Summary of some minor problems encountered while writing Rust

Automatic dereference (dereference coercion)

struct Car {
    width: u32,
    height: u32,
}

fn main() {
    let mut car1 = Car {
        width: 100,
        height: 200,
    };
    
    let car2 = &mut car1;
    car2.width = 200;
    println!("width: {}", car2.width);

    (*car2).width = 300;
    println!("width: {}", car2.width);
}

car2 is a mutable reference to car1. When we try to change width by car2, Rust will automatically dereference car2 to get the car1 instance and then modify the value of width. So car2.width = 200; is equal to (*car2).width = 300;.

But things are not so simple. We can see the following example. The first part is right. But for the second part, if we use y = 10, we will get an error. This is because the automatic deference doesn’t occur on primitive data types. So we must use * explicitly.

let car2 = &mut car1;
car2.width = 200;
println!("width: {}", car2.width);

let mut x = 5;
let y = &mut x;
*y = 10;
println!("y: {}", y);

Why? This may be a design consideration. For struct, it has many elements, so automatic dereference can make code simpler and more readable. But for primitive data types, if we don’t use *, it can make people misunderstand that y is reassigned rather than that the value which y refers to is changed.

And I need to add some things. * doesn’t get the ownership and just can calculate or other things.

There is another situation. It can work but this isn’t automatic dereference. This depends on + which has an Add trait.

let a = 12;
let b = &a;
let c = &a;
println!("{}", b + c);

That’s not all. I need to add something about deref coercion. The following code has the same effect and the first includes the second. This is because deref coercion, which can convert &Vec<&str> to &[&str]. [&str] is a slice, which represents a partial view of an array or other continuous data structures. So it can include Vec<&str>.

pub fn capitalize_words_string(words: &[&str]) -> String {}
pub fn capitalize_words_string(words: &Vec<&str>) -> String {}

parse()

It can change the type of the value.

We must specify its type when using parse().

let a: i32 = (&a[1..]).parse().unwrap();

When this value will return, we don’t need to specify its type. Because the function signature is already specified. But I think we should still do this, which makes the code clearer.

collect()

It can take the iterator value and convert them to Vec or other things.

We must specify its type when using collect(). This is because that it can handle not only Vec but also other things like HashMap.

let cells: Vec<_> = first_second.spilt('_').collect();

Using mutable objects continuously

According to borrow rules, we can’t have mutable and immutable borrow at the same time. This value is mutable borrow from self. And self.double_get continues to borrow self. So this is a wrong way, which could lead to the data race.

for (key, value) in self.real_data.iter_mut() {
  *value = self.double_get("get".to_string() + key.as_str());
}

Other example:

match boxes.get_mut(other) {
  Some(v) => {
	parse_command(v, turtle, variables, boxes);
  }
  None => {
	return Err(logo::CommandError::WrongCommandErr(other.to_string()));
  }
}

unwrap() VS except() VS ?

For Result:
unwrap(): If Result is Err(), the program directly panic(); If Result is Ok(), it will return the value in Ok().
except(): It is similar to unwrap(), except that it can panic() custom errors.
?: If Result is Err(), it will return Err(); If Result is Ok(), it will return the value in Ok().

For Option():
unwrap(): If Option is None, the program directly panic(); If Option is Some(), it will return the value in Ok().
except(): It is similar to unwrap(), except that it can panic() custom errors.
?: If Option is None, it will return None; If Option is Some(), it will return the value in Some().

Of course, we can also use match for more detailed processing.

match VS if let VS let if

let f = match f {
	Ok(file) => file,
	Err(error) => {
		panic!("Problem opening the file: {:?}", error)
	},
};

if let 1 = 1 {
	println!("1");
}

let a = if true {  // This must be a bool value
	1
} else {
	2
};

Why does Rust prohibit declaring multiple mutable references or mutable and immutable references

There could be many reasons. But I just explain two.
For the data race, suppose you declare one mutable reference and an immutable reference. You use mutable references to change the original data. When you use an immutable references to read the same data, you will find that the data has changed. But this is immutable reference.

For memory security, suppose you declare one mutable reference and one immutable reference. If you delete all the data by mutable reference. If you use immutable reference to read the data, they will have the problem.

The mut and &mut in the function signature

fn conmand(mut turtle: Turtle, variables: &mut variables)

iter() VS iter_mut() VS into_iter()

// iter() can get the immutable iterator of a collection.
let vec = vec![1, 2, 3, 4];
for value in vec.iter() {
    println!("{}", value);
}

// iter_mut() can get the mutable iterator of a collection. It applies to change a collection.
let mut vec = vec![1, 2, 3, 4];
for value in vec.iter_mut() {
    *value += 10;
}

// into_iter() can get the ownership.
let vec = vec![1, 2, 3, 4];
let new_vec: Vec<_> = vec.into_iter().map(|x| x * 10).collect();

The struct of the struct

When we use the struct like the following, we must use VariablesCommand::Make{name, value}. That means we can’t use directly a specific value.

pub enum VariablesCommand {
    Make { name: String, value: f32 },
    AddAssign { name: String, value: f32 },
}

the pattern matching of map()

As you can see, x is &&strbut &x is &str. Actually, this &is pattern matching rather than reference.

the usage of slice

Whether it’s &str or String, you have to use it by & when you use slice.

let a = "abcdedf";
let b = "abcdedf".to_string();
let c = &a[1..];
let d = &b[1..];
println!("c: {}", c);
println!("d: {}", d);

points to note when using Hashmap()

First, when you use get() or remove(), you must use & like the following. Because you don’t have to take the ownership and the reference is more reasonable.

hashmap.remove(&i);

Second, the key of the Hashmap doesn’t support the float due to some security reasons. Java or Python supports that.

points to note when using Vec()

Vec() uses usize.

points to note when using match

The following example is wrong because the elements of the match need to have the same return value.

let result = match value {
  0 => "zero",
  1 => 1,
  _ => "many",
};

one example understanding trait

pub enum Die {
    D4,
    D6,
    D8,
    D10,
    D12,
    D20,
}

pub struct Coin;

pub trait Rollable {
    fn roll(&self) -> String;
}

impl Rollable for Die {
    fn roll(&self) -> String {
        let max = match self {
            Die::D4 => 4,
            Die::D6 => 6,
            Die::D8 => 8,
            Die::D10 => 10,
            Die::D12 => 12,
            Die::D20 => 20,
        };
        get_random_value(max).to_string()
    }
}

impl Rollable for Coin {
    fn roll(&self) -> String {
        let max = match self {
            Coin => 2,
        };
        get_random_value(max).to_string()
    }
}

pub fn roll<T: Rollable> (item: T) -> u8 {
    item.roll().parse().unwrap()
}

the passing of reference in the struct

Why does person.greetings need &? person: &Person indicates that person is a reference, which is immutable. But person.greetings will get Box<dyn Greeting> rather than &Box<dyn Greeting>. This is because person: &Person can only indicate that person.greetings is immutable. But it can’t indicate that person.greetings is &. So the mutability is passable but the reference is not.

struct Person {
    name: String,
    greetings: Vec<Box<dyn Greeting>>,
}
fn speak_all_greetings(person: &Person) {
    println!("{} says:", person.name);
    for greeting in &person.greetings {
        greeting.greet();
    }
}

dyn

Why use dyn? This indicates that the interface (trait) is polymorphic, which means it can reference different types of instances that implement a particular trait.

struct English;
struct Spanish;
struct French;

trait Greeting {
    fn greet(&self);
}

impl Greeting for English {
    fn greet(&self) {
        println!("Hello!");
    }
}

impl Greeting for Spanish {
    fn greet(&self) {
        println!("Hola!");
    }
}

impl Greeting for French {
    fn greet(&self) {
        println!("Bonjour!");
    }
}

struct Person {
    name: String,
    greetings: Vec<Box<dyn Greeting>>,
}

the case without

<T> already apeared after the impl. So new doesn’t need to declare <T>.

impl<T> Point<T> {
    pub fn new (x: T, y: T) -> Self {
        Point { x, y }
    }
}

the signature of <T> and Box for closure

f is a closure. Why can’t we just use FnOnce(i32) -> i32 instead of <T>. Because when declaring the signature, Rust needs to know its exact size, and Fn(i32) -> i32 size is uncertain. So actually we can use Box which is dynamic distribution but it will have performance loss. The following way is static distribution.

And why we can’t use T: Fn(i32) -> i32 + Fn(i64) -> i64? Because Fn stands for closure and a closure just can have one trait about closure.

impl MyOption {
    fn map<T>(self, f: T) -> MyOption
    where
        T: FnOnce(i32) -> i32,
    {
        match self {
            MyOption::Some(x) => MyOption::Some(f(x)),
            MyOption::None => MyOption::None,
        }
    }
}

the directory structure of the package

Why Rust uses mod.rs? This approach is more flexible than Java’s, which means it can have more other operations.

src/
 |__ bin/
      |__ main.rs
 |__ lib/
      |__ mod.rs
      |__ translate.rs
      |__ music/
           |__ mod.rs
           |__ flac.rs
           |__ mp3.rs
 |__ Cargo.toml