33 releases
0.1.32 | Mar 18, 2023 |
---|---|
0.1.31 | Mar 18, 2023 |
0.1.27 | Jan 18, 2023 |
0.1.22 | Dec 24, 2022 |
0.1.7 | Nov 28, 2022 |
#120 in Programming languages
281 downloads per month
Used in 2 crates
(via toad)
160KB
2.5K
SLoC
naan
deliciously succinct
naan is a functional programming prelude for the Rust language that is:
- easy
- useful
std
- andalloc
-optional- FAST - exclusively uses concrete types (no
dyn
amic dispatch) meaning near-zero perf cost
Table of Contents
Higher-Kinded Types
HKT - What it is
When talking about types, it can be useful to be able to differentiate between a concrete type (u8
, Vec<u8>
, Result<File, io::Error>
)
and a generic type without its parameters supplied. (Vec
, Option
, Result
)
For example, Vec
is a 1-argument (unary) type function, and Vec<u8>
is a concrete type.
Kind refers to how many (if any) parameters a type has.
HKT - Why it's useful
Top · Up - HKTs
In vanilla Rust, Result::map
and Option::map
have very similar shapes:
impl<A, E> Result<A, E> {
fn map<B>(self, f: impl FnMut(A) -> B) -> Result<B, E>;
}
impl<A> Option<A> {
fn map<B>(self, f: impl FnMut(A) -> B) -> Option<B>;
}
it would be useful (for reasons we'll expand on later) to have them
both implement a Map
trait:
trait Map<A> {
fn map<B>(self: Self<A>, f: impl FnMut(A) -> B) -> Self<B>;
}
impl<A> Map<A> for Option<A> {
fn map<B>(self, f: impl FnMut(A) -> B) -> Option<B> {
Option::map(self, f)
}
}
but this code snippet isn't legal Rust because Self
needs to be generic and in vanilla Rust Self
must be a concrete type.
HKT - How it's done
With the introduction of Generic associated types, we can write a trait that can effectively replace a "generic self" feature.
Now we can actually write the trait above in legal, stable rust:
trait HKT {
type Of<A>;
}
struct OptionHKT;
impl HKT for OptionHKT {
type Of<A> = Option<A>;
}
trait Map<M, A>
where M: HKT<Of<A> = Self>
{
fn map<B, F>(self, f: F) -> M::Of<B>
where F: FnMut(A) -> B;
}
impl<A> Map<OptionHKT, A> for Option<A> {
fn map<B, F>(self, f: F) -> Option<B>
where F: FnMut(A) -> B
{
Option::map(self, f)
}
}
Currying
Top · Prev - HKT · Next - Function Composition
Currying - What it is
Currying is the technique where naan
gets its name. Function currying is the strategy of splitting functions that
accept more than one argument into multiple functions.
Example:
fn foo(String, usize) -> usize;
foo(format!("bar"), 12);
would be curried into:
fn foo(String) -> impl Fn(usize) -> usize;
foo(format!("bar"))(12);
Currying - Why it's useful
Currying allows us to provide some of a function's arguments and provide the rest of this partially applied function's arguments at a later date.
This allows us to use functions to store state, and lift functions that accept any number
of parameters to accept Results using Apply
EXAMPLE: reusable function with a stored parameter
use std::fs::File;
use naan::prelude::*;
fn copy_file_to_dir(dir: String, file: File) -> std::io::Result<()> {
// ...
# Ok(())
}
fn main() {
let dir = std::env::var("DEST_DIR").unwrap();
let copy = copy_file_to_dir.curry().call(dir);
File::open("a.txt").bind1(copy.clone())
.bind1(|_| File::open("b.txt"))
.bind1(copy.clone())
.bind1(|_| File::open("c.txt"))
.bind1(copy);
}
/*
equivalent to:
fn main() {
let dir = std::env::var("DEST_DIR").unwrap();
copy_file_to_dir(dir.clone(), File::open("a.txt")?)?;
copy_file_to_dir(dir.clone(), File::open("b.txt")?)?;
copy_file_to_dir(dir, File::open("c.txt")?)?;
}
*/
EXAMPLE: lifting a function to accept Results (or Options)
use std::fs::File;
use naan::prelude::*;
fn append_contents(from: File, to: File) -> std::io::Result<()> {
// ...
# Ok(())
}
fn main() -> std::io::Result<()> {
Ok(append_contents.curry()).apply1(File::open("from.txt"))
.apply1(File::open("to.txt"))
.flatten()
}
/*
equivalent to:
fn main() -> std::io::Result<()> {
let from = File::open("from.txt")?;
let to = File::open("to.txt")?;
append_contents(from, to)
}
*/
Currying - How it's done
naan introduces a few new function traits that add
ergonomics around currying and function composition;
F1
, F2
and F3
. These traits extend the builtin function
traits Fn
and FnOnce
with methods that allow currying and function
composition.
(note that each arity has a "callable multiple times"
version and a "callable at least once" version. The latter traits are
denoted with a suffix of Once
)
F2
and F2Once
Definitions
F2
and F2Once
Definitionspub trait F2Once<A, B, C>: Sized {
/// The concrete type that `curry` returns.
type Curried;
/// Call the function
fn call1(self, a: A, b: B) -> C;
/// Curry this function, transforming it from
///
/// `fn(A, B) -> C`
/// to
/// `fn(A) -> fn(B) -> C`
fn curry(self) -> Self::Curried;
}
pub trait F2<A, B, C>: F2Once<A, B, C> {
/// Call the function with all arguments
fn call(&self, a: A, b: B) -> C;
}
impl<F, A, B, C> F2<A, B, C> for F where F: Fn(A, B) -> C { /* <snip> */ }
impl<F, A, B, C> F2Once<A, B, C> for F where F: FnOnce(A, B) -> C { /* <snip> */ }
Function Composition
Top · Prev - Currying · Next - Typeclasses
Composition - What it is
Top · Up - Function Composition
Function composition is the strategy of chaining functions sequentially by automatically passing the output of one function to the input of another.
This very powerful technique lets us concisely express programs in terms of data that flows through pipes, rather than a sequence of time-bound statements:
use naan::prelude::*;
struct Apple;
struct Orange;
struct Grape;
#[derive(Debug, PartialEq)]
struct Banana;
fn apple_to_orange(a: Apple) -> Orange {
Orange
}
fn orange_to_grape(o: Orange) -> Grape {
Grape
}
fn grape_to_banana(g: Grape) -> Banana {
Banana
}
fn main() {
let apple_to_banana = apple_to_orange.chain(orange_to_grape)
.chain(grape_to_banana);
assert_eq!(apple_to_banana.call(Apple), Banana)
}
Typeclasses
Top · Prev - Function Composition
Some of the most powerful & practical types in programming are locked behind a feature that many languages choose not to implement in Higher-Kinded Types.
Utilities like map
, unwrap_or
, and and_then
are enormously useful tools
in day-to-day rust that allow us to conveniently skip a lot of hand-written control flow.
Comparing and_then
and map
to their desugared equivalent
and_then
and map
to their desugared equivalentuse std::io;
fn network_fetch_name() -> io::Result<String> {
Ok("harry".into())
}
fn network_send_message(msg: String) -> io::Result<()> {
Ok(())
}
fn global_state_store_name(name: &str) -> io::Result<()> {
Ok(())
}
// Declarative
fn foo0() -> io::Result<()> {
network_fetch_name().and_then(|name| {
global_state_store_name(&name)?;
Ok(name)
})
.map(|name| format!("hello, {name}!"))
.and_then(network_send_message)
}
// Idiomatic
fn foo1() -> io::Result<()> {
let name = network_fetch_name()?;
global_state_store_name(&name)?;
network_send_message(format!("hello, {name}!"))
}
// Imperative
fn foo2() -> io::Result<()> {
let name = match network_fetch_name() {
| Ok(name) => name,
| Err(e) => return Err(e),
};
match global_state_store_name(&name) {
| Err(e) => return Err(e),
| _ => (),
};
network_send_message(format!("hello, {name}!"))
}
A couple notes:
- the "idiomatic" implementation is the most brief and scannable
- the idiomatic and imperative implementations are more difficult to refactor due to scope sharing; imperative statements depend on the previous statements in order to be meaningful, while declarative expressions have little to no coupling to state or scope.
The value proposition of these typeclasses is that they allow us to think of types like Result, Option and Iterators as being abstract containers.
We don't need to know much about their internals to know how to use them effectively and productively.
This extremely simple but powerful metaphor allows us to solve some very complex problems with data structures that have a shared set of interfaces.
Semigroup and Monoid
Combining two values of a concrete type
Semigroup
is the name we give types that support some associative combination
of two values (a.append(b)
).
🔎 Associative means a.append( b.append(c) )
must equal a.append(b).append(c)
.
Examples:
- integer addition
1 * (2 * 3) == (1 * 2) * 3
- integer multiplication
1 + (2 + 3) == (1 + 2) + 3
- string concatenation
"a".append("b".append("c")) == "a".append("b").append("c") == "abc"
Vec<T>
concatenationvec![1].append(vec![2].append(vec![3])) == vec![1, 2, 3]
Option<T>
(only whenT
implementsSemigroup
)Some("a").append(Some("b")) == Some("ab")
Result<T, _>
(only whenT
implementsSemigroup
)Ok("a").append(Ok("b")) == Ok("ab")
Monoid
extends Semigroup
with an "identity" or "empty" value, that will do nothing when appended to another.
Examples:
- 0 in integer addition
0 + 1 == 1
- 1 in integer multiplication
1 * 2 == 2
- empty string
String::identity() == ""
"".append("a") == "a"
Vec<T>
Vec::<u32>::identity() == vec![]
vec![].append(vec![1, 2]) == vec![1, 2]
These are defined as:
pub trait Semigroup {
// 🔎 Note that this can be **any** combination of 2 selves,
// not just concatenation.
//
// The only rule is that implementations have to be associative.
fn append(self, b: Self) -> Self;
}
pub trait Monoid: Semigroup {
fn identity() -> Self;
}
Alt and Plus
Combining two values of a generic type
Alt
is the name we give to generic types that support an associative operation
on 2 values of the same type (a.alt(b)
).
🔎 Alt
is identical to Semigroup
, but the implementor is generic.
🔎 alt
is identical to Result::or
and Option::or
.
Examples:
Vec<T>
vec![1].alt(vec![2]) == vec![1, 2]
Result<T, _>
Ok(1).alt(Err(_)) == Ok(1)
Option<T>
None.alt(Some(1)) == Some(1)
Plus
extends Alt
with an "identity" or "empty" value, that will do nothing when alt
ed to another.
🔎 Plus
is identical to Monoid
, but the implementor is generic.
Examples:
Vec<T>
(Vec::empty() == vec![]
)Option<T>
(Option::empty() == None
)
These are defined as:
// 🔎 `Self` must be generic over some type `A`.
pub trait Alt<F, A>
where Self: Functor<F, A>,
F: HKT1<T<A> = Self>
{
fn alt(self, b: Self) -> Self;
}
pub trait Plus<F, A>
where Self: Alt<F, A>,
F: HKT1<T<A> = Self>
{
fn empty() -> F::T<A>;
}
Functor
using a function to transform values within a container
Functor
is the name we give to types that allow us to take a function from A -> B
and effectively "penetrate" a type and apply it to some F<A>
, yielding F<B>
(a.fmap(a_to_b)
).
🔎 This is identical to Result::map
and Option::map
.
🔎 There is a separate trait FunctorOnce
which extends Functor
to know that the mapping function will only be called once.
Functor
is defined as:
// 🔎 `Self` must be generic over some type `A`
pub trait Functor<F, A> where F: HKT1<T<A> = Self>
{
// 🔎 given a function `A -> B`,
// apply it to the values of type `A` in `Self<A>` (if any),
// yielding `Self<B>`
fn fmap<AB, B>(self, f: AB) -> F::T<B> where AB: F1<A, B>;
}
Bifunctor
mapping types with 2 generic parameters
Bifunctor
is the name we give to types that have 2 generic parameters,
both of which can be map
ped.
Bifunctor
requires:
bimap
- transforms
T<A, B>
toT<C, D>
, given a functionA -> C
and anotherB -> D
.
- transforms
Bifunctor
provides 2 methods:
lmap
(map left type)T<A, B> -> T<C, B>
rmap
(map right type)T<A, B> -> T<A, D>
🔎 There is a separate trait BifunctorOnce
which extends Bifunctor
to know that the mapping functions will only be called once.
Bifunctor
is defined as:
pub trait Bifunctor<F, A, B>
where F: HKT2<T<A, B> = Self>
{
/// 🔎 In Result, this combines `map` and `map_err` into one step.
fn bimap<A2, B2, FA, FB>(self, fa: FA, fb: FB) -> F::T<A2, B2>
where FA: F1<A, A2>,
FB: F1<B, B2>;
/// 🔎 In Result, this maps the "Ok" type and is equivalent to `map`.
fn lmap<A2, FA>(self, fa: FA) -> F::T<A2, B>
where Self: Sized,
FA: F1<A, A2>
{
self.bimap(fa, |b| b)
}
/// 🔎 In Result, this maps the "Error" type and is equivalent to `map_err`.
fn rmap<B2, FB>(self, fb: FB) -> F::T<A, B2>
where Self: Sized,
FB: F1<B, B2>
{
self.bimap(|a| a, fb)
}
}
Foldable
Unwrapping & transforming entire data structures
Types that are Foldable
can be unwrapped and collected into a new value.
Fold is a powerful and complex operation because of how general it is; if something
is foldable, it can be folded into practically anything.
🔎 There is a separate trait FoldableOnce
which extends Foldable
to know that the folding function can only be called once.
Folding can be thought of as a series of steps:
- Given some foldable
F<T>
, and you want aR
- I have a
Vec<Option<u32>>
and I want to sum the u32s that are Some, and discard the Nones
- I have a
- Start with some initial value of type
R
- I want a sum of u32s, so I'll start with zero.
- Write a function of type
Fn(R, T) -> R
. This will be called with the initialR
along with a value of typeT
from withinF<T>
. The function will be called repeatedly with theR
returned by the last call until there are no moreT
s inF<T>
.|sum_so_far, option_of_u32| sum_so_far + option_of_u32.unwrap_or(0)
- This function will be called for every
T
contained inF<T>
, collecting them into the initial valueR
you provided.vec![Some(1), None, Some(2), Some(4)].fold(|sum, n| sum + n.unwrap_or(0)) == 7
Examples
Result to Option
use naan::prelude::*;
fn passing() -> Result<u32, ()> {
Ok(0)
}
fn failing() -> Result<u32, ()> {
Err(())
}
assert_eq!(match passing() {
| Ok(t) => Some(t),
| _ => None,
},
Some(0));
assert_eq!(passing().fold1(|_, t| Some(t), None), Some(0));
assert_eq!(failing().fold1(|_, t| Some(t), None), None);
Collapse a Vec
use naan::prelude::*;
assert_eq!(vec![1, 2, 3].foldl(|sum, n| sum + n, 0), 6);
assert_eq!(vec![2, 4, 6].foldl(|sum, n| sum * n, 1), 48);
assert_eq!(vec!["a", "b", "c"].foldl(|acc, cur| format!("{acc}{cur}"), String::from("")),
"abc");
Foldable
is defined as:
pub trait Foldable<F, A> where F: HKT1<T<A> = Self>
{
/// Fold the data structure from left -> right
fn foldl<B, BAB>(self, f: BAB, b: B) -> B
where BAB: F2<B, A, B>;
/// Fold the data structure from right -> left
fn foldr<B, ABB>(self, f: ABB, b: B) -> B
where ABB: F2<A, B, B>;
/// Fold the data structure from left -> right
fn foldl_ref<'a, B, BAB>(&'a self, f: BAB, b: B) -> B
where BAB: F2<B, &'a A, B>,
A: 'a;
/// Fold the data structure from right -> left
fn foldr_ref<'a, B, ABB>(&'a self, f: ABB, b: B) -> B
where ABB: F2<&'a A, B, B>,
A: 'a;
}
🔎 Foldable
provides many additional methods derived from the required methods above. Full documentation can be found here.
use naan::prelude::*;
fn is_odd(n: &usize) -> bool {
n % 2 == 1
}
fn is_even(n: &usize) -> bool {
n % 2 == 0
}
assert_eq!(Some("abc".to_string()).fold(), "abc".to_string());
assert_eq!(Option::<String>::None.fold(), "");
let abc = vec!["a", "b", "c"].fmap(String::from);
assert_eq!(abc.clone().fold(), "abc");
assert_eq!(abc.clone().intercalate(", ".into()), "a, b, c".to_string());
assert_eq!(vec![2usize, 4, 8].any(is_odd), false);
assert_eq!(vec![2usize, 4, 8].all(is_even), true);
Lazy IO
License
Licensed under either of
- Apache License, Version 2.0, (LICENSE-APACHE or https://www.apache.org/licenses/LICENSE-2.0)
- MIT license (LICENSE-MIT or https://opensource.org/licenses/MIT)
at your option.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.