Issue Review

Copy local flakes

[tom]: forgot to push "experimental branch"
needs final review https://github.com/NixOS/nix/pull/6233/commits/e1f308a1ec3c395cd4978b45400f7a45adcea0dc
@regnat reviewed it, requesting more backwards-compat
@tomberek addressed that (just pushed), will need another round of review

Still waiting on the debugger PR

Problem: Nix takes into account the git repo it’s working in. Which can be nice, but also creates some friction and unintuitive behavior

@eelco’s work on lazy trees makes this nicer because it’s possible to display a better error message saying that the file exists but isn’t known to Git, so it should be checked-in
But not enough: Since Nix knows how to fix it, it could just fix it himself
Occasionnally, ppl want to use untracked files in a flake
Also means that adding a Git repo changes significantly the behavior of Nix
The behaviour is weird wrt the mental model of Git: The thing that Nix sees is the commit that would have been produced by git commit --all. Which isn’t intuitive at all
This decision isn’t documented anywhere
Motivations behind this design choice:
- Tying Nix to Git is done for hermeticity
- Copying everything to the store is (was) the easiest way to implement this hermetic model
Alternatives
- Stick to the last commit
- Only consider the staged chunks
- Consider the whole non-ignored workspace
…
That boils down to a matter of Git wokflow:
- Ppl who keep a clean worktree will prefer having everything (non-ignored) included
- Other ppl prefer having to explicitely add stuff
Other use-case: people who can’t commit the flake.nix (or other Nix files) but still want to use the flakes in-tree
Proposal by @layus : Keep mostly the same behavior, except that when the evaluator wants to eval a file that’s not tracked in Git it should just warn about it and evaluate it anyways

Regardless of what ends-up being done, the motivation should be recorded

Progress-bar (@garbas)