::named-generics-bundle
Convenience macros to help with the "bundle multiple generic params with a helper trait" pattern

API summary
# use ::named_generics_bundle::named_generics_bundle;
# use Sized as SomeBounds;
#
#[named_generics_bundle]
trait SomeTrait {
type SomeAssocType: SomeBounds + Clone;
}
fn some_api<S: SomeTrait>(nameable: S::SomeAssocType) {
nameable.clone();
}
type Example = SomeTrait![
# SomeAssocType = (),
];
some_api::<Example>(
# (),
);
some_api::<SomeTrait![SomeAssocType = i32]>(42);
Motivation
As your Rust projects grows in scope and functionality, your generic types may end up with more and
more generic parameters:
# use ::core::marker::PhantomData;
# pub trait Burns {}
# enum Uranium {} impl Burns for Uranium {}
# pub trait EnergyForm {}
# enum Beam {} impl EnergyForm for Beam {}
# pub trait YieldsEnergy { fn yield_energy<E: EnergyForm>(&self, _: &mut impl Burns) -> E { todo!() } }
# enum FluxCapacitor {} impl YieldsEnergy for FluxCapacitor {}
#
#[derive(Debug)]
struct Device<Fuel, Engine, Output>
where
Fuel: Burns,
Engine: YieldsEnergy,
Output: EnergyForm,
{
fuel: Fuel,
engine: Engine,
_p: PhantomData<fn() -> Output>,
}
impl<Fuel, Engine, Output> Device<Fuel, Engine, Output>
where
Fuel: Burns,
Engine: YieldsEnergy,
Output: EnergyForm,
{
pub fn assemble(fuel: Fuel, engine: Engine) -> Self {
Self { fuel, engine, _p: PhantomData }
}
pub fn frobnicate(&mut self) -> Output {
self.engine.yield_energy(&mut self.fuel)
}
}
fn get_away_from_the_beam() -> ! {
let mut device = Device::<Uranium, FluxCapacitor, Beam>::assemble(
# todo!(), todo!(),
);
let beam: Beam = device.frobnicate();
match beam {}
}
It turns out, you can make this already significantly more convenient from the point of view of
the callee by using a helper trait to bundle all the generic parameters as associated types, and
making the callees generic only over that helper trait.
# pub trait Burns {}
# enum Uranium {} impl Burns for Uranium {}
# pub trait EnergyForm {}
# enum Beam {} impl EnergyForm for Beam {}
# pub trait YieldsEnergy { fn yield_energy<E: EnergyForm>(&self, _: &mut impl Burns) -> E { todo!() } }
# enum FluxCapacitor {} impl YieldsEnergy for FluxCapacitor {}
#
pub trait DeviceSetup {
type Fuel: Burns;
type Output: EnergyForm;
type Engine: YieldsEnergy;
}
#[derive(Debug)]
struct Device<S: DeviceSetup> {
fuel: S::Fuel,
engine: S::Engine,
}
impl<S: DeviceSetup> Device<S> {
pub fn assemble(fuel: S::Fuel, engine: S::Engine) -> Self {
Self { fuel, engine }
}
pub fn frobnicate(&mut self) -> S::Output {
self.engine.yield_energy(&mut self.fuel)
}
}
fn get_away_from_the_beam() -> ! {
enum BlackMesaStyle {}
impl DeviceSetup for BlackMesaStyle {
type Fuel = Uranium;
type Engine = FluxCapacitor;
type Output = Beam;
}
let mut device = Device::<BlackMesaStyle>::assemble(
# todo!(), todo!(),
);
let beam: Beam = device.frobnicate();
match beam {}
}
This is already quite a useful trick, but it comes with certain caveats:
-
The most notable one, is that you can no longer inline-turbofish the generics: it is
necessary to define a helper type, and this can get unwieldy when there are outer generic
parameters in scope.
-
There are also some secondary issues, such as slapping #[derive(Clone)] on the struct Engine
and this resulting in Output having to be clone, even though it is not part of the actual
fields of Engine (it is a mere dummy PhantomData instead).
- To be fair, this is a limitation/bug stemming from
#[derive(Clone)] itself, and other
similar stdlib #[derive()]s, being rather dumb w.r.t. the impls they generate, w.r.t.
adding unnecessary bounds on the params themselves rather than focusing on bounding the
field types (there is a desire to do the latter, called "perfect derives", but they only
want to do so once all edge cases, such as recursive types, can be handled. There are also
some "implicit SemVer" considerations involved as well).
So you:
- either make
Output : Clone even if the only thing susceptible of being .clone()d is
the Device,
- or you stop using
#[derive(Clone)], and manually impl the trait in question. Which is
quite cumbersome! Or you involve some third-party lib to help you with that in a smarter or at least more tweakable manner than the stdlib, such as ::derivative (but this is a rather old crate, nowadays, and most
people feel like it is unnecessary to add a dependency for such a small thing).
Enters this crate!
Detailed Example
# pub trait Burns {}
# enum Uranium {} impl Burns for Uranium {}
# pub trait EnergyForm {}
# enum Beam {} impl EnergyForm for Beam {}
# pub trait YieldsEnergy { fn yield_energy<E: EnergyForm>(&self, _: &mut impl Burns) -> E { todo!() } }
# enum FluxCapacitor {} impl YieldsEnergy for FluxCapacitor {}
#
#[::named_generics_bundle::named_generics_bundle] pub trait DeviceSetup {
type Fuel: Burns;
type Output: EnergyForm;
type Engine: YieldsEnergy;
}
#[derive(Debug)]
struct Device<S: DeviceSetup> {
fuel: S::Fuel,
engine: S::Engine,
}
impl<S: DeviceSetup> Device<S> {
pub fn assemble(fuel: S::Fuel, engine: S::Engine) -> Self {
Self { fuel, engine }
}
pub fn frobnicate(&mut self) -> S::Output {
self.engine.yield_energy(&mut self.fuel)
}
}
fn get_away_from_the_beam() -> ! {
type BlackMesaStyle = DeviceSetup![ Fuel = Uranium,
Engine = FluxCapacitor,
Output = Beam,
];
let mut device = Device::<BlackMesaStyle>::assemble(
# todo!(), todo!(),
);
let beam: Beam = device.frobnicate();
match beam {}
}
# struct HandheldPortalDevice {}
# impl YieldsEnergy for HandheldPortalDevice {}
enum Cake {}
# impl EnergyForm for Cake {}
struct TestSubject {}
impl Burns for TestSubject {}
fn the_cake_is_a_lie() -> ! {
let testing_chamber =
Device::<DeviceSetup![
Engine = HandheldPortalDevice,
Output = Cake,
Fuel = TestSubject,
]>::assemble(
# todo!(), todo!(),
)
;
let puzzle_solved: Cake = testing_chamber.frobnicate();
match puzzle_solved {}
}
See the docs of #[named_generics_bundle] for more info.