Iterating over a vector of mutable references to trait objects

前端 未结 2 1429
旧巷少年郎
旧巷少年郎 2020-12-06 17:41

I have a struct that holds mutable references to trait objects:

trait Task {
    fn do_it(&mut self);
}

struct Worker<\'a> {
    task         


        
相关标签:
2条回答
  • 2020-12-06 18:03

    You need to have a mutable reference to each item. iter returns immutable references. And a immutable reference to a mutable variable is not itself mutable. Use iter_mut or for task in &mut self.tasks instead.

    Then, the easiest thing to do is to inline work_one into work:

    pub fn work(&mut self) {
        for task in self.tasks.iter_mut() {
            task.do_it()
        }
    }
    

    Unfortunately, splitting this into two functions is quite painful. You have to guarantee that calling self.work_one will not modify self.tasks. Rust doesn't track these things across function boundaries, so you need to split out all the other member variables and pass them separately to a function.

    See also:

    • cannot borrow `self.x` as immutable because `*self` is also borrowed as mutable
    • Why are borrows of struct members allowed in &mut self, but not of self to immutable methods?
    0 讨论(0)
  • 2020-12-06 18:27

    You are calling tasks.iter() which produces immutable references to the elements of Vec. You actually get back &&mut Task, an immutable reference to a mutable one (that is why the Rust compiler is complaining).

    To solve this, call tasks.iter_mut() to get an iterator of mutable references.

    The second problem is calling defining work_one as a method. You already borrow a mutable reference from self when iterating, so you cannot get another borrow.

    Working example (playground):

    trait Task {
        fn do_it(&mut self);
    }
    
    struct Worker<'a> {
        tasks: Vec<&'a mut Task>,
    }
    
    impl<'a> Worker<'a> {
        pub fn work(&mut self) {
            for task in self.tasks.iter_mut() {
                Worker::work_one(*task);
            }
        }
    
        fn work_one(task: &mut Task) {
            task.do_it();
        }
    }
    

    To still have access to self in work_one this workaround can be used. This basically just swaps the two vectors so you do not actually borrow self when iterating and then swapping it back. This is ugly, there may be a better pattern here, maybe someone else will suggest something better.

    pub fn work(&mut self) {
        let mut tasks = vec![];
        mem::swap(&mut tasks, &mut self.tasks);
        for task in tasks.iter_mut() {
            self.work_one(*task);
        }
        mem::swap(&mut tasks, &mut self.tasks);
    }
    

    A nicer alternative suggested by @Veedrac:

    fn work(&mut self) {
        let mut tasks = mem::replace(&mut self.tasks, Vec::new());
        for task in &mut tasks {
            self.work_one(*task);
        }
        self.tasks = tasks;
    }
    
    0 讨论(0)
提交回复
热议问题