LGTM. I'll set up a hydra jobset

https://hydra.nixos.org/jobset/nix/PR-1643

Could we have a test? :)

@domenkozar Sorry for taking a long time, I've added a test!

I've rebased the patches but for some reason nix-copy-closure test fails. Any ideas why? I'd very much like to see this fixed in Nix 2.0.

@abbradar That's a bug in the test that I thought @edolstra had fixed... @edolstra should that block merging this?

@abbradar You need the nix-2.0 branch of Nixpkgs.

Don't know if changing the behaviour of readFile just before release is a good idea though.

@edolstra I see this as a bug because it can silently break users' systems after nix-collect-garbage if they have fancy enough configuration (example). If I understand correctly this fix (in theory) just adds more paths into closure of some outputs so it should be fairly harmless. However I don't feel qualified to say this after I have already broken things with an older fix ;)

@edolstra I'd really prefer we branch off 2.0 if we're going to start delaying PRs because they shouldn't be in 2.0... That way we can keep moving despite being feature frozen (or perhaps feature chilled)

Can we proceed with this now that 2.0 release has been branched?

@edolstra Any concerns?

So, any movement here? I've run into this bug yet again, it's pretty annoying.

I marked this as stale due to inactivity. → More info

I have rebased this and resolved conflicts. Not sure if using printStorePathSet here is a good idea.

I understand that life and other priorities can get in the way, but it still feels frustrating to keep rebasing this patchset for so many years without any movement. Maybe now that 2.0 is released long time ago we can move forward with this?

The risk with this PR is that it can cause evaluation to produce a different result depending on whether the input is in the Nix store (where it could have references). In the past we had issues like that with Hydra (which copies sources to the Nix store). However, it should be okay so long as the source tree has no references.

Thanks! Let's hope this won't break corner cases for people like it did before @shlevy helped fix it. If anything arises, ping me.