Rust AsMut copying rather than taking mutable reference? - rust

My understanding of AsMut is that it is supposed to provide a generic way to take an argument which is "equivalent* to a mutable reference, i.e. it can be cheaply converted to a mutable reference.
However, I have encountered the following example code, I am trying to use AsMut just to be generic over slice, arr and Vec, but it seems to just be copying my array, rather than passing a mutable reference to it and modifying it in place:
pub fn uses_asmut<T, M>(mut m: M)
where M: AsMut<[T]> {
m.as_mut().swap(0,1);
}
#[test]
pub fn test_swap() {
let arr = [1,2];
uses_asmut(arr);
assert_eq!(arr, [2,1]);
}
(note, I know something must be wrong since I apparently pass ownership of arr to uses_asmut as an argument, but then the borrow checker doesn't complain on the next line when I use arr again! If I change it to uses_asmut(&mut arr) the test passes, but I think the code as written shouldn't even compile!)

The conversion should be cheap. But you're passing the array by value, and that was never said to be cheap. You could pass &mut arr to not copy the array.
When we say the conversion should be cheap, we mean "don't do something like str::from_utf8_mut()" that needs to scan the whole string. Indeed, the conversion from [T; N] to [T] is extremely cheap: so cheap, that it happens automatically by the compiler (coercion).
But it does not mean it is equivalent to mutable reference, because it is not a mutable reference. It is a generic type. If you want a mutable reference, take a mutable reference. You can even use AsMut, like &mut impl AsMut<[T]>.
There is no way AsMut could prevent you from moving (or copying) things but also not require you to type the &mut at the call site.

Related

Ergonomically passing a slice of trait objects

I am converting a variety of types to String when they are passed to a function. I'm not concerned about performance as much as ergonomics, so I want the conversion to be implicit. The original, less generic implementation of the function simply used &[impl Into<String>], but I think that it should be possible to pass a variety of types at once without manually converting each to a string.
The key is that ideally, all of the following cases should be valid calls to my function:
// String literals
perform_tasks(&["Hello", "world"]);
// Owned strings
perform_tasks(&[String::from("foo"), String::from("bar")]);
// Non-string types
perform_tasks(&[1,2,3]);
// A mix of any of them
perform_tasks(&["All", 3, String::from("types!")]);
Some various signatures I've attempted to use:
fn perform_tasks(items: &[impl Into<String>])
The original version fails twice; it can't handle numeric types without manual conversion, and it requires all of the arguments to be the same type.
fn perform_tasks(items: &[impl ToString])
This is slightly closer, but it still requires all of the arguments to be of one type.
fn perform_tasks(items: &[&dyn ToString])
Doing it this way is almost enough, but it won't compile unless I manually add a borrow on each argument.
And that's where we are. I suspect that either Borrow or AsRef will be involved in a solution, but I haven't found a way to get them to handle this situation. For convenience, here is a playground link to the final signature in use (without the needed references for it to compile), alongside the various tests.
The following way works for the first three cases if I understand your intention correctly.
pub fn perform_tasks<I, A>(values: I) -> Vec<String>
where
A: ToString,
I: IntoIterator<Item = A>,
{
values.into_iter().map(|s| s.to_string()).collect()
}
As the other comments pointed out, Rust does not support an array of mixed types. However, you can do one extra step to convert them into a &[&dyn fmt::Display] and then call the same function perform_tasks to get their strings.
let slice: &[&dyn std::fmt::Display] = &[&"All", &3, &String::from("types!")];
perform_tasks(slice);
Here is the playground.
If I understand your intention right, what you want is like this
fn main() {
let a = 1;
myfn(a);
}
fn myfn(i: &dyn SomeTrait) {
//do something
}
So it's like implicitly borrow an object as function argument. However, Rust won't let you to implicitly borrow some objects since borrowing is quite an important safety measure in rust and & can help other programmers quickly identified which is a reference and which is not. Thus Rust is designed to enforce the & to avoid confusion.

Calculate object A, then return object B that references A in Rust

In my code I often want to calculate a new value A, and then return some view of that value B, because B is a type that's more convenient to work with. The simplest case is where A is a vector and B is a slice that I would like to return. Let's say I want to write a function that returns a set of indices. Ideally this would return a slice directly because then I can use it immediately to index a string.
If I return a vector instead of a slice, I have to use to_slice:
fn all_except(except: usize, max:usize) -> Vec<usize> {
(0..except).chain((except + 1)..max).collect()
}
"abcdefg"[all_except(1, 7)]
string indices are ranges of `usize`
the type `str` cannot be indexed by `Vec<usize>`
help: the trait `SliceIndex<str>` is not implemented for `Vec<usize>`
I can't return a slice directly:
fn all_except(except: usize, max:usize) -> &[usize] {
(0..except).chain((except + 1)..max).collect()
}
"abcdefg"[all_except(1, 7)]
^ expected named lifetime parameter
missing lifetime specifier
help: this function's return type contains a borrowed value with an elided lifetime, but the lifetime cannot be derived from the arguments
help: consider using the `'static` lifetime
I can't even return the underlying vector and a slice of it, for the same reason
pub fn except(index: usize, max: usize) -> (&[usize], Vec<usize>) {
let v = (0..index).chain((index + 1)..max).collect();
(v, v.as_slice)
}
"abcdefg"[all_except(1, 7)[1]
Now it may be possible to hack this particular example using deref coercion (I'm not sure), but I have encountered this problem with more complex types. For example, I have a function that loads an ndarray::Array2<T> from CSV file, then want to split it into two parts using array.split_at(), but this returns two ArrayView2<T> which reference the original Array2<T>, so I encounter the same issue. In general I'm wondering if there's a solution to this problem in general. Can I somehow tell the compiler to move A into the parent frame's scope, or let me return a tuple of (A, B), where it realises that the slice is still valid because A is still alive?
Your code doesn't seem to make any sense, you can't index a string using a slice. If you could the first snippet would have worked with just an as_slice in the caller or something, vecs trivially coerce to slices. That's exactly what the compiler error is telling you: the compiler is looking for a SliceIndex and a Vec (or slice) is definitely not that.
That aside,
Can I somehow tell the compiler to move A into the parent frame's scope, or let me return a tuple of (A, B), where it realises that the slice is still valid because A is still alive?
There are packages like owning_ref which can bundle owner and reference to avoid extra allocations. It tends to be somewhat fiddly.
I don't think there's any other general solution, because Rust reasons at the function level, the type checker has no notion of "tell the compiler to move A into the parent scope". So you need a construct which works around borrow checker.

Cannot add-assign within `Vec` of user-defined type

Consider this code (Rust Playground):
#[derive(Clone, Copy, Debug)]
struct X(i32);
impl std::ops::AddAssign for X {
fn add_assign(&mut self, rhs: Self) {
self.0 += rhs.0;
}
}
fn main() {
let mut ary_i32 = [1_i32; 2];
ary_i32[0] += ary_i32[1]; // OK
let mut ary_x = [X(1); 2];
ary_x[0] += ary_x[1]; // OK
let mut vec_i32 = vec![1_i32; 2];
vec_i32[0] += vec_i32[1]; // OK
let mut vec_x = vec![X(1); 2];
vec_x[0] += vec_x[1]; // error[E0502]: cannot borrow `vec_x` as immutable because it is also borrowed as mutable
}
Why I get E0502 only on vec_x line?
I could not understand why only the operations for ary_x and vec_i32 are permitted. Does borrow checker treat builtin types (i32, array) specially?
I researched some resources and read MIR of my code, and managed to understand what is going on.
The comment by #trentcl will be the best answer.
I write the details as possible.
For array, Index and IndexMut traits are not used and compiler directly manipulates array elements (you can see this with MIR). So, borrowing problem does not exist here.
Explanating for Vec, rustc guide is useful.
First, Two-phase borrow is not applied to vec_foo[0] += vec_foo[1] statement.
And, the difference between i32 and X is caused by operator lowering.
Basically, statements like vec_user_defined[0] += vec_user_defined[1] are converted to function calls like add_assign(index_mut(...), *index(...)), and function arguments are evaluated from left to right. So, index_mut() borrows x mutably and index() tries to borrow x, and fails.
But for builtin types like i32, compound assignment operator is not converted to function call, and rhs is evaluated before lhs (you can see index() is called before index_mut() with MIR). So, for builtin types, vec_builtin[0] += vec_builtin[1] works.
I know these things from lo48576's article (Japanese).
I considered some workarounds:
Just use an intermediate variable as #sebpuetz said.
Convert Vec to slice as #trentcl said. But this doesn't work well for multidimensional Vec.
Write some macro to automatically introduce an intermediate variable. I found rhs_first_assign crate does such works.
Rust arrays live on the stack, are predictably sized, and therefore have stronger borrow checker guarantees. Vectors are smart pointers on the stack pointing at data that can grow and shrink on the Heap. Because the final example uses the Vector type, the borrow checker considers the entire Vector as a single mutably borrowed object when loading it from the Heap.
As you've observed, the borrow checker can create a mutable reference to a single element to something living on the Stack, whereas it creates a mutable reference to the Vector's smart pointer on the Stack, and then a further mutable reference to the data on the heap. This is why the immutable reference to vec_vec_x[1][1] fails.
As #sebpuetz noted in a comment, you can solve this by first copying an immutable reference to vec_vec_x[1][1], then creating an immutable reference.

Avoiding borrowing mutable and immutable at the same time

To add the elements of two Vecs I wrote a function like
fn add_components(dest: &mut Vec<i32>, first: &Vec<i32>, second: &Vec<i32>){
for i in 0..first.len() {
dest[i] = first[i] + second[i];
}
}
And this works fine when dest is another Vec.
let mut new_components = Vec::with_capacity(components.len());
Vector::add_components(&mut new_comps, &components, &other_components);
But it blows up when I am trying to add in-place:
Vector::add_components(&mut components, &components, &other_components);
because now I borrow components as mutable and immutable at the same time. But this obviously is what I am trying to achieve.
Are there any conventional and general (meaning not only concerning Vecs) solutions to this problem which don't involve unsafe code and pointer magic?
Another example of this problem:
Suppose I want to overload AddAssign for a numeric type like
impl AddAssign<Output=&NumericType> for NumericType {
fn add_assign(&mut self, other: &NumericType) {
unimplemented!() // concrete implementation is not important
}
}
Notice that I want to take a reference as second argument to avoid copying. This works fine when adding two different objects, but adding an object to itself creates the exact same scenario:
let mut num = NumericType{};
num += &num
I am borrowing num mutably and immutably at the same time. So obviously this should work and is safe, but it also is against Rust's borrowing rules.
What are the best practices (apart from copying of course) to deal with this issue, which arises in many forms?
There is no generic solution to this. Rust can't generically abstract over mutability in borrow checking.
You will need to have two versions of the function for in-place and destination versions.
Rust has strict aliasing rules, so dest[i] = first[i] + second[i] actually compiles to different code depending on whether the compiler has a guarantee that dest and first are different. Don't try to fudge it with unsafe, because it will be Undefined Behavior and will get miscompiled.

Why does the argument for the find closure need two ampersands?

I have been playing with Rust by porting my Score4 AI engine to it - basing the work on my functional-style implementation in OCaml. I specifically wanted to see how Rust fares with functional-style code.
The end result: It works, and it's very fast - much faster than OCaml. It almost touches the speed of imperative-style C/C++ - which is really cool.
There's a thing that troubles me, though — why do I need two ampersands in the last line of this code?
let moves_and_scores: Vec<_> = moves_and_boards
.iter()
.map(|&(column,board)| (column, score_board(&board)))
.collect();
let target_score = if maximize_or_minimize {
ORANGE_WINS
} else {
YELLOW_WINS
};
if let Some(killer_move) = moves_and_scores.iter()
.find(|& &(_,score)| score==target_score) {
...
I added them is because the compiler errors "guided" me to it; but I am trying to understand why... I used the trick mentioned elsewhere in Stack Overflow to "ask" the compiler to tell me what type something is:
let moves_and_scores: Vec<_> = moves_and_boards
.iter()
.map(|&(column,board)| (column, score_board(&board)))
.collect();
let () = moves_and_scores;
...which caused this error:
src/main.rs:108:9: 108:11 error: mismatched types:
expected `collections::vec::Vec<(u32, i32)>`,
found `()`
(expected struct `collections::vec::Vec`,
found ()) [E0308]
src/main.rs:108 let () = moves_and_scores;
...as I expected, moves_and_scores is a vector of tuples: Vec<(u32, i32)>. But then, in the immediate next line, iter() and find() force me to use the hideous double ampersands in the closure parameter:
if let Some(killer_move) = moves_and_scores.iter()
.find(|& &(_,score)| score==target_score) {
Why does the find closure need two ampersands? I could see why it may need one (pass the tuple by reference to save time/space) but why two? Is it because of the iter? That is, is the iter creating references, and then find expects a reference on each input, so a reference on a reference?
If this is so, isn't this, arguably, a rather ugly design flaw in Rust?
In fact, I would expect find and map and all the rest of the functional primitives to be parts of the collections themselves. Forcing me to iter() to do any kind of functional-style work seems burdensome, and even more so if it forces this kind of "double ampersands" in every possible functional chain.
I am hoping I am missing something obvious - any help/clarification most welcome.
This here
moves_and_scores.iter()
gives you an iterator over borrowed vector elements. If you follow the API doc what type this is, you'll notice that it's just the iterator for a borrowed slice and this implements Iterator with Item=&T where T is (u32, i32) in your case.
Then, you use find which takes a predicate which takes a &Item as parameter. Sice Item already is a reference in your case, the predicate has to take a &&(u32, i32).
pub trait Iterator {
...
fn find<P>(&mut self, predicate: P) -> Option<Self::Item>
where P: FnMut(&Self::Item) -> bool {...}
... ^
It was probably defined like this because it's only supposed to inspect the item and return a bool. This does not require the item being passed by value.
If you want an iterator over (u32, i32) you could write
moves_and_scores.iter().cloned()
cloned() converts the iterator from one with an Item type &T to one with an Item type T if T is Clone. Another way to do it would be to use into_iter() instead of iter().
moves_and_scores.into_iter()
The difference between the two is that the first option clones the borrowed elements while the 2nd one consumes the vector and moves the elements out of it.
By writing the lambda like this
|&&(_, score)| score == target_score
you destructure the "double reference" and create a local copy of the i32. This is allowed since i32 is a simple type that is Copy.
Instead of destructuring the parameter of your predicate you could also write
|move_and_score| move_and_score.1 == target_score
because the dot operator automatically dereferences as many times as needed.

Resources