fish: default --no-config prevents history from being saved or recalled

[adrian-gierakowski]

I think it might be better to remove it and set XDG_CONFIG_HOME to empty instead

[adrian-gierakowski]

XDG_CONFIG_HOME to empty instead

I guess that's a bad idea since other programs within the shell sessions would inherit it

Might be an upstream bug: fish-shell/fish-shell#12711

[BirdeeHub]

Does it prevent it?

Or does it just not save history unless you set the option of where to save it and enable it or whatever?

It may just be that it needs configuration to save it?

If that is the case, maybe we should add an option with a default value for whatever that fish option is.

[BirdeeHub]

Oh.

I just looked at your issue you filed for fish

Yeah it is definitely enabling private mode.

It probably should not be doing that.

I think that is an upstream bug yeah.

[adrian-gierakowski]

Oh.

I just looked at your issue you filed for fish

Yeah it is definitely enabling private mode.

It probably should not be doing that.

I think that is an upstream bug yeah.

Yeah, for now I'm running a fork which removes private mode from --no-config

[ormoyo]

So that's why I didn't have history while testing.
I thought I had misconfigured the shell or because the nature of the "nix run" command.

Maybe we should point that out in meta.description until it's fixed in fish?

[BirdeeHub]

probably a good idea. Alongside instructions to do flags."--no-config" = false; and move your regular fish config file to workaround it if desired.

[adrian-gierakowski]

The problems I encountered are:

1. There is no flag (apart from --no-config) or env var which can be used to prevent fish from loading config from /etc
2. If you use XDG_ env vars to set custom config and history dirs, it will also affect the programs you run within fish

So there is not really a good workaround, apart from forking fish

[BirdeeHub]

ah. fun.

The patched version you said you are using, have you submitted it as a PR?

[adrian-gierakowski]

ah. fun.

The patched version you said you are using, have you submitted it as a PR?

No, I was waiting for maintainers to acknowledge the problem. In the meantime I just patched it inline in an overlay:

      fish = super.fish.overrideAttrs (oldAttrs: {
        patches = (oldAttrs.patches or [ ]) ++ [
          (self.writeText "fish-remove-no-config-private.patch" ''
            --- a/src/bin/fish.rs
            +++ b/src/bin/fish.rs
            @@ -275,8 +275,6 @@
                         'l' => opts.is_login = true,
                         'N' => {
                             opts.no_config = true;
            -                // --no-config implies private mode, we won't be saving history
            -                opts.enable_private_mode = true;
                         }
                         'n' => opts.no_exec = true,
                         RUSAGE_ARG => opts.print_rusage_self = true,
          '')
        ];
      });

but it looks like someone else took the initiative: fish-shell/fish-shell#12741 :D

[krobelus]

I don't see a problem with this patch, but why are you using --no-config for an interactive shell?
Won't users expect that ~/.config/fish/config.fish is sourced etc?

[adrian-gierakowski]

@krobelus

The purpose of wrappers in this repo is to package executables and their configs into self contained units, which work the same way anywhere without requiring any ambient state present on the system.