Use YAML for config

[shivaraj-bh]

resolves #343

If the flake url is a local path and om.yaml exists in the project root, omnix will not invoke nix eval .#om.

Benchmarked om develop —pre-shell . && om develop —post-shell . in https://github.com/nammayatri/nammayatri, using YAML config is 14 times faster:

❯ nix run github:nixos/nixpkgs/nixpkgs-unstable#hyperfine -- --warmup 10 './with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .' './without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .'
Benchmark 1: ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .
  Time (mean ± σ):      48.8 ms ±   1.2 ms    [User: 20.7 ms, System: 30.4 ms]
  Range (min … max):    46.9 ms …  52.8 ms    59 runs

Benchmark 2: ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .
  Time (mean ± σ):     725.6 ms ±  13.6 ms    [User: 514.6 ms, System: 258.1 ms]
  Range (min … max):   708.6 ms … 751.8 ms    10 runs

Summary
  ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell . ran
   14.88 ± 0.46 times faster than ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .

om.yaml used for the benchmark:

health:
  default:
    caches:
      required:
        - "https://nammayatri.cachix.org"
    direnv:
      required: true
    system:
      min_ram: "24G"

[srid]

This looks good so far. Can we improve the OmConfig::get interface so as to resolve and avoid unnecessary nix command runs? See #339 (comment)

[shivaraj-bh]

This looks good so far. Can we improve the OmConfig::get interface so as to resolve and avoid unnecessary nix command runs? See #339 (comment)

Aren’t those nix commands the result of the checks in preShell?

omnix/crates/omnix-develop/src/core.rs

Lines 45 to 57 in af87950

	pub async fn develop_on_pre_shell(prj: &Project) -> anyhow::Result<()> {
	// Run relevant `om health` checks
	let health = NixHealth::from_om_config(&prj.om_config)?;
	let nix_info = NixInfo::get()
	.await
	.as_ref()
	.with_context(|| "Unable to gather nix info")?;
	let mut relevant_checks: Vec<&'_ dyn Checkable> =
	vec![&health.nix_version, &health.rosetta, &health.max_jobs];
	if !health.caches.required.is_empty() {
	relevant_checks.push(&health.trusted_users);
	};

[srid]

Aren’t those nix commands the result of the checks in preShell?

While we can't avoid running nix commands in preShell (for now), we certainly can avoid that in postShell (which too will be run as part of direnv invocation):

[shivaraj-bh]

While we can't avoid running nix commands in preShell (for now), we certainly can avoid that in postShell (which too will be run as part of direnv invocation):

Fixed in 1b7c152.

❯ just run develop --stage post-shell .
cargo run -p omnix-cli develop --stage post-shell .
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.28s
     Running `target/debug/om develop --stage post-shell .`
⌨️  Preparing to develop project: .

🍾 Welcome to the omnix project

To run omnix,

────────────────────
just watch <args>
────────────────────

(Now, as you edit the Rust sources, the above will reload!)

🍎🍎 Run 'just' to see more commands. See https://nixos.asia/en/vscode for IDE setup.

[shivaraj-bh]

Did a benchmark of running both postShell and preShell comparing master with changes in yaml-config branch using https://github.com/sharkdp/hyperfine

❯ nix run github:nixos/nixpkgs/nixpkgs-unstable#hyperfine -- --warmup 10 './with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .' './without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .'
Benchmark 1: ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .
  Time (mean ± σ):     208.7 ms ±   3.7 ms    [User: 6.6 ms, System: 2.3 ms]
  Range (min … max):   203.5 ms … 216.7 ms    14 runs
 
Benchmark 2: ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .
  Time (mean ± σ):     603.8 ms ±   6.9 ms    [User: 93.0 ms, System: 92.1 ms]
  Range (min … max):   592.2 ms … 612.5 ms    10 runs
 
Summary
  ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell . ran
    2.89 ± 0.06 times faster than ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .

Here are individual outputs of with-yaml and without-yaml respectively:

❯ ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .
⌨️  Preparing to develop project: .
❄️  nix --accept-flake-config --version️
❄️  nix --extra-experimental-features nix-command --accept-flake-config show-config --json️
✅ Nix environment is healthy.
⌨️  Preparing to develop project: .

🍾 Welcome to the omnix project

To run omnix,

────────────────────
just watch <args>
────────────────────

(Now, as you edit the Rust sources, the above will reload!)

🍎🍎 Run 'just' to see more commands. See https://nixos.asia/en/vscode for IDE setup.

❯ ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .
❄️  nix --accept-flake-config --version️
❄️  nix --extra-experimental-features nix-command --accept-flake-config show-config --json️
❄️  nix --accept-flake-config eval '.#om' --json️
⌨️  Preparing to develop project: .
✅ Nix environment is healthy.
❄️  nix --accept-flake-config --version️
❄️  nix --extra-experimental-features nix-command --accept-flake-config show-config --json️
❄️  nix --accept-flake-config eval '.#om' --json️
⌨️  Preparing to develop project: .

🍾 Welcome to the project

(Want to show custom instructions here? Add them to the om.develop.default.readme field in your flake.nix file)

[shivaraj-bh]

The above numbers are from omnix repo, I will run them on https://github.com/nammayatri/nammayatri, the result of which I will add to the description of the PR.

[srid]

Great

[srid]

The above numbers are from omnix repo, I will run them on https://github.com/nammayatri/nammayatri, the result of which I will add to the description of the PR.

Sure, I'll merge this in the meanwhile.

[shivaraj-bh]

Using YAML config is 14 times faster than flake config in Nammayatri on x86_64-linux (Updated the PR description).

[shivaraj-bh]

On MBP M1 pro it is ~10x faster:

nammayatri on  main via 🐍 v2.7.9 took 4s
❯ nix run github:nixos/nixpkgs/nixpkgs-unstable#hyperfine -- --warmup 10 './with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .' './without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .'
Benchmark 1: ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell .
  Time (mean ± σ):     185.6 ms ±   3.7 ms    [User: 6.2 ms, System: 2.7 ms]
  Range (min … max):   178.6 ms … 193.8 ms    15 runs

Benchmark 2: ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .
  Time (mean ± σ):      1.781 s ±  0.058 s    [User: 0.399 s, System: 0.326 s]
  Range (min … max):    1.716 s …  1.922 s    10 runs

Summary
  ./with-yaml/bin/om develop --stage pre-shell . && ./with-yaml/bin/om develop --stage post-shell . ran
    9.59 ± 0.37 times faster than ./without-yaml/bin/om develop --stage pre-shell . && ./without-yaml/bin/om develop --stage post-shell .

[srid]

Clearly an improvement. Let's hope nix config show doesn't become a bottleneck, since that will be run during every direnv refresh/load.