cannot borrow as mutable because it is also borrowed as immutable (likely regression)

[Dylan-DPC-zz]

From a crater run:

 Compiling web-dom v0.0.8
[INFO] [stderr]    Compiling webcomponent v0.3.3 (/opt/crater/workdir)
[INFO] [stderr] error[E0502]: cannot borrow `*controller` as mutable because it is also borrowed as immutable
[INFO] [stderr]   --> src/lib.rs:27:13
[INFO] [stderr]    |
[INFO] [stderr] 27 |             (controller.processor)(controller, &tag, el);
[INFO] [stderr]    |             ----------------------^^^^^^^^^^^^^^^^^^^^^^
[INFO] [stderr]    |             |
[INFO] [stderr]    |             mutable borrow occurs here
[INFO] [stderr]    |             immutable borrow occurs here
[INFO] [stderr]    |             immutable borrow later used by call
[INFO] [stderr] 
[INFO] [stderr] error: aborting due to previous error

crate(s) affected: webcomponent (0.3)
author: @richardanaya
log: https://crater-reports.s3.amazonaws.com/beta-1.35-1/beta-2019-04-11/reg/webcomponent-0.3.3/log.txt

[matthewjasper]

cc #59500

This can be fixed by using (*controller.processor) to avoid the new implementation in that PR.

[nikomatsakis]

@matthewjasper can you elaborate a bit on how #59500 is related here?

[nagisa]

Would be great to have a bisection and/or reduction of the reproducer. Cursory look at the contents of the web-dom crate indicates that it should be fairly easy to make a reproducer here.

[lqd]

Reduction: playground

struct CustomElements {
    processor: Box<fn(&mut CustomElements)>,
}

fn main() {
    move |controller: &mut CustomElements| {
        (controller.processor)(controller);
    };
}

---

edit Bisection points to nightly-2019-04-06 and acd8dd6 which is indeed #59500

[nikomatsakis]

OK, wait, I didn't read #59500 closely enough. I can see why this regression occurred. We didn't used to consider Box<fn(...)> to implement Fn, but we do now (because of the impls added in #59500). This means that (controller.processor) winds up being desugared as a call through Fn to the Box<..>, instead of inserting an auto-deref. This is the sort of breakage we typically accept (for better or worse) as a result of adding new impls.

[nikomatsakis]

The workaround is to change to (*controller.processor), although I suspect the better fix is to remove the Box altogether and just store the fn(&mut CustomElements) directly. =) (but maybe there's another reason to have the Box)

[pnkfelix]

Closing, as the code breakage described here was an expected consequence of teh impl that were added in PR #59500.

[nikomatsakis]

Discussed at the compiler meeting today. We are closing this as a permissible regression -- as I noted earlier, adding the impl caused the () operator to insert fewer dereferences than is used to, which leads to a compiler error. This is unfortunate but it is the sort of breakage that we current permit.