rustc / cargo panics on NixOS when compiling backtrace

[tazjin]

Hey!

System info:

~ $ cat /etc/os-release 
NAME=NixOS
ID=nixos
VERSION="17.09.2182.7f6f0c49f0 (Hummingbird)"
VERSION_CODENAME=hummingbird
VERSION_ID="17.09.2182.7f6f0c49f0"
PRETTY_NAME="NixOS 17.09.2182.7f6f0c49f0 (Hummingbird)"
HOME_URL="https://nixos.org/"
SUPPORT_URL="https://nixos.org/nixos/support.html"
BUG_REPORT_URL="https://github.com/NixOS/nixpkgs/issues"
~ $ rustc --version
rustc 1.22.1 (05e2e1c41 2017-11-22)
~ $ cargo --version
cargo 0.23.0 (61fa02415 2017-11-22)

---

I'm getting the following panic when building backtrace:

   Compiling rustc-demangle v0.1.5
   Compiling libc v0.2.33
   Compiling cc v1.0.3
   Compiling cfg-if v0.1.2
   Compiling backtrace-sys v0.1.16
   Compiling backtrace v0.3.4
thread '<unnamed>' panicked at 'failed to acquire jobserver token: Resource temporarily unavailable (os error 11)', /checkout/src/librustc_trans/back/write.rs:1619:28
note: Run with `RUST_BACKTRACE=1` for a backtrace.
error: failed to acquire jobserver token: Resource temporarily unavailable (os error 11)

error: Could not compile `backtrace`.

To learn more, run the command again with --verbose.

---

This is reproducible for me by just running cargo init and adding backtrace = "0.3.4" as the only dependency to the project.

Interestingly running cargo build a second time will succeed! I've tried setting build parallelism flags, e.g. -j1 to disable parallel builds, but it doesn't seem to make a difference.

Will try to add more information to the issue as debugging continues, but thought that you may have an idea what this is about :)

[alexcrichton]

Could you confirm where you're acquiring Cargo from? This was fixed in the jobserver crate 0.1.8 and it may not be pulled into Cargo locally for you perhaps?

[tazjin]

It's the nixpkgs-overlay which - afaik - uses the same channels as rustup. I'll double-check that in a bit

[tazjin]

@alexcrichton Which version of Cargo is supposed to have jobserver >= 0.1.8? As far as I can tell both Cargo master and the last two releases use 0.1.6.

Trying to figure out what exactly has caused this to suddenly be an issue, because reverting to Rust 1.21.0 (release from 2017-10-12) doesn't fix this again 🤔

[alexcrichton]

@tazjin a specification of jobserver = "0.1.6 is actually >= 0.1.6, < 0.2 so you need to look in the lock file to see what was actually selected. On Rust right now at least that's 0.1.8. That may not have been updated for 1.21.0 though and just 1.22.0

[matthiasbeyer]

It seems I'm running into this. Reported here.

[matthiasbeyer]

After building cargo 0.18 from source I was able to release my crates.

[pmahoney]

Just another datapoint, unrelated to backtrace-rs: I'm seeing the same error also under NixOS building letsencrypt-rs using rustc 1.21.0, cargo 0.22.0. Repeated builds show the error happening while compiling different dependencies (the error is not deterministic).

Running cargo build with -j1 did not help, but running with -j19 seems to allow everything to build OK, so that may be a workaround if anyone else is running into the same bug. (-j19 was just a large number I happened to choose, there is no reason behind it; watching htop while building, I don't think I ever saw more than five instances of rustc).

I'm not using the Mozilla overlay, just the stock rust packages in NixOS 18.03pre123130.1bc288591ea.

[alexcrichton]

I think this was a bug in rustc, so closing