accept that box dyn error is not and will never be an error

[yaahc]

r? @dtolnay since this is also relevant to you and anyhow.

I'm adding this negative impl to make it so I can eventually add the following impl to eyre:

    impl StdError for Box<dyn std::error::Error + Send + Sync + 'static> {
        fn ext_report<D>(self, msg: D) -> Report
        where
            D: Display + Send + Sync + 'static,
        {
            self.wrap_err(msg)
        }
    }

Which lets me call .wrap_err on Result<T, Box<dyn Error + Send + Sync + 'static>>. This is currently impossible because it produces a future incompatibility error:

error[E0119]: conflicting implementations of trait `context::ext::StdError` for type `std::boxed::Box<(dyn std::error::Error + std::marker::Send + std::marker::Sync + 'static)>`
  --> src/context.rs:30:5
   |
18 | /     impl<E> StdError for E
19 | |     where
20 | |         E: std::error::Error + Send + Sync + 'static,
21 | |     {
...  |
27 | |         }
28 | |     }
   | |_____- first implementation here
29 | 
30 |       impl StdError for Box<dyn std::error::Error + Send + Sync + 'static> {
   |       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ conflicting implementation for `std::boxed::Box<(dyn std::error::Error + std::marker::Send + std::marker::Sync + 'static)>`
   |
   = note: upstream crates may add a new impl of trait `std::error::Error` for type `std::boxed::Box<(dyn std::error::Error + std::marker::Send + std::marker::Sync + 'static)>` in future versions

For more information about this error, try `rustc --explain E0119`.
error: could not compile `eyre` due to previous error

But this error is incorrect because we cannot add the proper Error impl to Box<dyn Error> without lattice specialization, which is not happening any time soon, if ever. Adding this negative impl promises to never impl Error for box dyn error in the future, so this will need an FCP.

[yaahc]

Hmm, for some reason it did not work, I tried testing the change in eyre with a locally built toolchain and it just slightly changed the error msg instead of allowing the impl...

eyre on  helpful [$!] is 📦 v0.6.5 via 🦀 v1.60.0-dev on ☁️  [email protected] 
❯ cargo build
   Compiling eyre v0.6.5 (/home/jlusby/git/yaahc/eyre)
   Compiling once_cell v1.8.0
   Compiling indenter v0.3.3
error[E0119]: conflicting implementations of trait `context::ext::StdError` for type `std::boxed::Box<(dyn std::error::Error + std::marker::Send + std::marker::Sync + 'static)>`
  --> src/context.rs:30:5
   |
18 | /     impl<E> StdError for E
19 | |     where
20 | |         E: std::error::Error + Send + Sync + 'static,
21 | |     {
...  |
27 | |         }
28 | |     }
   | |_____- first implementation here
29 | 
30 |       impl StdError for Box<dyn std::error::Error + Send + Sync + 'static> {
   |       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ conflicting implementation for `std::boxed::Box<(dyn std::error::Error + std::marker::Send + std::marker::Sync + 'static)>`

For more information about this error, try `rustc --explain E0119`.
error: could not compile `eyre` due to previous error

cc @spastorino, can you check if this is a bug in the current impl of rust-lang/negative-impls-initiative#1 and if so use it as a testcase?

[yaahc]

As a minor update, I talked to niko today and he mentioned that he doesn't think that lattice specialization is likely to never happen. Given that I don't think we should be making this commitment, at least not until we've completely ruled out lattice specialization as a possible feature. I'm going to close this for now though I still would like to see if this can be made to work with negative-impls incase we ever decide to add it again in the future.