downloads: Add entry for NixOS

[maralorn]

As discussed in #170 (comment)

I have appended the NixOS entry to be polite. I wonder if there could be a more principled approach to ordering the entries.
At the moment I guess it’s "felt importance"? Alphabetically might be an improvement. Sadly the Haskell survey does not ask for the distributions that people primarily use.

Note: It is our aim to improve and move the documentation. Still the current documentation is a lot better than nothing. So I think we should put the link here for the while. I will update it as soon as we made progress on this.

[tomjaguarpaw]

Thanks! I'm not a Nix user, but as I understand it, Nix and NixOS are slightly different. Are you sure you're happy having NixOS as a subentry under Linux or should it belong at the top level? (I don't mind either way)

[davean]

Is nix still "jailbreak"ing packages and removing their version bound requirements? I know I've had data loss because they "didn't believe me" that I required a newer version of 'directory' in the past. If they're still doing that at all I'd want a very large warning.

[maralorn]

Thanks! I'm not a Nix user, but as I understand it, Nix and NixOS are slightly different. Are you sure you're happy having NixOS as a subentry under Linux or should it belong at the top level? (I don't mind either way)

I have considered this and given this a second attempt.

Is nix still "jailbreak"ing packages and removing their version bound requirements? I know I've had data loss because they "didn't believe me" that I required a newer version of 'directory' in the past. If they're still doing that at all I'd want a very large warning.

Yes, we still do jailbreaking, mostly because devs don‘t keep up with relaxing their bounds. Your story sounds terrible and is a good reminder to be careful with this. I cannot promise that this has actually become better, because I have not been in the business very long. But these days we are careful about this. In general when bounds seem weird, especially lower bounds, we check changelogs for explanations/breaking changes and contact upstream about it. We have policies helping with this, which we enforce in code review. At the very least I can say that Nix is relatively wide spread in the Haskell community and I haven‘t seen any complaints about issues like that in the last years.

I would not like to have a warning like that.

[tomjaguarpaw]

Thanks! Looks fine to me. Would be great to have a review by someone familiar with Nix (@TikhonJelvis, @rebeccaskinner?).

[TikhonJelvis]

Looks fine to me. Those seems like reasonable links and more than enough detail for the haskell.org page—it's good to leave more thorough instructions on the NixOS/Nixpkgs side.