Regresion in rustls-0.5.7, Rust 1.17

[brson]

/~https://github.com/ctz/rustls

130 brian@ip-10-145-43-250:~/dev/rustls⟫ git log -1
commit b149eb78187b97f057e3ee1aade6be67bf093229
Author: Joseph Birr-Pixton <jpixton@gmail.com>
Date:   Mon Feb 27 19:30:00 2017 +0000

    0.5.7

brian@ip-10-145-43-250:/mnt2/dev⟫ rustc +nightly -Vv
rustc 1.18.0-nightly (5c94997b6 2017-03-30)
binary: rustc
commit-hash: 5c94997b6b1cca3e65ecfc5ba40c033d6838019b
commit-date: 2017-03-30
host: x86_64-unknown-linux-gnu
release: 1.18.0-nightly
LLVM version: 3.9

brian@ip-10-145-43-250:~/dev/rustls⟫ cargo +nightly test
   Compiling rustls v0.5.7 (file:///mnt2/dev/rustls)
error: The attribute `export_macro` is currently unknown to the compiler and may have meaning added to it in the future (see issue #29642)
 --> src/msgs/macros.rs:3:1
  |
3 | #[export_macro]
  | ^^^^^^^^^^^^^^^
  |
  = help: add #![feature(custom_attribute)] to the crate attributes to enable

error: aborting due to previous error

error: Could not compile `rustls`.
Build failed, waiting for other jobs to finish...
error: The attribute `export_macro` is currently unknown to the compiler and may have meaning added to it in the future (see issue #29642)
 --> src/msgs/macros.rs:3:1
  |
3 | #[export_macro]
  | ^^^^^^^^^^^^^^^
  |
  = help: add #![feature(custom_attribute)] to the crate attributes to enable

error: aborting due to previous error

error: build failed

cc @cristicbz

[cristicbz]

@brson I think you meant @ctz

[ctz]

This was #40487 and rustls/rustls#57, fixed in rustls 0.5.8

[alexcrichton]

Out of curiosity, do we know the PR to rust that caused this regression?

[brson]

@jseyfried look familiar?

[brson]

Would be good to have the PR that caused this.

[brson]

@nagisa says this is on beta too.

[withoutboats]

@jseyfried I don't understand the connection between the PR you linked to and this regression. Can you explain more?

[jseyfried]

@withoutboats Whoops, though I was commenting on #40966 -- moved the comment there.

[nikomatsakis]

I'm confused why this would have ever worked? Isn't the correct name #[macro_export]?

[nagisa]

My guess would be "for some reason the check for unused attributes used to not work within that context for some reason before."

[nagisa]

IMO removing the error is hardly justifiable, and should be considered a bug fix.

[nikomatsakis]

@nagisa

IMO removing the error is hardly justifiable, and should be considered a bug fix.

By this do you mean "it's ok that we fixed this bug, even though it caused a regression"? If so, I am inclined to agree; plausible we could use a warning period, but maybe not worth it if only this crate is affected.

[pnkfelix]

As discussed above, this outcome is the result of a bug fix (where a use of an unknown attribute is now being rejected by the compiler).

Compiler team has decided that Adding a warning cycle for this bug at this point is not warranted