Date: Fri, 7 Jun 2024 13:08:25 +0200
From: Alice Ryhl <aliceryhl@...gle.com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: a.hindborg@...sung.com, alex.gaynor@...il.com, ardb@...nel.org, 
	benno.lossin@...ton.me, bjorn3_gh@...tonmail.com, boqun.feng@...il.com, 
	gary@...yguo.net, jbaron@...mai.com, jpoimboe@...nel.org, 
	linux-kernel@...r.kernel.org, linux-trace-kernel@...r.kernel.org, 
	mathieu.desnoyers@...icios.com, mhiramat@...nel.org, 
	miguel.ojeda.sandonis@...il.com, ojeda@...nel.org, rostedt@...dmis.org, 
	rust-for-linux@...r.kernel.org, wedsonaf@...il.com
Subject: Re: [PATCH 1/3] rust: add static_call support

On Fri, Jun 7, 2024 at 12:52 PM Peter Zijlstra <peterz@...radead.org> wrote:
>
> On Fri, Jun 07, 2024 at 09:43:29AM +0000, Alice Ryhl wrote:
> > Peter Zijlstra <peterz@...radead.org> wrote:
> > > On Thu, Jun 06, 2024 at 09:09:00PM +0200, Miguel Ojeda wrote:
> > > > On Thu, Jun 6, 2024 at 7:19 PM Peter Zijlstra <peterz@...radead.org> wrote:
> > > > >
> > > > > This is absolutely unreadable gibberish -- how am I supposed to keep
> > > > > this in sync with the rest of the static_call infrastructure?
> > > >
> > > > Yeah, they are macros, which look different from "normal" Rust code.
> > >
> > > Macros like CPP ?
> >
> > Yes, this patch series uses declarative macros, which are the closest
> > that Rust has to the C preprocessor. They are powerful, but just like
> > CPP, they can become pretty complicated and hard to read if you are
> > doing something non-trivial.
> >
> > The macro_rules! block is how you define a new declarative macro.
>
> I'm sorry, but 30+ years of reading ! as NOT (or factorial) isn't going
> to go away. So I'm reading your macros do NOT rule.

So you already understand ! in two ways, but you don't want to add a
third? That seems to violate the Zero One Infinity rule. :)

https://en.wikipedia.org/wiki/Zero_one_infinity_rule

> > The ($name:ident($($args:expr),* $(,)?)) part defines the arguments to
> > the declarative macro. This syntax means:
> >
> > 1. The input starts with any identifier, which we call $name.
> > 2. Then there must be a ( token.
>
> The above exaple fails, because the next token is :ident, whatever the
> heck that might be. Also, extra points for line-noise due to lack of
> whitespace.

The :ident part means that $name should be parsed as an identifier.
Similarly, the :expr part means that $args should be parsed as an
expression. It doesn't mean that the input should literally contain
":ident".

> > So for example, you might invoke the macro like this:
> >
> > static_call!(tp_func_my_tracepoint(__data, &mut my_task_struct));
>
> static_call NOT (blah dog blah);
>
> > Inside the macro, you will see things such as:
> > $crate::macros::paste! { $crate::bindings:: [<__SCK__ $name >]; }
> >
> > The Rust syntax for invoking a macro has an exclamation mark after the
>
> Like I said before, the creator of Rust must've been an esoteric
> language freak and must've wanted to make this unreadable on purpose :/
>
> Also, why the white space beteen the :: scope operator and the [< thing?
> that's just weird. I would then expect the output to be:
>
>   ...::bindings:: __SCK__my_static_key

Sorry, you are right. There is a space in the output.

> > name, so you know that $crate::macros::paste is a macro. The `paste`
> > macro just emits its input unchanged, except that any identifiers
> > between [< and >] are concatenated into a single identifier. So if $name
> > is my_static_key, then the above invocation of paste! emits:
> >
> >       $crate::bindings::__SCK__my_static_key;
>
> But it doesn't, so it isn't unmodified, it seems to strip whitespace.

Thanks for the correction. The actual output is:

$crate::bindings:: __SCK__my_static_key;

However, although whitespace is generally not used here, the syntax allows it.

> > The $crate::bindings module is where the output of bindgen goes, so this
> > should correspond to the C symbol called __SCK__my_static_key.
> >
> > > > Is there something we could do to help here? I think Alice and others
> > > > would be happy to explain how it works and/or help maintain it in the
> > > > future if you prefer.
> > >
> > > Write a new language that looks more like C -- pretty please ? :-)
> > >
> > > Mostly I would just really like you to just use arm/jump_label.h,
> > > they're inline functions and should be doable with IR, no weirdo CPP
> > > involved in this case.
> >
> > I assume that you're referring to static_key_false here? I don't think
> > that function can be exposed using IR because it passes the function
> > argument called key as an "i" argument to an inline assembly block. Any
> > attempt to compile static_key_false without knowing the value of key at
> > compile time will surely fail to compile with the
> >
> >       invalid operand for inline asm constraint 'i'
> >
> > error.
>
> You can have clang read the header files and compile them into
> Intermediate-Representation, and have it splice the lot into the Rust
> crap's IR and voila, compile time.
>
> You just need to extend the rust thing to be able to consume C header
> files.

I wish! There are people, including me, who want this. See e.g. this
very recent document:
https://rust-lang.github.io/rust-project-goals/2024h2/Seamless-C-Support.html

But there are also people who dislike the idea, so it does not have
unanimous support yet, unfortunately.

Ultimately, I have to work with what exists today.

Alice

Powered by blists - more mailing lists