nh shell and nh run?

[iFreilicht]

It's been a major annoyance for a long time with nix-command that you always have to write out the name of the flake explicitly when using nix shell or nix run.

The popular nix-dram patches this functionality in, and it was requested for addition to lix, but as this is a breaking change, it was rejected.

Maybe this would be a good functionality to add to nh?

Being able to simply run nh shell cargo gcc would be a QoL improvement and not too difficult to implement. nh search already always uses nixpkgs, so there's less of an argument to be made against this being incompatible with other flakes.

[viperML]

I don't want to implement that unless we can get completion support, and I don't have the bandwidth to look into implementing that.

[iFreilicht]

I understand that. I may have a look at implementing this myself, but if anyone else reads this feel free to take it on.

Playing around a bit, it seems feasible to just shell out to nix eval. time nix eval nixpkgs#legacyPackages.aarch64-darwin --apply "builtins.attrNames" reports an average of 1.4s of runtime on my machine, so it would be an option to just read that, cache it in /tmp, and then search through that list.

You could search with eval directly: nix eval --json nixpkgs#legacyPackages.aarch64-darwin --apply 'pkgs: builtins.filter (x: (builtins.match "^ne.*" x) != null) (builtins.attrNames pkgs)'. This takes around half a second, no matter the query, so caching the full list will likely be quicker.

[viperML]

From a quick search, this seems to be the way to implement it:

https://docs.rs/clap_complete/latest/clap_complete/engine/struct.ArgValueCompleter.html
clap-rs/clap#1232
clap-rs/clap#3166

[viperML]

Also note that after #151, nh will not be a flake-first tool, but rather support -f, -A, etc.