Regression in time to compilation failure

[jrobsonchase]

There seems to have been a regression a couple of weeks ago in the time it takes rustc to exit after reporting an error. Using the nightly from 2016-10-21, build failures exit in under half a second, but on the nightlies after that (starting on 2016-10-25), they hang for ~20 seconds.

jchase@tethys test_regression $ rustup default nightly-2016-10-21 && time rustc src/lib.rs; echo -e '\n\n'; rustup default nightly && time rustc src/lib.rs
info: using existing install for 'nightly-2016-10-21-x86_64-unknown-linux-gnu'
info: default toolchain set to 'nightly-2016-10-21-x86_64-unknown-linux-gnu'

  nightly-2016-10-21-x86_64-unknown-linux-gnu unchanged - rustc 1.14.0-nightly (f09420685 2016-10-20)

error: expected one of `!` or `::`, found `#`
 --> src/lib.rs:2:1
  |
2 | #[cfg(test)]
  | ^

error: aborting due to previous error


real	0m0.237s
user	0m0.220s
sys	0m0.013s



info: using existing install for 'nightly-x86_64-unknown-linux-gnu'
info: default toolchain set to 'nightly-x86_64-unknown-linux-gnu'

  nightly-x86_64-unknown-linux-gnu unchanged - rustc 1.14.0-nightly (5665bdf3e 2016-11-02)

error: expected one of `!` or `::`, found `#`
 --> src/lib.rs:2:1
  |
2 | #[cfg(test)]
  | ^

error: aborting due to previous error


real	0m22.905s
user	0m22.827s
sys	0m0.070s

jchase@tethys test_regression $ cat src/lib.rs 
asdf
#[cfg(test)]
mod tests {
    #[test]
    fn it_works() {
    }
}

Fixing the error (asdf -> fn main() {}) results in the expected:

jchase@tethys test_regression $ rustup default nightly-2016-10-21 && time rustc src/lib.rs; echo -e '\n\n'; rustup default nightly && time rustc src/lib.rs
info: using existing install for 'nightly-2016-10-21-x86_64-unknown-linux-gnu'
info: default toolchain set to 'nightly-2016-10-21-x86_64-unknown-linux-gnu'

  nightly-2016-10-21-x86_64-unknown-linux-gnu unchanged - rustc 1.14.0-nightly (f09420685 2016-10-20)


real	0m0.172s
user	0m0.133s
sys	0m0.033s



info: using existing install for 'nightly-x86_64-unknown-linux-gnu'
info: default toolchain set to 'nightly-x86_64-unknown-linux-gnu'

  nightly-x86_64-unknown-linux-gnu unchanged - rustc 1.14.0-nightly (5665bdf3e 2016-11-02)


real	0m0.181s
user	0m0.163s
sys	0m0.007s

[oli-obk]

It's not just rustc. If a run-pass test fails with compiletest-rs, the same hang can be seen for reporting the stacktrace of the failed test

[jrobsonchase]

Any movement on this? Still seeing the same behavior on nightly - stable and beta are fine.

[jrobsonchase]

Just a little more information: When I compile rust myself, I don't see the issue - even from the same commit. strace shows a huge jump in calls to mmap during the hang-time from ~600 calls with beta to ~9000 calls with nightly.

Edit: Scratch the caveat around compiling it myself - I was doing it wrong.

[sfackler]

I think someone ran into something similar before and it ended up being an issue with libbacktrace's custom allocator implementation.