Open lunik1 opened 4 months ago
I have not seen this error before, and it does not look like a Nix, Arion or Haskell error. Podman previously did not invest in a major update migration (2-3? 3-4? I don't recall.), so I'm not too surprised that something would break. Do you have state from 4.0 on your machine?
The arion tests pass for nixos-unstable
, podman 5.0.2.
I'll dig into it a bit more. Rolling back to a pre-podman 5.0 nixpkgs commit did fix it for me.
Still not sure what's going on entirely, but I now see this on two systems. The error message looks like it is coming from Go. I have tried a podman system prune
and reboot to get clear any old podman state.
Running strace
shows what command sets it off
[pid 76385] execve("/nix/store/491k4bspg098nm2jqk9p81hrn4wx5d01-docker-compose-1.29.2/bin/.docker-compose-wrapped", ["/nix/store/491k4bspg098nm2jqk9p8"..., "--file", "./.tmp-arion-docker-compose76199"..., "--project-name", "dionysus", "up"], 0x2225640 /* 31 vars */) = 0
ERROR: runtime error: index out of range [0] with length 0
and indeed running
/nix/store/491k4bspg098nm2jqk9p81hrn4wx5d01-docker-compose-1.29.2/bin/.docker-compose-wrapped --file docker-compose.yaml up
where the docker-compose.yaml is generated by arion show
triggers the same error.
Another thing you could try is start with an empty docker compose file and then add things until it fails, or insert half of it and do like a bisect of sorts.
Even the trivial
version: '3.4'
triggers it. But I have confirmed it is not an arion bug. It seems to be a bug between docker compose v1 and podman v5.0. I only see this error when using docker-compose v1 and from nixpkgs commit f9be31de1b7b5f3738f23a0c04a20bc524cab30c
onwards, which introduced podman v5.
I have not seen this reported elsewhere yet, but fyi podman no longer test against docker compose v1, see https://github.com/containers/podman/pull/22421
I'm also fairly certain now that https://github.com/hercules-ci/arion/issues/230 is a docker-compose v1 bug, too.
I get the above runtime error when running
aup
on the latest unstable NixOS, reproducible with the "minimal" example in this repo. I haven't had time to confirm this, but I suspect it could be connected to the update to podman 5.0.