Looks sensible at a first glance. I just hope that the extension plumbing code is really operationally equivalent …

Looks sensible at a first glance. I just hope that the extension plumbing code is really operationally equivalent …

This can be tested by running something like nix-instantiate --expr (import ./. { system = /.../; }).haskellPackages`, before and after the change and checking for any differences in the reported list of drvs.

I'd appreciate if someone could review this changes in depth. I've looked over it already, but it's easy to miss stuff when doing refactors like this. I know it's tedious and I'd be very grateful :)

Also I've confirmed that e8ac57e does not change a single outpath (using nix-env-diff).

May be best to split the hash-changing changes into a separate PR? That way you can use ofborg results.

May be best to split the hash-changing changes into a separate PR? That way you can use ofborg results.

With the way the darwin overrides were done I don't think it's entirely possible and it'd be a nasty rebase nevertheless. Not sure how aarch64 fits into the picture for ofborg, I've picked some of the commits onto the haskell-updates branch now.

Fair enough.

I have stared at this for quite a while and this all seems very reasonable.

I have also tried to check that you didn‘t lose any overrides.

Thx very much! I like this refactor!