NixOS
Get rid of `impureOutputHash`; fix possible bug
Motivation
I do not believe there is any problem with computing hashDerivationModulo the normal way with impure derivations.
Conversely, the way this used to work is very suspicious because two almost-equal derivations that only differ in depending on different impure derivations could have the same drv hash modulo. That is very suspicious because there is no reason to think those two different impure derivations will end up producing the same content-addressed data!
Context
Checklist for maintainers
Maintainers: tick if completed or explain if not relevant
- [ ] agreed on idea
- [ ] agreed on implementation strategy
- [ ] tests, as appropriate
- functional tests -
tests/**.sh - unit tests -
src/*/tests - integration tests
- functional tests -
- [ ] documentation in the manual
- [ ] code and comments are self-explanatory
- [ ] commit message explains why the change was made
- [ ] new feature or bug fix: updated release notes
The point of making DerivationType::ContentAddressed a record (and derivation type no longer an enum) was that we already handled this case and didn't need a new variant.
I don't agree that this already handles that case. An impure derivation is not conceptually the same as a derivation that isn't sandboxed and not fixed-output. You could imagine derivations that are not sandboxed and not fixed-output, but are still expected to have a reproducible result (e.g. a derivation that uses ccache to build something remotely).
We conversely, one might want to cache actually impure derivations more normally, in a "per transaction" substitution map.
Still, I will split this up so the second commit doesn't hold back the first.
In the meeting today @edolstra mentioned that, as we discussed in the past, we don't necessary want to make impure derivations work "just like" CA derivations.
A thing I forgot to point out was that floating CA derivations are already not handled specially. The only special case in hashDerivationModulo is for fixed CA derivations -- input address and floating CA are handled the same way.
Impure derivations are certainly not fixed output, so I don't think we are going to run into any problems here.
This pull request has been mentioned on NixOS Discourse. There might be relevant details there:
https://discourse.nixos.org/t/2024-02-12-nix-team-meeting-minutes-212-5/60216/1
