X-Git-Url: https://git.ralfj.de/rust-101.git/blobdiff_plain/17ab30e2988868e5f59b36bb0364cadb0a1c42f8..09a36e34a7b4f163c25fb971771bc4c7edd63e2b:/src/part05.rs?ds=inline diff --git a/src/part05.rs b/src/part05.rs index 25c98e2..d7cf64a 100644 --- a/src/part05.rs +++ b/src/part05.rs @@ -1,9 +1,7 @@ -// Rust-101, Part 05: Copy, Clone -// ============================== - -use std::cmp; -use std::ops; +// Rust-101, Part 05: Clone +// ======================== +// ## Big Numbers // In the course of the next few parts, we are going to build a data-structure for // computations with *bug* numbers. We would like to not have an upper bound // to how large these numbers can get, with the memory of the machine being the @@ -64,6 +62,7 @@ impl BigInt { } } +// ## Cloning // If you have a close look at the type of `BigInt::from_vec`, you will notice that it // consumes the vector `v`. The caller hence loses access. There is however something // we can do if we don't want that to happen: We can explicitly `clone` the vector, @@ -71,9 +70,12 @@ impl BigInt { // `clone` takes a borrowed vector, and returns a fully owned one. fn clone_demo() { let v = vec![0,1 << 16]; - let b1 = BigInt::from_vec(v.clone()); + let b1 = BigInt::from_vec((&v).clone()); let b2 = BigInt::from_vec(v); } +// Rust has special treatment for methods that borrow its `self` argument (like `clone`, or +// like `test_invariant` above): It is not necessary to explicitly borrow the receiver of the +// method. Hence you could replace `(&v).clone()` by `v.clone()` above. Just try it! // To be clonable is a property of a type, and as such, naturally expressed with a trait. // In fact, Rust already comes with a trait `Clone` for exactly this purpose. We can hence @@ -84,54 +86,61 @@ impl Clone for BigInt { } } // Making a type clonable is such a common exercise that Rust can even help you doing it: -// If you add `#[derive(Clone)]' right in front of the definition of `BigInt`, Rust will -// generate an implementation of `clone` that simply clones all the fields. Try it! -// -// To put this in perspective, `clone` in Rust corresponds to what people usually manually do in -// the copy constructor of a C++ class: It creates new, independent instance containing the -// same values. Contrary to that, if you pass something to a function normally (like the -// second call to `from_vec` in `clone_demo`), only a *shallow* copy is created: The fields -// are copied, but pointers are simply duplicated. This corresponds to the default copy -// constructor in C++. Rust assumes that after such a copy, the old value is useless -// (as the new one uses the same pointers), and hence considers the data semantically -// moved to the copy. That's another explanation of why Rust does not let you access -// a vector anymore after you passed ownership to some function. +// If you add `#[derive(Clone)]` right in front of the definition of `BigInt`, Rust will +// generate an implementation of `Clone` that simply clones all the fields. Try it! -// With `BigInt` being about numbers, we should be able to write a version of `vec_min` -// that computes the minimum of a list of `BigInt`. We start by writing `min` for -// `BigInt`. Now our assumption of having no trailing zeros comes in handy! -impl BigInt { - fn min(self, other: Self) -> Self { - // Just to be sure, we first check that both operands actually satisfy our invariant. - // `debug_assert!` is a macro that checks that its argument (must be of type `bool`) - // is `true`, and panics otherwise. It gets removed in release builds, which you do with - // `cargo build --release`. - // - // If you carefully check the type of `BigInt::test_invariant`, you may be surprised that - // we can call the function this way. Doesn't it take `self` in borrowed form? Indeed, - // the explicit way to do that would be to call `(&other).test_invariant()`. However, the - // `self` argument of a method is treated specially by Rust, and borrowing happens automatically here. - debug_assert!(self.test_invariant() && other.test_invariant()); - // If the lengths of the two numbers differ, we already know which is larger. - if self.data.len() < other.data.len() { - self - } else if self.data.len() > other.data.len() { - other - } else { - // **Exercise**: Fill in this code. - panic!("Not yet implemented."); +// We can also make the type `SomethingOrNothing` implement `Clone`. However, that +// can only work if `T` is `Clone`! So we have to add this bound to `T` when we introduce +// the type variable. +use part02::{SomethingOrNothing,Something,Nothing}; +impl Clone for SomethingOrNothing { + fn clone(&self) -> Self { + match *self { + Nothing => Nothing, + // In the second arm of the match, we need to talk about the value `v` + // that's stored in `self`. However, if we would write the pattern as + // `Something(v)`, that would indicate that we *own* `v` in the code + // after the arrow. That can't work though, we have to leave `v` owned by + // whoever called us - after all, we don't even own `self`, we just borrowed it. + // By writing `Something(ref v)`, we borrow `v` for the duration of the match + // arm. That's good enough for cloning it. + Something(ref v) => Something(v.clone()), } } } +// Again, Rust will generate this implementation automatically if you add +// `#[derive(Clone)]` right before the definition of `SomethingOrNothing`. -fn vec_min(v: &Vec) -> Option { - let mut min: Option = None; - for e in v { - // In the loop, `e` now has type `&i32`, so we have to dereference it. - min = Some(match min { - None => e.clone(), - Some(n) => e.clone().min(n) - }); +// ## Mutation + aliasing considered harmful (part 2) +// Now that we know how to borrow a part of an `enum` (like `v` above), there's another example for why we +// have to rule out mutation in the presence of aliasing. First, we define an `enum` that can hold either +// a number, or a string. +enum Variant { + Number(i32), + Text(String), +} +// Now consider the following piece of code. Like above, `n` will be a borrow of a part of `var`, +// and since we wrote `ref mut`, they will be mutable borrows. In other words, right after the match, `ptr` +// points to the number that's stored in `var`, where `var` is a `Number`. Remember that `_` means +// "we don't care". +fn work_on_variant(mut var: Variant, text: String) { + let mut ptr: &mut i32; + match var { + Variant::Number(ref mut n) => ptr = n, + Variant::Text(_) => return, } - min + /* var = Variant::Text(text); */ + *ptr = 1337; } +// Now, imagine what would happen if we were permitted to also mutate `var`. We could, for example, +// make it a `Text`. However, `ptr` still points to the old location! Hence `ptr` now points somewhere +// into the representation of a `String`. By changing `ptr`, we manipulate the string in completely +// unpredictable ways, and anything could happen if we were to use it again! (Technically, the first field +// of a `String` is a pointer to its character data, so by overwriting that pointer with an integer, +// we make it a completely invalid address. When the destructor of `var` runs, it would try to deallocate +// that address, and Rust would eat your laundry - or whatever.) +// +// I hope this example clarifies why Rust has to rule out mutation in the presence of aliasing *in general*, +// not just for the specific + +// [index](main.html) | [previous](part04.html) | [next](part06.html)