1 // Rust-101, Part 07: Operator Overloading, Tests, Formatting
2 // ==========================================================
4 pub use part05::BigInt;
6 // With our new knowledge of lifetimes, we are now able to write down the desired type of `min`:
7 //@ We want the function to take two borrows *of the same lifetime*, and then
8 //@ return a borrow of that lifetime. If the two input lifetimes would be different, we
9 //@ would not know which lifetime to use for the result.
11 fn min<'a>(&'a self, other: &'a Self) -> &'a Self;
14 //@ Now we can implement a generic function `vec_min` that works on above trait.
15 //@ The code is pretty much straight-forward, and Rust checks that all the
16 //@ lifetimes actually work out. Observe that we don't have to make any copies!
17 pub fn vec_min<T: Minimum>(v: &Vec<T>) -> Option<&T> {
18 let mut min: Option<&T> = None;
20 min = Some(match min {
27 //@ Notice that the return type `Option<&T>` is technically (leaving the borrowing story aside) a
28 //@ pointer to a `T`, that could optionally be invalid. In other words, it's just like a pointer in
29 //@ C(++) or Java that can be `NULL`! However, thanks to `Option` being an `enum`, we cannot forget
30 //@ to check the pointer for validity, avoiding the safety issues of C(++). <br/>
31 //@ Also, if you are worried about wasting space, notice that Rust knows that `&T` can never be
32 //@ `NULL`, and hence optimizes `Option<&T>` to be no larger than `&T`. The `None` case is represented
33 //@ as `NULL`. This is another great example of a zero-cost abstraction: `Option<&T>` is exactly like
34 //@ a pointer in C(++), if you look at what happens during execution - but it's much safer to use.
36 // **Exercise 07.1**: For our `vec_min` to be usable with `BigInt`, you will have to provide an implementation of
37 // `Minimum`. You should be able to pretty much copy the code you wrote for exercise 06.1. You should *not*
38 // make any copies of `BigInt`!
39 impl Minimum for BigInt {
40 fn min<'a>(&'a self, other: &'a Self) -> &'a Self {
45 // ## Operator Overloading
46 //@ How can we know that our `min` function actually does what we want it to do? One possibility
47 //@ here is to do *testing*. Rust comes with nice built-in support for both unit tests and integration
48 //@ tests. However, before we go there, we need to have a way of checking whether the results of function calls are
49 //@ correct. In other words, we need to define how to test equality of `BigInt`. Being able to
50 //@ test equality is a property of a type, that - you guessed it - Rust expresses as a trait: `PartialEq`.
52 //@ Doing this for `BigInt` is fairly easy, thanks to our requirement that there be no trailing zeros. We simply
53 //@ re-use the equality test on vectors, which compares all the elements individually.
54 //@ The `inline` attribute tells Rust that we will typically want this function to be inlined.
55 impl PartialEq for BigInt {
57 fn eq(&self, other: &BigInt) -> bool {
58 debug_assert!(self.test_invariant() && other.test_invariant());
59 self.data == other.data /*@*/
63 //@ Since implementing `PartialEq` is a fairly mechanical business, you can let Rust automate this
64 //@ by adding the attribute `derive(PartialEq)` to the type definition. In case you wonder about
65 //@ the "partial", I suggest you check out the documentation of [`PartialEq`](https://doc.rust-lang.org/std/cmp/trait.PartialEq.html)
66 //@ and [`Eq`](https://doc.rust-lang.org/std/cmp/trait.Eq.html). `Eq` can be automatically derived as well.
68 // Now we can compare `BigInt`s. Rust treats `PratialEq` special in that it is wired to the operator `==`:
69 //@ That operator can not be used on our numbers! Speaking in C++ terms, we just overloaded the `==` operator
70 //@ for `BigInt`. Rust does not have function overloading (i.e., it will not dispatch to different
71 //@ functions depending on the type of the argument). Instead, one typically finds (or defines) a
72 //@ trait that catches the core characteristic common to all the overloads, and writes a single
73 //@ function that's generic in the trait. For example, instead of overloading a function for all
74 //@ the ways a string can be represented, one writes a generic functions over [ToString](https://doc.rust-lang.org/std/string/trait.ToString.html).
75 //@ Usually, there is a trait like this that fits the purpose - and if there is, this has the great
76 //@ advantage that any type *you* write, that can convert to a string, just has to implement
77 //@ that trait to be immediately usable with all the functions out there that generalize over `ToString`.
78 //@ Compare that to C++ or Java, where the only chance to add a new overloading variant is to
79 //@ edit the class of the receiver.
81 //@ Why can we also use `!=`, even though we just overloaded `==`? The answer lies in what's called a *default implementation*.
82 //@ If you check out the documentation of `PartialEq` I linked above, you will see that the trait actually provides
83 //@ two methods: `eq` to test equality, and `ne` to test inequality. As you may have guessed, `!=` is wired to `ne`.
84 //@ The trait *definition* also provides a default implementation of `ne` to be the negation of `eq`. Hence you can just
85 //@ provide `eq`, and `!=` will work fine. Or, if you have a more efficient way of deciding inequality, you can provide
86 //@ `ne` for your type yourself.
87 fn compare_big_ints() {
88 let b1 = BigInt::new(13);
89 let b2 = BigInt::new(37);
90 println!("b1 == b1: {} ; b1 == b2: {}; b1 != b2: {}", b1 == b1, b1 == b2, b1 != b2);
94 // With our equality test written, we are now ready to write our first testcase.
95 //@ It doesn't get much simpler: You just write a function (with no arguments or return value), and give it
96 // the `test` attribute. `assert!` is like `debug_assert!`, but does not get compiled away in a release build.
99 let b1 = BigInt::new(1);
100 let b2 = BigInt::new(42);
101 let b3 = BigInt::from_vec(vec![0, 1]);
103 assert!(*b1.min(&b2) == b1); /*@*/
104 assert!(*b3.min(&b2) == b2); /*@*/
106 // Now run `cargo test` to execute the test. If you implemented `min` correctly, it should all work!
109 //@ There is also a macro `assert_eq!` that's specialized to test for equality, and that prints the two
110 //@ values (left and right) if they differ. To be able to do that, the macro needs to know how to format
111 //@ the value for printing. This means that we - guess what? - have to implement an appropriate trait.
112 //@ Rust knows about two ways of formatting a value: `Display` is for pretty-printing something in a way
113 //@ that users can understand, while `Debug` is meant to show the internal state of data and targeted at
114 //@ the programmer. The latter is what we want for `assert_eq!`, so let's get started.
116 // All formating is handled by [`std::fmt`](https://doc.rust-lang.org/std/fmt/index.html). I won't explain
117 // all the details, and refer you to the documentation instead.
120 //@ In the case of `BigInt`, we'd like to just output our internal `data` array, so we
121 //@ simply call the formating function of `Vec<u64>`.
122 impl fmt::Debug for BigInt {
123 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
127 //@ `Debug` implementations can be automatically generated using the `derive(Debug)` attribute.
129 // Now we are ready to use `assert_eq!` to test `vec_min`.
132 let b1 = BigInt::new(1);
133 let b2 = BigInt::new(42);
134 let b3 = BigInt::from_vec(vec![0, 1]);
136 let v1 = vec![b2.clone(), b1.clone(), b3.clone()];
137 let v2 = vec![b2.clone(), b3.clone()];
138 assert_eq!(vec_min(&v1), Some(&b1)); /*@*/
139 assert_eq!(vec_min(&v2), Some(&b2)); /*@*/
142 // **Exercise 07.1**: Add some more testcases. In particular, make sure you test the behavior of
143 // `vec_min` on an empty vector. Also add tests for `BigInt::from_vec` (in particular, removing
144 // trailing zeros). Finally, break one of your functions in a subtle way and watch the test fail.
146 // **Exercise 07.2**: Go back to your good ol' `SomethingOrNothing`, and implement `Display` for it. (This will,
147 // of course, need a `Display` bound on `T`.) Then you should be able to use them with `println!` just like you do
148 // with numbers, and get rid of the inherent functions to print `SomethingOrNothing<i32>` and `SomethingOrNothing<f32>`.
150 //@ [index](main.html) | [previous](part06.html) | [next](part08.html)