Hey all, I'm working on a bug I encountered related to a race condition between GC processes, but I'm a little stuck on a C++ thing (not my forte). All links will be to 2.24.12 since that's what I'm on at the moment, but I suspect the error is still present in recent versions.

If the GC socket is disconnected, you get errno = 32 here, and you throw SysError: https://github.com/NixOS/nix/blob/f22359ba1af6d976d248318aa14e6a6326682f5c/src/libutil/unix/file-descriptor.cc#L43

So, you would expect that to get caught in addTempRoot here: https://github.com/NixOS/nix/blob/f22359ba1af6d976d248318aa14e6a6326682f5c/src/libstore/gc.cc#L135

However, if I put some debug("I'm here")-style comments in that catch block, I don't see the exception caught. I know that the exception is being throw in writeFull because I can put a try-catch block around the body, catch the SysError, throw a std::exception instead, and catch that at this location in addTempRoot. It seems like "throw SysError here, catch SysError there" should be pretty straightforward, but it's not working, and I'm out of my C++-depth if there are any nuances involved.

:: (nixbld13) → 
  UID     PID    PPID STIME     TIME COMMAND
30013  978082  977874 Apr30 00:00:00 bash -e /nix/store/vj1c3wf9c11a0qs6p3ymfvrnsdgsdcbq-source-stdenv.sh /nix/store/shkw4qm9qcw5sc5n1k5jznc83ny02r39-default-builder.sh
30013 2051738  978082 Apr30 00:01:09 /nix/store/fqm9bqqlmaqqr02qbalm1bazp810qfiw-python3-3.12.9/bin/python3.12 /nix/store/jvcm0xph04gl36xmwclfkg26736javak-meson-1.7.0/bin/meson test --no-rebuild --p
30013 2051835 2051738 Apr30 00:00:00 /nix/store/9nw8b61s8lfdn8fkabxhbz0s775gjhbr-bash-5.2p37/bin/bash -x -e -u -o pipefail gc-auto.sh
30013 2054090  978082 Apr30 00:00:00 /nix/store/9nw8b61s8lfdn8fkabxhbz0s775gjhbr-bash-5.2p37/bin/bash -x -e -u -o pipefail gc-non-blocking.sh
30013 2054096 2054090 Apr30 00:00:00 nix-store --gc -vvvvv