| 21 May 2023 |
 Lily Foster | Also the navidrome update script fails because building .go-modules in general on the derivation fails due to buildGoModule propagating nativeBuildInputs to the fetcher (???) which obviously adds the npm hook which when run, expects to be building npm stuff, not fetching go modules | 21:38:51 |
| Lily Foster | I would say this is a bug in buildGoModule and not npmConfigHook because I imagine the same would happen with, e.g., the rust hooks | 21:39:12 |
| Lily Foster | * I would say this is a bug in buildGoModule and not npmConfigHook because I imagine the same would happen with, e.g., the cargo/rust hooks | 21:39:16 |
 hexa | hm ok | 21:39:38 |
| Lily Foster | Now for what to do about it, idk | 21:39:46 |
| Lily Foster | There's no comment in the buildGoModule code about why it propagates nativeBuildInputs to the fetcher -- do you know why it would be doing that? | 21:40:20 |
| Lily Foster | That seems... counterintuitive. Stuff like buildRustPackage don't do that | 21:41:11 |
| Lily Foster | * That seems... counterintuitive. Stuff like buildRustPackage doesn't do that | 21:41:24 |
| Lily Foster | (and you can't override the fetcher with buildGoModule either since the fetcher is still not separately exposed...) | 21:41:53 |
| Lily Foster | Apparently buildGoModule also doesn't propagate unpackPhase or its hooks to the fetcher despite propagating src, which seems sus... | 21:48:24 |
| Lily Foster | It looks like the evcc derivation has the exact same issue and fixes it via buildGoModule's overrideModAttrs to set a slimmer nativeBuildInputs | 21:56:58 |
| Lily Foster | Yeah, adding this to the derivation fixes building .go-modules and fixes the update script hexa:
overrideModAttrs = oldAttrs: {
nativeBuildInputs = lib.filter (drv: drv != npmHooks.npmConfigHook) oldAttrs.nativeBuildInputs;
preBuild = null;
};
| 22:03:18 |
| hexa | In reply to @lily:lily.flowers
There's no comment in the buildGoModule code about why it propagates nativeBuildInputs to the fetcher -- do you know why it would be doing that? no idea | 22:03:41 |
| hexa | In reply to @lily:lily.flowers
It looks like the evcc derivation has the exact same issue and fixes it via buildGoModule's overrideModAttrs to set a slimmer nativeBuildInputs oh, right 😲 | 22:04:10 |
| Lily Foster | In reply to @hexa:lossy.network
https://github.com/NixOS/nixpkgs/pull/230991#issuecomment-1552133508 I just realized that was before winter fixed the PR to actually work (which is why it ended up with a pushd/popd) | 22:05:11 |
| Lily Foster | So yeah the npmRoot PR needs to a rebase and then I'll merge and deal with further changes in follow-ups, to unblock other stuff | 22:06:33 |
| hexa | just the upper pushd hunk needs a massage | 22:11:08 |
| Lily Foster | Yep, I've already done it locally. Was gonna get permission from winter before just pushing to the branch myself though | 22:11:34 |
| 22 May 2023 |
 raitobezarius | is there a way with buildNpmPackage to just "install" the node_modules and be done? | 11:45:23 |
| raitobezarius | is dontNpmInstall the option I am looking for? I tried it and I'm not sure I am holding it correctly | 11:45:40 |
| Lily Foster | Is the default install hook over-copying or failing or something? | 11:48:19 |
| raitobezarius | > Executing npmInstallHook
> npm ERR! Invalid package, must have name and version
| 11:58:22 |
| raitobezarius | this is an invalid Node.js package per se | 11:58:26 |
| raitobezarius | because they reuse it in some Makefile later on | 11:58:31 |
| raitobezarius | And they really only need the node deps | 11:58:35 |
| Lily Foster | If the project has custom install logic, just override installPhase and do what you need to do | 12:05:52 |
| Lily Foster | Or if you want the default stdenv installPhase then dontNpmInstall does what you want | 12:07:42 |
| raitobezarius | Okay, thank you! | 12:09:20 |
| Lily Foster | You may need to prune node_modules first yourself though so you don't end up with dev deps in the output | 12:10:07 |
| Lily Foster | If the makefile doesn't | 12:10:17 |
