| 6 Nov 2023 |
 K900 | Can't reproduce here | 20:58:41 |
| K900 | This is just a random poetry2nix package that definitely does not touch gradio | 20:58:50 |
| K900 | As you can see, all the inputs are there | 20:59:06 |
 matthewcroughan - nix.how | in poetry2nix defaultPoetryOverrides.extend (self: super: {}) does super use Nixpkgs ? | 20:59:55 |
| matthewcroughan - nix.how | or is that before the overlay is applied? | 21:00:10 |
| matthewcroughan - nix.how | because if so, I've been using super, assuming it means before and after in the context of extended | 21:00:50 |
| matthewcroughan - nix.how | * because if so, I've been using super, assuming it means before and after in the context of just the extended ovelray | 21:01:08 |
| matthewcroughan - nix.how | * because if so, I've been using super, assuming it means before and after in the context of just the extended overlay | 21:01:11 |
| matthewcroughan - nix.how | * because if not, I've been using super, assuming it means before and after in the context of just the extended overlay | 21:01:23 |
| K900 | super uses the entire overlay stack | 21:01:25 |
| K900 | Which is like ten layers of overlays | 21:01:38 |
| K900 | Your overlay is always applied last | 21:01:48 |
| matthewcroughan - nix.how | https://github.com/MatthewCroughan/privateGPT/blob/main/flake.nix#L30-L41 | 21:02:23 |
| matthewcroughan - nix.how | maybe you can see it from the code then, I'm not sure how to debug this | 21:02:30 |
| matthewcroughan - nix.how | my gradio-client and gradio overrides shouldn't need to exist, since the inputs.nixpkgs already has those nativeBuildInputs | 21:02:59 |
| K900 | And what happens if you remove those? | 21:03:24 |
| matthewcroughan - nix.how | it'll crash saying that it doesn't have those nativeBuildInputs python3.11-gradio-client> hatchling.plugin.exceptions.UnknownPluginError: Unknown metadata hook: requirements_txt | 21:04:01 |
| matthewcroughan - nix.how | Also look at those vendored cargo.locks, there must be a way poetry2nix could use them automatically right? | 21:05:49 |
| K900 | No | 21:06:06 |
| K900 | That would be IFD | 21:06:10 |
| matthewcroughan - nix.how | not if you used builtins.fetchTarball for the fetching of the src right? | 21:06:27 |
| K900 | That would just be banned in nixpkgs | 21:06:37 |
| K900 | Just like IFD | 21:06:40 |
| matthewcroughan - nix.how | then src is an outPath with the cargo.lock in it | 21:06:40 |
| matthewcroughan - nix.how | * then src is an outPath with the Cargo.lock in it | 21:06:43 |
| matthewcroughan - nix.how | Oh, I didn't realise nixpkgs banned that | 21:06:48 |
| K900 | And if you're doing that, you might as well do IFD | 21:06:50 |
| matthewcroughan - nix.how | I don't understand why, or why that's bad, compared to IFD | 21:07:02 |
| K900 | The whole point of not doing IFD or anything IFD-adjacent is to make it so you can evaluate nixpkgs offline | 21:07:04 |
| matthewcroughan - nix.how | Oh, I can actually do that? | 21:07:21 |
