X-Git-Url: https://git.ralfj.de/rust-101.git/blobdiff_plain/09a36e34a7b4f163c25fb971771bc4c7edd63e2b..2e8bcf46df767e7cbceb1ae92ba2ec8e4996c241:/src/part08.rs?ds=sidebyside diff --git a/src/part08.rs b/src/part08.rs index db8e56c..e01a35b 100644 --- a/src/part08.rs +++ b/src/part08.rs @@ -1,21 +1,41 @@ // Rust-101, Part 08: Associated Types, Modules // ============================================ -use std::cmp; -use std::ops; -use std::fmt; +use std::{cmp,ops}; use part05::BigInt; -// Add with carry, returning the sum and the carry +//@ As our next goal, let us implement addition for our `BigInt`. The main issue here will be dealing with the overflow. +//@ First of all, we will have to detect when an overflow happens. This is stored in a so-called *carry* bit, and we have to carry this +//@ information on to the next pair of digits we add. The core primitive of addition therefore is to add two digits *and* a +//@ carry, and to return the sum digit and the next carry. + +// So, let us write a function to "add with carry", and give it the appropriate type. Notice Rust's native support for pairs. fn overflowing_add(a: u64, b: u64, carry: bool) -> (u64, bool) { + //@ Rust's stanza on integer overflows may be a bit surprising: In general, when we write `a + b`, an overflow is + //@ considered an *error*. If you compile your program in debug mode, Rust will actually check for that error and panic + //@ the program in case of overflows. For performance reasons, no such checks are currently inserted for release builds. + //@ The reason for this is that many serious security vulnerabilities have been caused by integer overflows, so just assuming + //@ "per default" that they are intended is dangerous.
+ //@ If you explicitly *do* want an overflow to happen, you can call the `wrapping_add` + //@ function (see [the documentation](http://doc.rust-lang.org/stable/std/primitive.u64.html#method.wrapping_add), + //@ there are similar functions for other arithmetic operations). There are also similar functions + //@ `checked_add` etc. to enforce the overflow check. let sum = u64::wrapping_add(a, b); - if sum >= a { // first addition did not overflow - unimplemented!() - } else { // first addition *did* overflow + // If an overflow happened, then the sum will be smaller than *both* summands. Without an overflow, of course, it will be + // at least as large as both of them. So, let's just pick one and check. + if sum >= a { + // The addition did not overflow.
+ // **Exercise 08.1**: Write the code to handle adding the carry in this case. unimplemented!() + } else { + // The addition *did* overflow. It is impossible for the addition of the carry + // to overflow again, as we are just adding 0 or 1. + (sum + if carry { 1 } else { 0 }, true) /*@*/ } } +// `overflow_add` is a sufficiently intricate function that a test case is justified. +// This should also help you to check your solution of the exercise. /*#[test]*/ fn test_overflowing_add() { assert_eq!(overflowing_add(10, 100, false), (110, false)); @@ -25,12 +45,106 @@ fn test_overflowing_add() { assert_eq!(overflowing_add(1 << 63, (1 << 63) -1 , true), (0, true)); } -impl ops::Add for BigInt { +// ## Associated Types +//@ Now we are equipped to write the addition function for `BigInt`. As you may have guessed, the `+` operator +//@ is tied to a trait (`std::ops::Add`), which we are now going to implement for `BigInt`. +//@ +//@ In general, addition need not be homogeneous: For example, we could add a vector (in 3-dimensional +//@ space, say) to a point. So when implementing `Add` for a type, one has to specify the type of +//@ the other operand. In this case, it will also be `BigInt` (and we could have left it away, since that's the default). +impl ops::Add for BigInt { + //@ Besides static functions and methods, traits can contain *associated types*: This is a type + //@ chosen by every particular implementation of the trait. The methods of the trait can then + //@ refer to that type. In the case of addition, it is used to give the type of the result. + //@ (Also see the [documentation of `Add`](http://doc.rust-lang.org/stable/std/ops/trait.Add.html).) + //@ + //@ In general, you can consider the two `BigInt` given above (in the `impl` line) *input* types of trait search: When + //@ `a + b` is invoked with `a` having type `T` and `b` having type `U`, Rust tries to find an implementation of `Add` for + //@ `T` where the right-hand type is `U`. The associated types, on the other hand, are *output* types: For every combination + //@ of input types, there's a particular result type chosen by the corresponding implementation of `Add`. + + // Here, we choose the result type to be again `BigInt`. type Output = BigInt; + + // Now we can write the actual function performing the addition. fn add(self, rhs: BigInt) -> Self::Output { - let mut result_vec:Vec = Vec::with_capacity(cmp::max(self.data.len(), rhs.data.len())); + // We know that the result will be *at least* as long as the longer of the two operands, + // so we can create a vector with sufficient capacity to avoid expensive reallocations. + let max_len = cmp::max(self.data.len(), rhs.data.len()); + let mut result_vec:Vec = Vec::with_capacity(max_len); + let mut carry = false; /* the current carry bit */ + for i in 0..max_len { + // Compute next digit and carry. Store the digit for the result, and the carry for later. + let lhs_val = if i < self.data.len() { self.data[i] } else { 0 }; + let rhs_val = if i < rhs.data.len() { rhs.data[i] } else { 0 }; + let (sum, new_carry) = overflowing_add(lhs_val, rhs_val, carry); /*@*/ + result_vec.push(sum); /*@*/ + carry = new_carry; /*@*/ + } + // **Exercise 08.2**: Handle the final `carry`, and return the sum. unimplemented!() } } -// [index](main.html) | [previous](part07.html) | [next](main.html) +// ## Traits and borrowed types +//@ If you inspect the addition function above closely, you will notice that it actually requires +//@ *ownership* of its arguments: Both operands are consumed to produce the result. This is, of +//@ course, in general not what we want. We'd rather like to be able to add two `&BigInt`. + +// Writing this out becomes a bit tedious, because trait implementations (unlike functions) require full explicit annotation +// of lifetimes. Make sure you understand exactly what the following definition says. +impl<'a, 'b> ops::Add<&'a BigInt> for &'b BigInt { + type Output = BigInt; + fn add(self, rhs: &'a BigInt) -> Self::Output { + // **Exercise 08.3**: Implement this function. + unimplemented!() + } +} + +// ## Modules +//@ As you learned, tests can be written right in the middle of your development in Rust. However, it is +//@ considered good style to bundle all tests together. This is particularly useful in cases where +//@ you wrote utility functions for the tests, that no other code should use. + +// Rust calls a bunch of definitions that are grouped together a *module*. You can put definitions in a submodule as follows. +mod my_mod { + type MyType = i32; + fn my_fun() -> MyType { 42 } +} +//@ As already mentioned, outside of the module, only those items declared public with `pub` may be used. Submodules can access +//@ everything defined in their parents. Modules themselves are also hidden from the outside per default, and can be made public +//@ with `pub`. When you use an identifier (or, more general, a *path* like `mod1::submod::name`), it is interpreted as being +//@ relative to the current module. So, for example, to access `overflowing_add` from within `my_mod`, you would have to give a more +//@ explicit path by writing `super::overflowing_add`, which tells Rust to look in the parent module. +//@ +//@ You can make names from other modules available locally with `use`. Per default, `use` works globally, so e.g. +//@ `use std;` imports the *global* name `std`. By adding `super::` or `self::` to the beginning of the path, you make it relative +//@ to the parent or current module, respectively. (You can also explicitly construct an absolute path by starting it with `::`, +//@ e.g., `::std::cmp::min`). You can say `pub use path;` to simultaneously *import* names and make them publicly available to others. +//@ Finally, you can import all public items of a module at once with `use module::*;`. +//@ +//@ Modules can be put into separate files with the syntax `mod name;`. To explain this, let me take a small detour through +//@ the Rust compilation process. Cargo starts by invoking`rustc` on the file `src/lib.rs` or `src/main.rs`, depending on whether +//@ you compile an application or a library. When `rustc` encounters a `mod name;`, it looks for the files `name.rs` and +//@ `name/mod.rs` and goes on compiling there. (It is an error for both of them to exist). You can think of the contents of the +//@ file being embedded at this place. However, only the file where compilation started, and files `name/mod.rs` can load modules +//@ from other files. This ensures that the directory structure mirrors the structure of the modules, with `mod.rs`, `lib.rs` +//@ and `main.rs` representing a directory or crate itself (similar to, e.g., `__init__.py` in Python). + +// For the purpose of testing, one typically introduces a module called `tests` and tells the compiler +// (by means of the `cfg` attribute) to only compile this module for tests. +#[cfg(test)] +mod tests { + //@ If we added some functions here that are useful for testing, Rust would not bother compiling + //@ them when you just build your program for normal use. Other than that, tests work as usually. + #[test] + fn test_add() { + let b1 = BigInt::new(1 << 32); + let b2 = BigInt::from_vec(vec![0, 1]); + + assert_eq!(&b1 + &b2, BigInt::from_vec(vec![1 << 32, 1])); + // **Exercise 08.4**: Add some more testcases. + } +} + +//@ [index](main.html) | [previous](part07.html) | [next](main.html)