Warning about missing `dyn` in `#[ derive() ]`

[najamelan]

warning: trait objects without an explicit `dyn` are deprecated>                                                                      
  -->
26 | #[ derive( Copy, Clone, Eq, PartialEq, Debug, Fail ) ]
   |                                               ^^^^ help: use `dyn`: `dyn (Fail)`
note: lint level defined here
  --> ...path
   |
1  | #![ warn   ( rust_2018_idioms  ) ] // Opt in to warnings about new 2018 idioms
   |              ^^^^^^^^^^^^^^^^
   = note: #[warn(bare_trait_objects)] implied by #[warn(rust_2018_idioms)]

This looks wrong to me, and the suggestion doesn't even compile either. Does the warning check whether we are in an attribute?

[olivia-fl]

I'm also experiencing this, and it's also worth noting that adding #[allow(bare_trait_objects)] to the struct in question doesn't remove the warning.

[oli-obk]

I guess derive macro expansions should have the edition of the derive macro crate and not of the one using the derive attribute?

[najamelan]

@withoutboats ah, so this would be some code in the crate failure_derive causing this warning and the warning would be correct?

#![ allow( bare_trait_objects ) ] on top of the file silences the warning.

These are the macro expansions from failure_derive. One of the mentions of Fail seems to have dyn but the other doesn't:

impl ::failure::Fail for ErrorKind {
   #[allow(unreachable_code)]
   fn cause(&self) -> ::std::option::Option<&::failure::Fail> {
       match *self {
           ErrorKind::NotAFile => { return None }
           ErrorKind::UnicodeError => { return None }
       }
       None
   }
   ...
}

impl Fail for MyError {
   fn cause(&self) -> Option<&dyn Fail> { self.inner.cause() }
   fn backtrace(&self) -> Option<&Backtrace> {
       self.inner.backtrace()
   }
}

[alexcrichton]

While during triage we placed this on the "Rust 2018 Preview 2" milestone I realize now this is related to the rust_2018_idioms lint which is less critical than the compatibility lint, so I'm going to reclassify as the release candidate

[alexcrichton]

I believe this has since been fixed in the interim, so closing.