41
\$\begingroup\$

While Rust is very, very rarely competitive in code golfing competitions (Java is often shorter), it can still be fun to golf in. What are some tricks for making Rust code shorter?

Please post only one tip in each answer.

\$\endgroup\$
0

26 Answers 26

27
\$\begingroup\$

Use closures instead of functions

A closure:

|n|n+1

is shorter than a function:

fn f(n:i32)->i32{n+1}

Closures longer than one statement need braces but are still far shorter than a function.

\$\endgroup\$
21
\$\begingroup\$

Converting &str to String

Never do these:

s.to_string()  // 13 bytes
s.to_owned()   // 12 bytes

This is always shorter:

s.repeat(1)    // 11 bytes

If s is a string literal:

format!(s)     // 10 bytes

For example: use format!("") instead of String::new() to save 2 bytes.

If type inference works:

s.into()       // 8 bytes
\$\endgroup\$
1
  • \$\begingroup\$ For when you're converting an iterator of &str to an iterator of String, .map(String::from) beats any of these save .map(|s|s.into()) (which wins by one character) by virtue of avoiding a closure, and unlike .map(|s|s.into()), it works without relying on type-inferencing. \$\endgroup\$ Commented Apr 9 at 18:52
19
\$\begingroup\$

Avoid .iter().enumerate()

Let's say you have some x that implements the IntoIterator Trait and you need to call a function f that takes the index of an element and a reference to it. The standard way of doing this is

x.iter().enumerate().map(f)

instead you can do

(0..).zip(x).map(f)

and save yourself not only the unusually long enumerate but also the call to iter!

\$\endgroup\$
4
  • \$\begingroup\$ Now I want to do this in regular Rust code. \$\endgroup\$
    – Joonazan
    Commented Jun 22, 2022 at 21:14
  • \$\begingroup\$ Unfortunately enumerate produces shorter assembly. godbolt.org/z/hnee1GK34 But zip is better in every other way. It doesn't require the index to be usize. \$\endgroup\$
    – Joonazan
    Commented Jun 22, 2022 at 21:22
  • \$\begingroup\$ it appears that they produce the same assembly starting with rust 1.60 \$\endgroup\$ Commented Jan 1, 2023 at 8:57
  • \$\begingroup\$ For the specific case of String/&str where you want to iterate the characters, s.char_indices() exists and beats s.chars().enumerate() (by 5), (0..).zip(s.chars()) (by 4), and s.chars().zip(0..) (by 2). \$\endgroup\$ Commented Apr 9 at 18:55
16
\$\begingroup\$

If you need many mutable variables, it can waste a lot of space declaring them and initializing them, since each requires the mut keyword and you can't do anything like a=b=c=0. A solution then is to declare a mutable array

let mut a=[0;5];

You spend 3 extra bytes each time you use them vs. a normal variable:

a[0]+=1;

but it can often still be worth it.

Using tuples for the same trick is often an even better option:

let mut t=(1,4,"this", 0.5, 'c');

This has the advantage of saving a byte on each use vs. the array:

t.0=2

It also lets them be of different types. On the downside, it requires more characters to initialize.

\$\endgroup\$
1
  • \$\begingroup\$ "you can't do anything like a=b=c=0" Except you can. You need to repeat the mut for each variable, but you can initialize them all to 0 at once, no matter how many there are, where [0;#] is a little longer. This doesn't mean the technique is useless or anything, just that it's harder for it to win, given the initializer costs more, and each use costs more, so you only save on the declaration. \$\endgroup\$ Commented Apr 19 at 17:33
16
\$\begingroup\$

Use @ to define multiple variables to the same value

@ is meant to define an alias for pattern matching (like destructuring an array but also getting the original array at the same time). But it works with plain variables too.

let a=2;let b=2; // 16 bytes
let(a,b)=(2,2);  // 15 bytes
let a@b=2;       // 10 bytes

It works in destructuring too:

let a=1;let b=1;let c=2;let d=2; // 32 bytes
let(a,b,c,d)=(1,1,2,2);          // 23 bytes
let(a@b,c@d)=(1,2);              // 19 bytes
\$\endgroup\$
1
  • 4
    \$\begingroup\$ If they need to be mutable (otherwise, what's the point), both variables need the mut keyword, e.g. let mut a@mut b=1; \$\endgroup\$
    – Jo King
    Commented Sep 24, 2022 at 1:24
13
\$\begingroup\$

When using string formatting for example with print!(), one can use both numbered and unnumbered formatters to save one byte per item to format:

Best shown with an example:

fn main() {
    print!(
        "{}{}{}. Yes, {0}{}{2}. All you other{1}{2}s are just imitating.",
        "I'm", " Slim", " Shady", " the real",
    );
}

Which outputs:

I'm Slim Shady. Yes, I'm the real Shady. All you other Slim Shadys are just imitating.

So the unnumbered formatters will get assigned to the items in order, this allows you to skip the index on them. Note that you can only use one unnumbered formatter per item to format, after that it will get used up.

\$\endgroup\$
11
\$\begingroup\$

Skipping trailing semicolons

In functions returning (), where the last expression is also of type (), you don't need the trailing ;:

fn main(){print!("Hello, world!")}
\$\endgroup\$
11
\$\begingroup\$

Replace all filter_map with flat_map

It works because:

  • filter_map takes a FnMut(Self::Item) -> Option<B>

  • flat_map takes a FnMut(Self::Item) -> impl IntoIterator

  • Option implements IntoIterator, which will return an iterator with one item for Some and empty iterator for None

\$\endgroup\$
9
\$\begingroup\$

When working with strings with newlines in them, you save one byte if you use a literal line break in the source code vs having a \n in the string and/or using println.

print!("Hello
World!
");

is 2 bytes less than:

println!("Hello\nWorld!");
\$\endgroup\$
1
  • 10
    \$\begingroup\$ Save another byte by using print with literal line break instead of println \$\endgroup\$
    – JayXon
    Commented Aug 2, 2020 at 1:44
8
\$\begingroup\$

Reading lines

After considering various things, I think that is generally the shortest way to retrieve a line. The line has a newline, it can be removed by trimming (.trim()) or if that cannot be done by slicing.

let y=&mut"".into();std::io::stdin().read_line(y);

For multiple lines, lines iterator can be used, the iterated line doesn't end in newline then. A glob import is needed to import BufRead, needed for lines to be available for StdinLock<'_> type.

use std::io::*;let y=stdin();y.lock().lines()
\$\endgroup\$
8
\$\begingroup\$

When using whole number floating point numbers, you can omit the trailing .0 to save one byte.

let a=1.

is 1 byte less than:

let a=1.0
\$\endgroup\$
1
  • 2
    \$\begingroup\$ When using FP numbers that have an integer part of 0, you can omit the leading zero as well; .0, .5 are both valid \$\endgroup\$
    – Starwort
    Commented Jun 6, 2023 at 20:33
8
\$\begingroup\$

Use ..=

Replace all ..i+1 with ..=i to save 1 byte, works on both array index a[i..=i] and range (0..=n)

\$\endgroup\$
2
  • 3
    \$\begingroup\$ clippy actually has a warning for this \$\endgroup\$
    – betseg
    Commented Feb 26, 2020 at 5:20
  • 3
    \$\begingroup\$ The reverse is better for numbers \$\endgroup\$
    – SuperPizz
    Commented Sep 30, 2021 at 20:08
8
\$\begingroup\$

Do-while loops

Rust does not have a C-style do-while loop:

do{do_work();}while(condition);

But blocks are also expressions in Rust, so you can write:

while{do_work();condition}{}
\$\endgroup\$
7
\$\begingroup\$

Declare multiple variables using pattern matching

(Somewhat of an extension of https://codegolf.stackexchange.com/a/99124/97519)

Using multiple let statements:

let a=x;let b=y;
let a=x;let b=y;let c=z;

Using a tuple match saves 1 char for 2 variables and 3 more chars for ever variable after:

let(a,b)=(x,y);
let(a,b,c)=(x,y,z);

Similarly, you can use array patterns for initializing multiple mutable variables to the same value (this works with immutable variables too but is pretty useless):

let[mut a,mut b]=[x;2];
\$\endgroup\$
2
  • 2
    \$\begingroup\$ Why is the immutable version useless? \$\endgroup\$
    – Wheat Wizard
    Commented Jul 31, 2020 at 20:48
  • 10
    \$\begingroup\$ @AdHocGarfHunter why would you want to initialize two immutable variables to the same value? \$\endgroup\$
    – Doorknob
    Commented Jul 31, 2020 at 22:10
6
\$\begingroup\$

Use closure parameter to define variable instead of let

{let(a,b)=(A,B);X}

can be replaced by

(|a,b|X)(A,B)

to save 5 bytes if the type of a and b can be inferred, it usually works when they are integers or are passed to another function directly in X.

This works well on 2 or more variables, but for a single variable, this could still save 2 bytes if it allows you to get rid of the {}.

\$\endgroup\$
6
\$\begingroup\$

Prefer todo! over panic!

todo! is shorter than panic! by 1 byte, so prefer to use it if you need to panic and exit the program. It's also shorter than print! by a byte, so it can shorten your program if you can output to stderr instead of stdout.

todo!("optional message for stderr")
\$\endgroup\$
4
\$\begingroup\$

Use the ? operator to unwrap infallible results. ? is used for error propagation in conjunction with the result and option types, and per this meta post returning them with infallible results is a standard output method. For example, the following code summing the value of a string when interpreted in base a and b, with ?:

|s,a,b|Ok(i32::from_str_radix(s,a)?+i32::from_str_radix(s,b)?)

and without:

|s,a,b|i32::from_str_radix(s,a).unwrap()+i32::from_str_radix(s,b).unwrap()

As you can see, with ? it is much shorter, and it works with options as well, although often for smaller gains.

\$\endgroup\$
2
  • \$\begingroup\$ The unwrap in this example is inside a map, usually filter_map is used in this case, but for code golf, just use flat_map instead: |n|n.to_string().chars().flat_map(|c|c.to_digit(10)).sum(), see my answer for why flat_map works \$\endgroup\$
    – JayXon
    Commented Dec 17, 2020 at 9:26
  • \$\begingroup\$ @JayXon you're right, I have a better example up now. \$\endgroup\$
    – Aiden4
    Commented Dec 17, 2020 at 16:01
4
\$\begingroup\$

if let can sometimes be shorter than match

Using match:

match a{M(v)=>expr,_=>other}

Using if let costs 2 bytes if there are only 2 branches and 1 is wildcard:

if let M(v)=a{expr}else{other}

More effective when your match needed {} anyways, like if the types don't match. For example:

match a{M(v)=>{expr;}_=>{other;}}

vs.:

if let M(v)=a{expr;}else{other;}

saves 1 byte.

\$\endgroup\$
0
3
\$\begingroup\$

Loops are sometimes shorter than iterators

Even with the extra curly braces, in some cases it can be shorter to declare a mutable variable and use a loop instead of using Iterator methods.

For example, to do something with every third element of a slice and then return the last index:

|n:&[_]|{let mut i=0;while i<n.len(){println!("{}",n[i]);i+=3}i-3}
|n:&[_]|(0..=n.len()/3).map(|i|{println!("{}",n[i*3]);i*3}).last().unwrap()
\$\endgroup\$
1
  • \$\begingroup\$ Using any(|x|f(x)>()) is generally shorter than using last() though you are correct loops can sometimes be even shorter \$\endgroup\$
    – mousetail
    Commented Aug 1, 2022 at 12:27
3
\$\begingroup\$

(0..).scan can be shorter than std::iter::successors

If you want to make endless sequences that are based on previous state:

std::iter::successors(Some(a),|x,_|Some(/*something*/))

This will be shorter by a character:

(0..).scan(a,|x,_|{let y=*x;*x=/*something*/;Some(y)})

If you don't mind dropping the first case, it can be even shorter:

(0..).scan(a,|x,_|{*x=/*something*/;Some(*x)})
\$\endgroup\$
3
\$\begingroup\$

Get the nth element of a String or &str slice

In rust you can't directly index a string or slice. If you want to build a string from indexes of another string you would typically do this:

s.chars().nth(n).unwrap() // char
s.bytes().nth(n).unwrap() // u8

Shorter is this, but it will depend s so won't work if s is generated from a expression:

s.as_bytes()[n]as char // char, supports only single byte characters
s.as_bytes()[n] // u8

However, if you are ok with a &str (like if you want to collect into a String or concatenate with a existing String). This also supports only single byte characters.

&s[n..][..1] // &str
&s[n..=n] // &str, if `n` is a single variable or sufficiently short expression

If your expression is a literal you can use byte string instead to index it directly:

b"hello world"[n] // u8
b"hello world"[n]as char // char
\$\endgroup\$
2
  • 1
    \$\begingroup\$ &s[n..][..1] and &s[n..n+1] have the same "supports only single byte characters" caveat; slicing strings at non-char boundaries leads to a panic. Still a great tip if that holds though! \$\endgroup\$ Commented Aug 1, 2022 at 19:37
  • 1
    \$\begingroup\$ s.as_bytes() does not consume s, s.into_bytes() does. also no space needed between ] and as, and &s[n..=n] saves one \$\endgroup\$
    – JayXon
    Commented Aug 2, 2022 at 21:42
2
\$\begingroup\$

scan and fold on Iterators can be used for stateful iteration

If your closure consists of variable declarations and iterating over something while keeping an internal state, you might be able to save some bytes by using scan or fold instead of a for loop:

|i:&str|{let(mut a,mut b,mut c)=(1,2,0);for x in i.chars(){/*AAA*/;/*BBB*/;c=/*CCC*/}c}

scan is good for handling many mutable variables, at the cost of some operations (such as assignment) requiring dereferencing with * and needing to collect the final value using a method like last:

|i:&str|i.chars().scan((1,2),|(a,b),x|{/*AAA*/;/*BBB*/;Some(/*CCC*/)}).last()

fold can also be used in a similar manner by passing values in the accumulator:

|i:&str|i.chars().fold((1,2,3),|(a,b,c),x|(/*AAA*/,/*BBB*/,/*CCC*/)).2
\$\endgroup\$
1
\$\begingroup\$

Using slice patterns to get both an array and variables with the same value

It is a common trick to use array slicing to compactly assign many variables to the same value:

let[a,b,c]=[0;3]
// vs
let(a,b,c)=(0,0,0)
// or
let a=0;let b=0;let c=0;

I've never really seen this used on arrays, even though it also saves bytes there:

let[a,..@b]=[0;99]; // store "the rest" of the slice in b
// vs
let(a,b)=(0,[0;98]);
// or
let a=0;let b=[0;98];

Saves just one byte so may not be worth it if it increases the array size by one byte.

It works for mutable variables too:

let[mut v,mut u@..]=[0;99];
// vs
let(mut u,mut v)=(0,[0;98]);
\$\endgroup\$
1
\$\begingroup\$

Use for loops or map to unwrap known Some/Ok values when ? isn't an option

When it's too expensive to declare a return value that would allow you to use ? to handle known Some/Ok values, and scope is not an issue, you can use the fact that Option and Result implement IntoIterator and a custom version of map to shorten variable declarations by using a for loop or their map method.

// Unwrapping Some
if let Some(x)=returnsoption(){}  // Idiomatic fallible unwrap
let x=returnsoption().unwrap();   // Unconditional unwrap saves 1, and avoids nested scope
for x in returnsoption(){}        // for loop saves five more, as cost of nested scope
returnsoption().map(|x|{});       // One longer with map if {} needed, and even nastier, scope-wise, and usually needs semi-colon
returnsoption().map(|x|);         // Map one shorter when {} not needed and scope issues not a problem; can be two shorter in cases where semi-colon not needed

// Unwrapping Ok
if let Ok(x)=returnsresult(){}  // Idiomatic fallible unwrap
let x=returnsresult().unwrap(); // Unconditional unwrap avoids nested scope, but costs 1
for x in returnsresult(){}      // for loop saves four over next best option
returnsresult().map(|x|{});        // One longer with map if {} needed, and even nastier, scope-wise, and usually needs semi-colon
returnsresult().map(|x|);          // Map one shorter when {} not needed and scope issues not a problem; can be two shorter in cases where semi-colon not needed

This even saves something when it means you need to use the turbofish, e.g., if the type of x can't be deduced:

let x:u32=s.parse().unwrap();
// vs.
for x in s.parse::<u32>(){}

saves two characters over let, even with the turbofish adding three characters, though map remains better if the scope issues aren't a problem:

// If scope issues aren't a problem, saves 2-4/3-5 (depending on need for semi-colon) more:
s.parse().map(|x:u32|{});
s.parse().map(|x:u32|);

All the uses of map will produce a compile-time warning due to the result being unused, so avoid them if that's a problem.

\$\endgroup\$
1
\$\begingroup\$

Fallible unpacking can let you avoid filtering

If you have an iterable input, e.g. command-line arguments, where some values are known to be bad (the program name is not useful), and others can be split into multiple discrete values, the normal solution to loop over .skip(n) to bypass the bad values (e.g. .skip(1) to skip the program name in the arguments), split/collect the rest to a Vec, and use the values by indexing (unless they're referred to many times, it's not worth the cost to unpack). For example, imagine a program that receives four numbers at a time, comma-separated as command-line arguments, and must compute a value using each value individually. The simple solution (using the flat_map trick to avoid a level of unwraping or use of filter_map) is:

fn main(){
  for a in std::env::args().skip(1){
     let v:Vec<u32>=a.split(',').flat_map(str::parse).collect();
     println!("{},{}",v[0]+v[1],v[2]+v[3])
  }
}

And that's not bad. But the .skip(1) costs eight characters, and all the indexing costs 12 more, for a total of 20 characters we'd like to avoid. By using if let with slice-unpacking, we can imply the skip (assuming the program name isn't four numbers separated by embedded commas), and unpack to individual names, while still saving characters:

fn main(){
  for a in std::env::args(){
    let v:Vec<u32>=a.split(',').flat_map(str::parse).collect();
    if let[m,n,o,p]=a.split(',').flat_map(str::parse).collect::<Vec<u32>>()[..]{
      println!("{},{}",m+n,o+p)
    }
  }
}

The difference here is pretty trivial (the second version shaves 8+12=20 characters by avoiding .skip(1) and indexing four times, but it costs 18 more to wrap in if let, use a turbofish on collect instead of letting type-inference handle it the pattern itself is longer (let[m,n,o,p] vs. let v), and [..] is needed to convert from Vec to slice for the pattern matching.

But:

  1. Unpacking more values only costs two characters more per name, and saves three characters the first time said value is used, and if you need to use any given value more than once, it saves three more characters per use, etc.
  2. Frequently, type-inference can figure out the type of the matched values, and that will allow you to replace iterator.collect::<Vec<u32>>() with Vec::from_iter(iterator), saving eight characters and making it much quicker to reach the break even point.

If the iterator had no invalid values to skip, this wouldn't be worth as much (the .skip(1) is 40% of the savings here), but the general pattern can be useful even then if you have enough things to unpack, and use them often enough.

Try both out on the Rust Playground

\$\endgroup\$
0
\$\begingroup\$

Abuse reduce to do something special with the first element of an iterator

It's very inconvenient when the first element of an iterator is a special case, since all ways to deal with it are quite long.

You can use .next(): (~73 bytes)

let mut a=m.iter();let f=a.next().unwrap();for i in a{do_something(f,i)};

A lot shorter is using collect. Depending on what you need later with the array this may be the shortest choice. (~62 bytes)

let a:Vec<_>=m.collect();for i in a[1..]{do_something(a[0],i)}

However, a pro move I saw JoKing use was to use reduce. reduce already handles the first case special so all that is necessary is to just return it and don't modify it. This is (~36 bytes)

m.reduce(|f,a|{do_something(f,a);f})
\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.