Revamp nix flake for nix-tools

[andreabedini]

This PR revamps the nix-tools flake and adds a binary tarball with static builds (for now only on x86_64-linux).

We already have a Hydra jobset monitoring the nix-tools (sub?)flake, refreshing every hour:
https://ci.zw3rk.com/jobset/input-output-hk-haskell-nix/nix-tools

I did a brief test of Haskell.nix with this branch and it seems ok. Let's see what the main CI says.

[andreabedini]

Integration with Haskell.nix still to do.

[angerman]

We should also try to compile some perf metrics. With/without; hot/cold.

[andreabedini]

@hamishmack what's the role of nix-tools/nix/hix.nix? is it needed? nix-tools is loaded in haskell.nix as a hix project.

[hamishmack]

You can remove nix/hix.nix (it does not look like we will need to pass any project args). If we were passing modules for something then we would want to keep it to make the hix commands see that.

[hamishmack]

nix-tools is loaded in haskell.nix as a hix project.

Will that be replaced with the derivations to download the static nix-tools?

[andreabedini]

I am going to include the required changes in haskell.nix in this PR.

[michaelpj]

Am I missing the motivation here? The tarball will still need to be built by haskell.nix, so won't we still have the same build-time work to do?

[andreabedini]

Will that be replaced with the derivations to download the static nix-tools?

@hamishmack The plan would be for haskell.nix optionally download a static build of nix-tools rather that bulding it as part of the bootstrap process. We can then asses whether this helps (e.g. with adding a new version of GHC).

Am I missing the motivation here? The tarball will still need to be built by haskell.nix, so won't we still have the same build-time work to do?

The tarball will still need to be built by haskell.nix that's correct. The difference is that, having its own flake, it can be built separately. I hope this can simplify (and shorten) the bootstrap process.