Remote builds break when remote shell is not-POSIXy-enough #805

New issue

Closed

opened 2025-04-22 06:11:00 +00:00 by k900 · 3 comments

k900 commented

2025-04-22 06:11:00 +00:00

Member

Describe the bug

This was surfaced in my setup by a combination of nushell and f92235e1d2.

Lix runs ssh -x user@remote "echo started; nix-daemon --stdio", and then waits for "started" to be printed. However, on nushell, the semicolon operator swallows stdout, so it never is printed, and we wait forever.

This also has the potential to explode in other interesting ways on other non-POSIX shells, so we should probably think of a better solution (SSH subsystem? hello message in the daemon? rewrite the whole thing to use libssh? idk).

Steps To Reproduce

nix store ping --store ssh://user@host, where user@host's login shell is nushell.

Additional context

Fuck my life.

## Describe the bug This was surfaced in my setup by a combination of nushell and f92235e1d2ebe7f5fa8048543c4f59661d52f787. Lix runs `ssh -x user@remote "echo started; nix-daemon --stdio"`, and then waits for `"started"` to be printed. However, on nushell, the semicolon operator swallows stdout, so it never _is_ printed, and we wait forever. This also has the potential to explode in other interesting ways on other non-POSIX shells, so we should probably think of a better solution (SSH subsystem? hello message in the daemon? rewrite the whole thing to use libssh? idk). ## Steps To Reproduce `nix store ping --store ssh://user@host`, where `user@host`'s login shell is nushell. ## Additional context Fuck my life.

k900 added the

bug

label

2025-04-22 06:11:00 +00:00

k900 added reference main

2025-04-22 06:11:05 +00:00

k900 commented

2025-04-22 06:16:47 +00:00

Author

Member

As a temporary workaround, I've created a new user that has its shell set to bash, and use it for remote builds only.

k900 commented

2025-04-22 06:42:34 +00:00

Author

Member

Never mind, that breaks nix-copy-closure. The new temporary workaround is https://gerrit.lix.systems/c/lix/+/3046 which is terrible and evil and I hate it.

Never mind, that breaks `nix-copy-closure`. The new temporary workaround is https://gerrit.lix.systems/c/lix/+/3046 which is terrible and evil and I hate it.

pennae added this to the 2.93 milestone

2025-04-25 17:26:03 +00:00

pennae added the

Affects/Only nightly

label

2025-04-25 17:26:10 +00:00

lix-bot commented

2025-04-26 07:44:33 +00:00

Member

This issue was mentioned on Gerrit on the following CLs:

commit message in cl/3046 ("libstore/ssh: fix the SSH connectivity check with non-POSIXy shells")

This issue was mentioned on Gerrit on the following CLs: * commit message in [cl/3046](https://gerrit.lix.systems/c/lix/+/3046) ("libstore/ssh: fix the SSH connectivity check with non-POSIXy shells")