Evaluating the Nickel flake is very slow

[thufschmitt]

Describe the bug

Evaluating the default output of the flake takes a very long time (roughly 18s on my machine). This is very annoying for projects like Organist that depend on Nickel by its flake as it adds a lot to their Nix evaluation time. And in the case of Organist, all the projects using Organist will have to pay for that.

The nickel derivation from Nixpkgs doesn't have this problem.

To Reproduce

$ nix run p#hyperfine -- -L FLAKEREF nickel\#default,nixpkgs\#nickel "nix eval {FLAKEREF}.outPath --option eval-cache false" --warmup 2
Benchmark 1: nix eval nickel#default.outPath --option eval-cache false
  Time (mean ± σ):     18.530 s ±  0.077 s    [User: 13.057 s, System: 2.872 s]
  Range (min … max):   18.363 s … 18.675 s    10 runs
 
Benchmark 2: nix eval nixpkgs#nickel.outPath --option eval-cache false
  Time (mean ± σ):      2.269 s ±  0.021 s    [User: 1.591 s, System: 0.415 s]
  Range (min … max):    2.239 s …  2.295 s    10 runs
 
Summary
  'nix eval nixpkgs#nickel.outPath --option eval-cache false' ran
    8.16 ± 0.08 times faster than 'nix eval nickel#default.outPath --option eval-cache false'

Expected behavior

Something fast :)

More seriously, I don't know whether that's something worth fixing since the evaluation time overhead is very likely due to Crane, and is hopefully repaid by a better build-time caching for iterative development. But it's worth being aware of it.

Environment

• OS name + version:
• Version of the code: 1921c316ad81bca8100c3a0c6ae2e3da974cdd51

Additional context

ø

[yannham]

In your opinion, what could be a way forward? Would adding an output nickel-fast that doesn't use crane make sense?

[thufschmitt]

Honestly: I have no idea :)

For Organist, I've vendored a modified copy of the Nix expressions from Nixpkgs (here). If you're willing to maintain something like that then yes, exposing them as a nickel-fast output could make sense.

Otherwise, just saying that it's a price to pay for living at HEAD and that the releases get a fast expression from Nixpkgs would be a fine tradeoff for me. I was mostly opening the issue to make sure that you were aware that there's a tradeoff to make.