X-Git-Url: https://git.ralfj.de/rust-101.git/blobdiff_plain/816ab35bec9dec1571988fcf97e57a38a32f5ff5..01999d0e811a07d9089c1b2158a6602ce6890f2f:/src/part02.rs?ds=inline diff --git a/src/part02.rs b/src/part02.rs index 4e0fa6b..bd8abf0 100644 --- a/src/part02.rs +++ b/src/part02.rs @@ -1,106 +1,127 @@ // Rust-101, Part 02: Generic types, Traits // ======================================== -use std; +//@ Let us for a moment reconsider the type `NumberOrNothing`. Isn't it a bit annoying that we +//@ had to hard-code the type `i32` in there? What if tomorrow, we want a `CharOrNothing`, and +//@ later a `FloatOrNothing`? Certainly we don't want to re-write the type and all its inherent methods. -// Let us for a moment reconsider the type `NumberOrNothing`. Isn't it a bit -// annoying that we had to hard-code the type `i32` in there? What if tomorrow, -// we want a `CharOrNothing`, and later a `FloatOrNothing`? Certainly we don't -// want to re-write the type and all its inherent methods. +// ## Generic datatypes -// The solution to this is called *generics* or *polymorphism* (the latter is Greek, -// meaning "many shapes"). You may know something similar from C++ (where it's called -// *templates*) or Java, or one of the many functional languages. So here, we define -// a generic type `SomethingOrNothing`. -enum SomethingOrNothing { +//@ The solution to this is called *generics* or *polymorphism* (the latter is Greek, +//@ meaning "many shapes"). You may know something similar from C++ (where it's called +//@ *templates*) or Java, or one of the many functional languages. So here, we define +//@ a generic type `SomethingOrNothing`. +pub enum SomethingOrNothing { Something(T), Nothing, } -use self::SomethingOrNothing::{Something,Nothing}; -// What this does is to define an entire family of types: We can now write -// `SomethingOrNothing` to get back our `NumberOrNothing`, but we -// can also write `SomethingOrNothing` or even `SomethingOrNothing>`. -// In fact, such a type is so useful that it is already present in the standard -// library: It's called an *option type*, written `Option`. -// Go check out its [documentation](http://doc.rust-lang.org/stable/std/option/index.html)! -// (And don't worry, there's indeed lots of material mentioned there that we did not cover yet.) +// Instead of writing out all the variants, we can also just import them all at once. +pub use self::SomethingOrNothing::*; +//@ What this does is to define an entire family of types: We can now write +//@ `SomethingOrNothing` to get back our `NumberOrNothing`. +type NumberOrNothing = SomethingOrNothing; +//@ However, we can also write `SomethingOrNothing` or even `SomethingOrNothing>`. +//@ In fact, such a type is so useful that it is already present in the standard library: It's called an +//@ *option type*, written `Option`. Go check out its [documentation](http://doc.rust-lang.org/stable/std/option/index.html)! +//@ (And don't worry, there's indeed lots of material mentioned there that we did not cover yet.) -// **Exercise**: Write functions converting between `SomethingOrNothing` and `Option`. You will have to use -// the names of the constructor of `Option`, which you can find in the documentation I linked above. -// -// Here's a skeleton for your solution, you only have to fill in the function bodies. -// (`panic!` is, again, a macro - this one terminates execution when it is reached). -// -// Notice the syntax for giving generic implementations to generic types: Think of the first `` -// as *declaring* a type variable ("I am doing something for all types `T`"), and the second `` as -// *using* that variable ("The thing I do, is implement `SomethingOrNothing`"). +// ## Generic `impl`, Static functions +//@ The types are so similar, that we can provide a generic function to construct a `SomethingOrNothing` +//@ from an `Option`, and vice versa. +//@ +//@ Notice the syntax for giving generic implementations to generic types: Think of the first `` +//@ as *declaring* a type variable ("I am doing something for all types `T`"), and the second `` as +//@ *using* that variable ("The thing I do, is implement `SomethingOrNothing`"). +//@ +// Inside an `impl`, `Self` refers to the type we are implementing things for. Here, it is +// an alias for `SomethingOrNothing`. +//@ Remember that `self` is the `this` of Rust, and implicitly has type `Self`. impl SomethingOrNothing { fn new(o: Option) -> Self { - panic!("Not yet implemented.") + match o { None => Nothing, Some(t) => Something(t) } /*@*/ } fn to_option(self) -> Option { - panic!("Not yet implemented.") + match self { Nothing => None, Something(t) => Some(t) } /*@*/ } } -// Inside an `impl`, `Self` refers to the type we are implementing things for. Here, it is -// an alias for `SomethingOrNothing`. -// Remember that `self` is the `this` of Rust, and implicitly has type `Self`. -// -// Observe how `new` does *not* have a `self` parameter. This corresponds to a `static` method -// in Java or C++. In fact, `new` is the Rust convention for defining constructors: They are -// nothing special, just static functions returning `Self`. - -// You can call static functions, and in particular constructors, as follows: +//@ Observe how `new` does *not* have a `self` parameter. This corresponds to a `static` method +//@ in Java or C++. In fact, `new` is the Rust convention for defining constructors: They are +//@ nothing special, just static functions returning `Self`. +//@ +// You can call static functions, and in particular constructors, as demonstrated in `call_constructor`. fn call_constructor(x: i32) -> SomethingOrNothing { SomethingOrNothing::new(Some(x)) } -// Now that we have a generic `SomethingOrNothing`, wouldn't it be nice to also gave a generic -// `vec_min`? Of course, we can't take the minimum of a vector of *any* type. It has to be a type -// supporting a `min` operation. Rust calls such properties that we may demand of types *traits*. +// ## Traits +//@ Now that we have a generic `SomethingOrNothing`, wouldn't it be nice to also gave a generic +//@ `vec_min`? Of course, we can't take the minimum of a vector of *any* type. It has to be a type +//@ supporting a `min` operation. Rust calls such properties that we may demand of types *traits*. -// So, as a first step towards a generic `vec_min`, we define a `Minimum` trait. -// For now, just ignore the `Copy`, we will come back to this point later. -// A `trait` is a lot like interfaces in Java: You define a bunch of functions -// you want to have implemented, and their argument and return types. -trait Minimum : Copy { - fn min(a: Self, b: Self) -> Self; +//@ So, as a first step towards a generic `vec_min`, we define a `Minimum` trait. +//@ For now, just ignore the `Copy`, we will come back to this point later. +//@ A `trait` is a lot like interfaces in Java: You define a bunch of functions +//@ you want to have implemented, and their argument and return types.
+//@ The function `min` takes to arguments of the same type, but I made the +//@ first argument the special `self` argument. I could, alternatively, have +//@ made `min` a static function as follows: `fn min(a: Self, b: Self) -> Self`. +//@ However, in Rust one typically prefers methods over static function wherever possible. +pub trait Minimum : Copy { + fn min(self, b: Self) -> Self; } -// Now we can write `vec_min`, generic over a type `T` that we demand to satisfy the `Minimum` trait. -// This is called a *trait bound*. -// The only difference to the version from the previous part is that we call `T::min` (the `min` -// function provided for type `T`) instead of `std::cmp::min`. -// -// Notice a crucial difference to templates in C++: We actually have to declare which traits -// we want the type to satisfy. If we left away the `Minimum`, Rust would have complained that -// we cannot call `min`. Just try it! There is no reason to believe that `T` provides such an operation. -// This is in strong contrast to C++, where the compiler only checks such details when the -// function is actually used. -fn vec_min(v: Vec) -> SomethingOrNothing { +//@ Next, we write `vec_min` as a generic function over a type `T` that we demand to satisfy the `Minimum` trait. +//@ This requirement is called a *trait bound*. +//@ The only difference to the version from the previous part is that we call `e.min(n)` instead +//@ of `std::cmp::min(n, e)`. Rust automatically figures out that `n` is of type `T`, which implements +//@ the `Minimum` trait, and hence we can call that function. +//@ +//@ There is a crucial difference to templates in C++: We actually have to declare which traits +//@ we want the type to satisfy. If we left away the `Minimum`, Rust would have complained that +//@ we cannot call `min`. Just try it!
+//@ This is in strong contrast to C++, where the compiler only checks such details when the +//@ function is actually used. +pub fn vec_min(v: Vec) -> SomethingOrNothing { let mut min = Nothing; for e in v { min = Something(match min { Nothing => e, - Something(n) => T::min(n, e) + // Here, we can now call the `min` function of the trait. + Something(n) => { + e.min(n) /*@*/ + } }); } min } +//@ Before going on, take a moment to ponder the flexibility of Rust's take on abstraction: +//@ We just defined our own, custom trait (interface), and then implemented that trait +//@ *for an existing type*. With the hierarchical approach of, e.g., C++ or Java, +//@ that's not possible: We cannot make an existing type suddenly also inherit from our abstract base class. +//@ +//@ In case you are worried about performance, note that Rust performs *monomorphisation* +//@ of generic functions: When you call `vec_min` with `T` being `i32`, Rust essentially goes +//@ ahead and creates a copy of the function for this particular type, filling in all the blanks. +//@ In this case, the call to `T::min` will become a call to our implementation *statically*. There is +//@ no dynamic dispatch, like there would be for Java interface methods or C++ `virtual` methods. +//@ This behavior is similar to C++ templates. The optimizer (Rust is using LLVM) then has all the +//@ information it could want to, e.g., inline function calls. -// To make the function usable with a `Vec`, we implement the `Minimum` trait for `i32`. +// ## Trait implementations +// To make `vec_min` usable with a `Vec`, we implement the `Minimum` trait for `i32`. impl Minimum for i32 { - fn min(a: Self, b: Self) -> Self { - std::cmp::min(a, b) + fn min(self, b: Self) -> Self { + if self < b { self } else { b } /*@*/ } } -// In order to run our code and see the result, we again provide a `print` function. -// This also shows that we can have multiple `impl` blocks for the same type, and we -// can provide some methods only for certain instances of a generic type. -impl SomethingOrNothing { - fn print(self) { +// We again provide a `print` function. +//@ This also shows that we can have multiple `impl` blocks +//@ for the same type (remember that `NumberOrNothing` is just a type alias for `SomethingOrNothing`), +//@ and we can provide some methods only for certain instances of a generic type. +impl NumberOrNothing { + pub fn print(self) { match self { Nothing => println!("The number is: "), Something(n) => println!("The number is: {}", n), @@ -108,50 +129,21 @@ impl SomethingOrNothing { } } -// Now we are again ready to run our code. Remember to change `main.rs` appropriately. -// Rust figures out automatically that we want the `T` of `vec_min` to be `i32`, and -// that `i32` implements `Minimum` and hence all is good. -// -// In case you are worried about performance, note that Rust performs *monomorphisation* -// of generic functions: When you call `vec_min` with `T` being `i32`, Rust essentially goes -// ahead and creates a copy of the function for this particular type, filling in all the blanks. -// In this case, the call to `T::min` will become a call to our implementation *statically*. There is -// no dynamic dispatch, like there would be for Java interface methods or C++ `virtual` methods. -// This behavior is similar to C++ templates. The optimizer (Rust is using LLVM) then has all the -// information it could want to, e.g., inline function calls. +// Now we are ready to run our new code. Remember to change `main.rs` appropriately. +//@ Rust figures out automatically that we want the `T` of `vec_min` to be `i32`, and +//@ that `i32` implements `Minimum` and hence all is good. fn read_vec() -> Vec { vec![18,5,7,3,9,27] } -pub fn part_main() { +pub fn main() { let vec = read_vec(); let min = vec_min(vec); min.print(); } -// If this printed `3`, then you generic `vec_min` is working! -// -// Before going on, take a moment to ponder the flexibility of Rust's take on abstraction: -// We just defined our own, custom trait (interface), and then implemented that trait -// *for an existing type*. With the hierarchical approach of, e.g., C++ or Java, -// that's not possible: We cannot make an existing type suddenly also inherit from our abstract base class. +//@ If this printed `3`, then you generic `vec_min` is working! So get ready for the next part. -// **Exercise**: Define a trait "Print" to write a generic version of `SomethingOrNothing::print`. -// Implement that trait for `i32`, and change the code above to use it. -// I will again provide a skeleton for this solution. It also shows how to attach bounds to generic -// implementations (just compare it to the `impl` block from the previous exercise). -// You can read this as "For all types `T` satisfying the `Print` trait, I provide an implementation -// for `SomethingOrNothing`". -// -// Notice that I called the function on `SomethingOrNothing` `print2` to disambiguate from the `print` defined above. -// -// *Hint*: There is a macro `print!` for printing without appending a newline. -trait Print { - /* Add things here */ -} -impl SomethingOrNothing { - fn print2(self) { - panic!("Not yet implemented.") - } -} +// **Exercise 02.1**: Change your program such that it computes the minimum of a `Vec` (where `f32` is the type +// of 32-bit floating-point numbers). You should not change `vec_min` in any way, obviously! -// [index](main.html) | [previous](part01.html) | [next](part03.html) +//@ [index](main.html) | [previous](part01.html) | [next](part03.html)