Nightly regression from stable - macro_rules macros collide with custom derive attributes

[Arnavion]

rustc 1.30.0-nightly (f8d34596f 2018-08-30)
binary: rustc
commit-hash: f8d34596ff74da91e0877212a9979cb9ca13eb7e
commit-date: 2018-08-30
host: x86_64-pc-windows-msvc
release: 1.30.0-nightly
LLVM version: 7.0

On nightly, building https://github.com/Arnavion/derive-error-chain/tree/v0.11.1/derive-error-chain-tests fails with:

error: macro `error_chain` may not be used in attributes
  --> src\main.rs:49:2
   |
49 |     #[error_chain(error = "Error", result_ext = "ResultExt", result = "Result")]
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

derive-error-chain is a custom derive that registers an attribute named error_chain

#[proc_macro_derive(ErrorChain, attributes(error_chain))]

If the crate that uses the custom derive also has the regular macro_rules macro named error_chain from the error-chain crate in scope, it fails with that error.

#[macro_use]
extern crate derive_error_chain;
#[macro_use]
extern crate error_chain; // Brings `error_chain!` into scope

#[derive(Debug, ErrorChain)]
#[error_chain(error = "Error", result_ext = "ResultExt", result = "Result")] // Uses the attribute registered by the custom derive.
pub enum ErrorKind {
	Msg(String),
}

Stable 1.28 does not have this problem.

I mentioned this in #38356 (comment) and #38356 (comment). @alexcrichton responded to the second one that it would not be an acceptable regression. Since it fails now without needing to enable any features, I assume this is will regress stable eventually, hence I'm opening this issue.

[petrochenkov]

Duplicate of #53583