macro_rules! macros can be extremely powerful, but their call-site ergonomics are sometimes not great, especially when decorating item definitions.

Indeed, compare:

foo! {
    struct Struct {
        some_field: SomeType,
    }
}

to:

#[foo]
struct Struct {
    some_field: SomeType,
}

1. The former does not scale well, since it leads to rightward drift and "excessive" braces.

2. But on the other hand, the latter requires setting up a dedicated crate for the compiler, a proc-macro crate. And 99% of the time this will pull the ::syn and ::quote dependencies, which have a non-negligible compile-time overhead (the first time they are compiled).

   • note: these crates are a wonderful piece of technology, and can lead to extremely powerful macros. When the logic of the macro is so complicated that it requires a recursive tt muncher when implemented as a macro_rules! macro, it is definitely time to be using a procedural macro.

     Anything involving ident generation / derivation, for instance, will very often require procedural macros, unless it is simple enough for ::paste to handle it.

Solution

Nicer derives

#[macro_use]
extern crate macro_rules_attribute;

// Easily define shorthand aliases for "derive groups"
derive_alias! {
    #[derive(Eq!)] = #[derive(Eq, PartialEq)];
    #[derive(Ord!)] = #[derive(Ord, PartialOrd, Eq!)];
    #[derive(Copy!)] = #[derive(Copy, Clone)];
    #[derive(StdDerives!)] = #[derive(Debug, Copy!, Default, Ord!, Hash)];
}

/// Strongly-typed newtype wrapper around a `usize`, to be used for `PlayerId`s.
#[derive(StdDerives!, Into!, From!)]
pub
struct PlayerId /* = */ (
    pub usize,
);

// You can also fully define your own derives using `macro_rules!` syntax
// (handling generic type definitions may be the only finicky thing, though…)
macro_rules! Into {(
    $( #[$attr:meta] )*
    $pub:vis
    struct $NewType:ident (
        $(#[$field_attr:meta])*
        $field_pub:vis
        $Inner:ty $(,

        $($rest:tt)* )?
    );
) => (
    impl ::core::convert::Into<$Inner> for $NewType {
        #[inline]
        fn into (self: $NewType)
          -> $Inner
        {
            self.0
        }
    }
)} use Into;

macro_rules! From {(
    $( #[$attr:meta] )*
    $pub:vis
    struct $NewType:ident (
        $(#[$field_attr:meta])*
        $field_pub:vis
        $Inner:ty $(,

        $(#[$other_field_attr:meta])*
        $other_field_pub:vis
        $Rest:ty )* $(,)?
    );
) => (
    impl ::core::convert::From<$Inner> for $NewType {
        #[inline]
        fn from (inner: $Inner)
          -> Self
        {
            Self(inner, $($Rest::default),*)
        }
    }
)} use From;
#
# fn main() {}

Have a -lite version of a proc-macro dependency that thus requires unergonomic macro_rules!?

Say you are writing a (pervasive and yet) tiny dependency within the async ecosystem.

• By virtue of working with async, you'll most probably need to deal with pin-projections, and thence, with ::pin-project.

• But by virtue of being (pervasive and yet) tiny, you don't want to depend on the quote / proc-macro2 / syn heavyweight[^only_full_syn_is_heavy] troika/trinity/triumvirate of more advanced proc-macro crates.

[^only_full_syn_is_heavy]: (note that only syn with the "full" features would be the truly heavyweight party)

Hence why you may reach for something such as ::pin-project-lite, and its pin_project! macro_rules!-based polyfill of the former's #[pin_project] attribute.

But this suddenly hinders the ergonomics of your type definitions, and, worse, would not be composable whenever the pattern were to be repeated for some other functionality (e.g., say a cell_project! similar macro).

Say no more! Time to #[apply] our neat trick:

#[macro_use]
extern crate macro_rules_attribute;

use {
    ::core::pin::{
        Pin,
    },
    ::pin_project_lite::{
        pin_project,
    },
};

#[apply(pin_project!)]
struct Struct<T, U> {
    #[pin]
    pinned: T,
    unpinned: U,
}

impl<T, U> Struct<T, U> {
    fn method(self: Pin<&mut Self>) {
        let this = self.project();
        let _: Pin<&mut T> = this.pinned; // Pinned reference to the field
        let _: &mut U = this.unpinned; // Normal reference to the field
    }
}
#
# fn main() {}

More ergonomic lazy_static!s

Say you had something like:

# use Sync as Logic;
#
static MY_GLOBAL: &dyn Logic = &Vec::<i32>::new();

and now you want to change the value of that MY_GLOBAL to something that isn't const-constructible, and yet would like to minimize the churn in doing so.

// (For those unaware of it, leaking memory to initialize a lazy static is
// a completely fine pattern, since it only occurs once, and thus, a bounded
// amount of times).
static MY_GLOBAL: &dyn Logic = Box::leak(Box::new(vec![42, 27])); // Error: not `const`!

You could directly use a lazy_static! or a OnceCell, but then the definition of your static will now appear noisier than it needs be. It's time for attribute-position polish!

First, define the helper around, say, OnceCell's Lazy type:

macro_rules! lazy_init {(
    $( #[$attrs:meta] )*
    $pub:vis
    static $NAME:ident: $Ty:ty = $init_value:expr ;
) => (
    $( #[$attrs] )*
    $pub
    static $NAME : ::once_cell::sync::Lazy<$Ty> =
        ::once_cell::sync::Lazy::new(|| $init_value)
    ;
)} pub(in crate) use lazy_init;

and now it is time to use it!:

# use Sync as Logic;
#
#[macro_use]
extern crate macro_rules_attribute;

#[apply(lazy_init)]
static MY_GLOBAL: &dyn Logic = Box::leak(Box::new(vec![42, 27]));
#
# macro_rules! lazy_init {(
#     $( #[$attrs:meta] )*
#     $pub:vis
#     static $NAME:ident : $Ty:ty = $init_value:expr ;
# ) => (
#     $( #[$attrs] )*
#     $pub
#     static $NAME : ::once_cell::sync::Lazy<$Ty> =
#         ::once_cell::sync::Lazy::new(|| $init_value)
#     ;
# )} use lazy_init;
#
# fn main() {}

# fn main() {}
#[macro_use]
extern crate macro_rules_attribute;

attribute_alias! {
    #[apply(complex_cfg!)] = #[cfg(
        any(
            any(
                foo,
                feature = "bar",
            ),
            all(
                target_os = "fenestrations",
                not(target_arch = "Pear"),
            ),
        ),
    )];
}

#[apply(complex_cfg!)]
mod some_item { /* … */ }

If you are allergic to #[macro_use] unscoped / globally-preluded semantics, you may not be fond of having to use:

#[macro_use]
extern crate macro_rules_attribute;
# fn main() {}

like this documentation pervasively does.

In that case, know that you may very well stick to using use imports:

use ::macro_rules_attribute::{derive, derive_alias, /* … */};
// or even
use ::macro_rules_attribute::*;

derive_alias! {
    #[derive(Copy!)] = #[derive(Clone, Copy)];
}

#[derive(Copy!)]
struct Foo;

or even inlining the fully qualified paths (but note that the …_alias! macros still take unqualified paths inside the definitions):

::macro_rules_attribute::derive_alias! {
    #[derive(Copy!)] = #[derive(Clone, Copy)];
}

#[::macro_rules_attribute::derive(Copy!)]
struct Foo;

I personally find these approaches too noisy to be worth it, despite the so gained "namespace purity", hence my not using that pattern across the rest of the examples.