Crates.io | macro_rules_attribute |
lib.rs | macro_rules_attribute |
version | 0.1.5 |
source | src |
created_at | 2019-09-05 17:19:43.8442 |
updated_at | 2023-05-13 19:00:23.748316 |
description | Use declarative macros in attribute or derive position |
homepage | https://crates.io/crates/macro_rules_attribute |
repository | https://github.com/danielhenrymantilla/macro_rules_attribute-rs |
max_upload_size | |
id | 162473 |
size | 47,809 |
::macro_rules_attribute
Use declarative macros in attribute or derive position.
macro_rules! my_fancy_decorator { /* … */ }
#[apply(my_fancy_decorator!)]
struct Foo { /* … */ }
macro_rules! MyFancyDerive { /* … */ }
#[derive(MyFancyDerive!)] /* using this crate's `#[derive]` attribute */
struct Foo { /* … */ }
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,
}
The former does not scale well, since it leads to rightward drift and "excessive" braces.
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 proc
edural
macro.
Anything involving ident
generation / derivation, for instance, will very
often require proc
edural macros, unless it is simple enough for
::paste
to handle it.
With this crate's #[apply]
and #[derive]
attributes, it is now possible to use proc_macro_attribute
syntax to apply a
macro_rules!
macro:
#[macro_use]
extern crate macro_rules_attribute;
macro_rules! foo {
// …
# ( $($tt:tt)* ) => ()
}
macro_rules! Bar {
// …
# ( $($tt:tt)* ) => ()
}
#[apply(foo)]
#[derive(Debug, Bar!)]
struct Struct {
some_field: SomeType,
}
#
# fn main() {}
without even depending on ::quote
, ::syn
or ::proc-macro2
, for
fast compile times.
#[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() {}
-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
heavyweight1
troika/trinity/triumvirate of more advanced proc-macro crates.
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() {}
lazy_static!
sSay 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() {}
An optional compilation feature, "verbose-expansions"
can be used to print at
compile-time the exact output of each macro invocation from this crate:
[dependencies]
macro_rules_attribute.version = "..."
macro_rules_attribute.features = ["verbose-expansions"]
derive
aliases# fn main() {}
#[macro_use]
extern crate macro_rules_attribute;
derive_alias! {
#[derive(Ord!)] = #[derive(PartialEq, Eq, PartialOrd, Ord)];
}
#[derive(Debug, Clone, Copy, Ord!)]
struct Foo {
// …
}
derive_alias!
and #[derive]
for more info.cfg
aliases# 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 { /* … */ }
#[macro_use] extern crate macro_rules_attribute
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.
(note that only syn
with the "full"
features would be the truly heavyweight party) ↩