I did not build anything but fine by me.

Both jdk8 and jdk11 build properly for me with this PR.

Someone please root cause this..

Okay, I think I know what's happening... Just need to confirm something 👀 🤣

Posted my root cause analysis in #425323 (comment)

Well that's extremely cursed.

I guess we could revert this and swap the outputs around here? I guess that's only a band-aid for the quite frankly spooky postFixup code...

Swapping the outputs doesn't solve the issue... The outputs array is converted by Nix to a dictionary ( 😩 )

[ "out" "jre" ] -> {"jre":"/nix/store/...","out":"/nix/store/..."}
[ "jre" "out" ] -> {"jre":"/nix/store/...","out":"/nix/store/..."}

You can confirm this with nix develop .#jdk8 and then cat $NIX_ATTRS_JSON_FILE or cat $NIX_ATTRS_SH_FILE

Huh. I always thought that:

(in nixpkgs)
$ nix-instantiate --json --eval -A openjdk8.outputs
["out","jre","debug"]

That's true, but outputs has a special meaning when __structuredAttrs = true...

By the way, I found this in the Nix manual (search for "outputs") https://nix.dev/manual/nix/2.30/development/json-guideline.html?highlight=outputs#json-guideline

OK, so I think it's safe to say this is actually properly documented, just not implemented correctly in nixpkgs? In that case we should probably raise an issue about that?

OK, so I think it's safe to say this is actually properly documented, just not implemented correctly in nixpkgs? In that case we should probably raise an issue about that?

I think it's the best approach.

There are some questions that should be answered before a global fix is proposed:

• is getAllOutputNames supposed to return output names in order or is this something that was never guaranteed and derivations that rely on this are simply wrong and need to be fixed?
• what is the best way to fix this? (I have some thoughts on this but I would like to hear more experienced people on this)

That and taking into consideration that a global fix would cause a mass rebuild is enough, imo, to warrant a "rfc" (idk if nixpkgs has rfcs or not) on this matter

I don't think relying on output order is the way tbh. The postFixup can be fixed to not do that fairly easily, so it probably should.

Good point! I also found this

Do you have any ideas on how to improve the postFixup hook?

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/nixos-rebuild-switch-building-openjdk-for-no-reason-and-failing-because-of-cyclic-dependency/66917/6