Feat/node insertion - Githubissues

DieracDelta / flake_generator

Generate Nix Flakes imperatively.

MIT License

20 stars 2 forks source link

Feat/node insertion #17

Open DieracDelta opened 3 years ago

DieracDelta commented 3 years ago

This is a draft PR for AST node insertion. Objectives:

[x] create AST nodes for attribute sets
[x] convert between Input type and attribute set
[ ] clean code
- [ ] so many places need &str instead of String
- [ ] better file naming to match rest of project
- [ ] clippy
[x] tie into main control loop so we can create inputs
[x] more tests

Resolves #18 and #14 .

DieracDelta commented 3 years ago

Input generation works, and I've added in a nixpkgs_fmt pass to make sure things look nice. I've added a couple of tests as well. Code quality is poor; lots of unused imports, dead code, etc. A bunch of the processing done in main.rs does not belong. Inputs_utils needs to be split out into something more modular that matches with the rest of the style. Not ready to merge, but getting there.

I also think that a bunch of the functions in utils.rs and input_utils.rs morally speaking belong in rnix. I'm just filling in the gaps done by TODOs. I was hoping for more discussion on the issue I opened there...

DieracDelta commented 3 years ago

Also, inputs need to be (1) checked, and (2) there should be some sort of type generation. Ideally, we should support other types of flake inputs that covers the entire spec. Path/git/, enabling submodules, revisions, etc. Essentially, everything specified here. I think I'd prefer to do these in separate PRs though. The scaffolding here can define the basis for further work.

DieracDelta commented 3 years ago

@zseri sorry I haven't pulled in your changes yet. I think that in order for this to work in a scalable manner (we'd like to pull in some sort of evaluation mechanism so we can meaningfully make edits and check their correctness), the IR needs to be a bit more flushed out. Going to focus on that for a bit then cleanup in the input code. In essence we're making our own IR here. I think it should at a higher level than the output of the parser, and we should have some sort of mechanism to switch between IR and GreenNodes. We'll parse the structure into this higher level IR, make changes based on user request, then push down to GreenNodes again.