Rust has a very powerful type system, but as a result it has some quirks, some would say cursed expressions. There’s a test file, weird-expr.rs, in the rust repository that tests for some of these and makes sure there consistent between updates. So I wanted to go over each of these and explain how it’s valid rust.
Note that these are not bugs, but rather extreme cases of rust features like loops, expressions, coercion and so on.
Strange
fn strange() -> bool {let _x:bool = return true;}The expression return true has the type !. The never type can coerce into any other type, so we can assign it to a boolean.
Funny
fn funny(){
fn f(_x: ()){}
f(return);
}The function f has a single parameter of () type, we can again pass return because ! will be coerced into ().
What
use std::cell::Cell;
fn what(){
fn the(x: &Cell<bool>){
return while !x.get() {x.set(true);};
}
let i = &Cell::new(false);
let dont = {||the(i)};
dont();
assert!(i.get());
}The the function takes a reference to a Cell<bool>. Inside the function, we use a while loop
while !x.get() {x.set(true);}to set the cells contains to true if its contents are false and we return that while loop which has the type ().
Next we create a variable i which is a reference to a Cell<bool> and bind a closure that calls the with i as the parameter, we then call that closure and assert that i is true.
Zombie jesus
fn zombiejesus() {
loop {
while (return) {
if (return) {
match (return) {
1 => {
if (return) {
return
} else {
return
}
}
_ => { return }
};
} else if (return) {
return;
}
}
if (return) { break; }
}
}The expression (return) has the type never, since the never type can coerce into any type we can use it in all these places.
In if and while statements it gets coerced into a boolean, in a match statement it gets coerced into anything.
let screaming = match(return){
"aahhh" => true,
_ => false
};Not sure
use std::mem::swap;
fn notsure() {
let mut _x: isize;
let mut _y = (_x = 0) == (_x = 0);
let mut _z = (_x = 0) < (_x = 0);
let _a = (_x += 0) == (_x = 0);
let _b = swap(&mut _y, &mut _z) == swap(&mut _y, &mut _z);
}We have an uninitialised variable _x, we assign _y to (_x = 0) == (_x = 0). (_x = 0) evaluates to the unit type so _y is true. Similar thing with _z and _a, except _z is false since () is not less than itself. _b is also true because swap returns ().
Cant touch this
fn canttouchthis() -> usize {
fn p() -> bool { true }
let _a = (assert!(true) == (assert!(p())));
let _c = (assert!(p()) == ());
let _b: bool = (println!("{}", 0) == (return 0));
}The function p() function returns that a boolean, the assert! macro returns (), so _a and _c are both true.
In the final line _b is assigned to the expression
(println!("{}"),0) == (return 0))The println! returns macro returns (), and (return 0) is ! which gets coerced into () so the expression is valid, this line also returns 0 which makes the function signature valid.
Angry dome
fn angrydome() {
loop { if break { } }
let mut i = 0;
loop {
i += 1;
if i == 1 {
match (continue) {
1 => { },
_ => panic!("wat") }
}
break;
}
}In the first line we immediately exit the loop, because break is a valid expression, which has the type !, we can use it in an if statement.
In the next part we assign i to 0. We increment i in the loop, the if statement will run in the first iteration because i is now 1. We match (continue) which is !, the loop skips to the next iteration, we increment i again so it’s now 2. The if statement doesn’t run so the loop exits and the function returns.
Union
fn union() {
union union<'union> { union: &'union union<'union>, }
}Rust has three categories of keywords:
- Strict keywords, which can only be used in their correct contexts
- Reserved keywords, which have been reserved for future use, but have the same limitations as strict keywords
- Weak keywords, which only have special meaning in certain contexts
union is a weak keyword and is only a keyword when used in a union declaration, allowing us to it to be used in other contexts, such as function names.
Punch card
fn punch_card() -> impl std::fmt::Debug {
..=..=.. .. .. .. .. .. .. .. .. .. .. .. .. ..
..=.. ..=.. .. .. .. .. .. .. .. .. .. ..=.. ..
..=.. ..=.. ..=.. ..=.. .. ..=..=.. ..=..=..=..
..=..=.. .. ..=.. ..=.. ..=.. .. .. .. ..=.. ..
..=.. ..=.. ..=.. ..=.. .. ..=.. .. .. ..=.. ..
..=.. ..=.. ..=.. ..=.. .. .. ..=.. .. ..=.. ..
..=.. ..=.. .. ..=..=.. ..=..=.. .. .. ..=..=..
}In rust .. represents an unbounded range (std::ops::RangeFull) usually used in slices. Similarly ..= represents a range up to and including a value (std::ops::RangeToInclusive). All the different ranges have types which you can see in the std::ops module docs.
Ranges can be combined into whatever amalgamation you would like:
use std::ops::{RangeFull, RangeTo, RangeToInclusive};
let _a: RangeToInclusive<RangeTo<RangeFull>> = ..=.. .. ;All of these range types implement Debug, which satisfies the impl std::fmt::Debug return type.
Monkey barrel
fn monkey_barrel() {
let val: () = ()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=()=();
assert_eq!(val, ());
}In rust an assignment expression consists of a left assignee expression, an equals sign (=) and a right value expression. A tuple pattern can be used an assignee expression, which means it can appear on the left part of an assignment expression. Most of the times we use this to assign destructure values.
let (x,y) = (110.0,50.5);But the tuple can also be empty, which means we’re assigning it to the () type.
let () = ();Because assignments return () we can chain them
let () = ()=()=();Semi’s
fn semisemisemisemisemi() {
;;;;;;; ;;;;;;; ;;; ;;; ;;
;; ;; ;;;; ;;;; ;;
;;;;;;; ;;;;; ;; ;;;; ;; ;;
;; ;; ;; ;; ;; ;;
;;;;;;; ;;;;;;; ;; ;; ;;
}You can add a semi-colon anywhere in a block, which creates an empty statement with an empty value (). So these semi-colons just create a bunch of empty statements.
Useful syntax
fn useful_syntax() {
use {{std::{{collections::{{HashMap}}}}}};
use ::{{{{core}, {std}}}};
use {{::{{core as core2}}}};
}Rust allows grouped use statements to reduce boilerplate. These braces can also be used at the root of the statement, there’s also no limit to the number of braces you can use.
use {std::sync::Arc};
use core::{mem::{{transmute}}};Infinite modules
fn infcx() {
pub mod cx {
pub mod cx {
pub use super::cx;
pub struct Cx;
}
}
let _cx: cx::cx::Cx = cx::cx::cx::cx::cx::Cx;
}We declare a module cx, then we create another sub-module also named cx. The line
pub use super::cx;is re-exporting the module from itself, which means we can now call it recursively. It’s simpler to see if we change the names.
pub mod outer{
pub mod inner{
pub use super::inner;
pub struct Item;
}
}
let _item: outer::inner::Item = outer::inner::inner::inner::Item;Fish fight
fn fish_fight() {
trait Rope {
fn _____________<U>(_: Self, _: U) where Self: Sized {}
}
struct T;
impl Rope for T {}
fn tug_o_war(_: impl Fn(T, T)) {}
tug_o_war(<T>::_____________::<T>);
}The Rope trait has a provided method with one generic U, and it takes in two arguments, one of type Self and another of type U. We make a struct T and implement Rope for it. The tug_of_war function accepts any function or closure that implements Fn(T,T).
The expression <T>::_____________::<T> is a fully qualified function pointer, with T as the generic type (fn(T,T)). Because both parameters are of the same type, we can pass this into the tug_of_war.
Dots
fn dots() {
assert_eq!(String::from(".................................................."),
format!("{:?}", .. .. .. .. .. .. .. .. .. .. .. .. ..
.. .. .. .. .. .. .. .. .. .. .. ..));
}The range syntax (std::ops::RangeFull) implements Debug and gets formatted as "..". So we can chain them to get a string of dots.
u8
fn u8(u8: u8) {
if u8 != 0u8 {
assert_eq!(8u8, {
macro_rules! u8 {
(u8) => {
mod u8 {
pub fn u8<'u8: 'u8 + 'u8>(u8: &'u8 u8) -> &'u8 u8 {
"u8";
u8
}
}
};
}
u8!(u8);
let &u8: &u8 = u8::u8(&8u8);
crate::u8(0u8);
u8
});
}
}Let’s take this apart, we have a macro u8!, which declares a module u8 which declares a function u8 which takes a parameter named u8 of type u8 and returns a reference to a u8.
macro_rules! u8 {
(u8) => {
mod u8 {
pub fn u8<'u8: 'u8 + 'u8>(u8: &'u8 u8) -> &'u8 u8 {
"u8";
u8
}
}
};
}Next we call u8::u8(&8u8) and assign it to a variable (u8). The next line calls crate::u8(0u8), and finally we return the u8 variable from the entire expression.
Continue
fn 𝚌𝚘𝚗𝚝𝚒𝚗𝚞𝚎() {
type 𝚕𝚘𝚘𝚙 = i32;
fn 𝚋𝚛𝚎𝚊𝚔() -> 𝚕𝚘𝚘𝚙 {
let 𝚛𝚎𝚝𝚞𝚛𝚗 = 42;
return 𝚛𝚎𝚝𝚞𝚛𝚗;
}
assert_eq!(loop {
break 𝚋𝚛𝚎𝚊𝚔 ();
}, 42);
}These use unicode monospace characters, instead of normal ASCII characters, for identifiers, which don’t break rust’s rules of using keywords as identifiers.
Fishy
fn fishy() {
assert_eq!(
String::from("><>"),
String::<>::from::<>("><>").chars::<>().rev::<>().collect::<String>()
);
}Rust uses the turbo fish syntax when adding generics and lifetimes. We can use empty angle brackets to explicitly specify empty generics.
Special characters
fn special_characters() {
let val = !((|(..):(_,_),(|__@_|__)|__)((&*"\\",'🤔')/**/,{})=={&[..=..][..];})//
;
assert!(!val);
}Let’s decode the right expression:
let val = &[..=..][..];We create a reference to a slice containing a range &[..=..], then we take a full slice of that.
Now for the left expression:
let val = (|(..):(_,_),(|__@_|__)|__)((&*"\\",'🤔')/**/,{});We have a closure with two arguments, the first argument is a tuple, with auto-inferred types.
let val = |(..):(_,_)|{};The second argument is a closure which has an at binding, the variable __ is bound to a wildcard pattern (_), which will match anything.
let val = |(..):(_,_),(|__@_|__)|{};Then we immediately call that closure, passing in a tuple with a string and a char, and an empty block.
let val = (|(..):(_,_),(|__@_|__)|)((&*"\\",'🤔'),{})Match
fn r#match() {
let val: () = match match match match match () {
() => ()
} {
() => ()
} {
() => ()
} {
() => ()
} {
() => ()
};
assert_eq!(val, ());
}This is just matching nested match statements.
Match nested if
fn match_nested_if() {
let val = match () {
() if if if if true {true} else {false} {true} else {false} {true} else {false} => true,
_ => false,
};
assert!(val);
}This is a match guard with nested if statements.
Function
fn function() {
struct foo;
impl Deref for foo {
type Target = fn() -> Self;
fn deref(&self) -> &Self::Target {
&((|| foo) as _)
}
}
let foo = foo () ()() ()()() ()()()() ()()()()();
}The Deref trait is used when a type can be implicitly coerced into another type, it’s usually used by smart pointers so they can be implicitly used at the underlying type.
We implement Deref for foo into a function pointer that returns foo, which means we can call that foo again recursively.
Bathroom stall
fn bathroom_stall() {
let mut i = 1;
matches!(2, _|_|_|_|_|_ if (i+=1) != (i+=1));
assert_eq!(i, 13);
}
In a match arm multiple patterns can be matched in one arm, separated by |.
let foo = 'a';
match foo {
'a'..'c'|'x'..'z' => {}
_ => {}
}The matches! macro has the same syntax as a match statement so we can also chain multiple patterns, even if those are wildcard patterns.
matches!((),_|_|_|_|_|_)matches!(2, _|_|_|_|_|_ if (i+=1) != (i+=1));We have six different patterns here, which all do the same thing: we check if i +=1 != i += 1, which increments it twice, so each iteration is incrementing i by 2. 6 x 2 = 12 plus 1 (the initial value) and the final value is 13 so the assert assert_eq!(i,13) is true. The match!(2,..) doesn’t panic because it’s a wildcard pattern so any value could have been used. The if statement is always going to be false because the right expression will always be one more than the left so it will run until all the patterns have been tried.
Closure matching
fn closure_matching() {
let x = |_| Some(1);
let (|x| x) = match x(..) {
|_| Some(2) => |_| Some(3),
|_| _ => unreachable!(),
};
assert!(matches!(x(..), |_| Some(4)));
}x is a closure that takes in a parameter with an unspecified type, which will be inferred through its usage. Next we match x(..) which makes the type of the closure RangeFull. It looks like we’re matching closures but it’s really just multiple wildcard patterns.
The numbers also don’t matter even though it seems as though the function is being incremented each time, since it’s a wildcard anything would match.
Return already
fn return_already() -> impl std::fmt::Debug {
loop {
return !!!!!!!
break !!!!!!1111
}
}The break expression is repeatedly applying a not operation on an integer, while the return expression is also repeatedly applying a not operation on the break expression.
Fake macros
fn fake_macros() -> impl std::fmt::Debug {
loop {
if! {
match! (
break! {
return! {
1337
}
}
) {
}
} {
}
}
}Let’s isolate the return statement:
fn fake_macros() -> impl std::fmt::Debug{
return! { 1337 }
}This is doing a not operation on the inner expression. Next we wrap that expression in a loop.
fn fake_macros() -> impl std::fmt::Debug{
loop {
break! {
return! {
1337
}
}
}
}The break!{ } is also doing a not operation on the return! { 1337 }, which has the type !. Now the functions return type is inferred from both the loop and the return statement. Divergent function?
Next we wrap everything inside the loop in a match statement
fn fake_macros() -> impl std::fmt::Debug{
loop {
match!(
break! {
return! {
1337
}
}
){
}
}
}We don’t have to add any patterns to the match statement since we’re matching never. And finally we wrap this in an if statement.
fn fake_macros() -> impl std::fmt::Debug{
loop {
if! {
match! (
break! {
return! {
1337
}
}
)
} {
}
}
}So to sum it up:
return! { 1337 }makes the return type of the function ani32, which implementsDebugbreak! { ... }makes the return type of the loop!, because of the innerreturn, which also implementsDebug- We match the break statement and leave out the patterns since it is
! - Wrap the match statement in an if statement