timely-0.1.15 beta regression

[est31]

timely version 0.1.15 regressed from stable to beta (full log):

Aug 04 00:58:05.190 INFO kablam! error: no rules expected the token `EventStreamLogger<OperatesEvent, File>`
Aug 04 00:58:05.190 INFO kablam!    --> src/logging.rs:160:35
Aug 04 00:58:05.190 INFO kablam!     |
Aug 04 00:58:05.190 INFO kablam! 160 |               __thread_local_inner!($t, $init);
Aug 04 00:58:05.190 INFO kablam!     |                                     ^^
Aug 04 00:58:05.190 INFO kablam! ...
Aug 04 00:58:05.190 INFO kablam! 164 | / thread_local!{
Aug 04 00:58:05.190 INFO kablam! 165 | |     /// Logs operator creation.
Aug 04 00:58:05.190 INFO kablam! 166 | |     pub static OPERATES: EventStreamLogger<OperatesEvent, File> = EventStreamLogger::new();
Aug 04 00:58:05.190 INFO kablam! 167 | |     /// Logs channel creation.
Aug 04 00:58:05.190 INFO kablam! ...   |
Aug 04 00:58:05.190 INFO kablam! 178 | |     pub static GUARDED_PROGRESS: EventStreamLogger<bool, File> = EventStreamLogger::new();
Aug 04 00:58:05.190 INFO kablam! 179 | | }
Aug 04 00:58:05.191 INFO kablam!     | |_- in this macro invocation

cc @frankmcsherry

[arielb1]

cc @jseyfried

[arielb1]

Actually, maybe this is using the "pirate" __thread_local_inner from std?

[arielb1]

Yup. This compiles on stable and not on beta:

fn xyz() {
   __thread_local_inner!(Vec<u32>, Vec::new());
}

fn main() {}

Please don't use obviously private libstd macros.

[arielb1]

timely 0.2.0 uses thread_local!, stabilized in 1.11.0. Maybe we want to do this in timely 0.1.16?

[durka]

It looks like old versions of timely copied the new thread_local macro from my PR before it was merged. Unfortunately(?) stability attributes on macros don't do anything.

[alexcrichton]

The libs team discussed this today and we decided that we're not going to actively try to fix this regression, so closing.