation from closure to try blocks without having to
> edit all use sites?
> 

My preference to function instead of macro here is because I want to
avoid the extra level of abstraction and make things explict, so that
users and reviewers can understand the API behavior solely based on
Rust's types, functions and closures: they are simpler than macros, at
least to me ;-)

Speak of future changes in the implementation:

First, I think the macro version here is just a poor-man's try block, in
other words, I'd expect explicit use of try blocks intead of
`from_kernel_result` when the feature is ready. If that's the case, we
need to change the use sites anyway.

Second, I think the semantics is a little different between try block
implementation and closure implemention? Consider the following case:

	// the outer function return a Result<i32>

	let a = from_kernel_result!({
		...
		return Some(0); // x is i32
		..
	});

	return Some(a + 1);

Do both implementation share the same behavior?

No one can predict the future, but being simple at the beginning sounds
a good strategy to me.

Regards,
Boqun

> Best,
> Gary