| 28 Nov 2025 |
 Acid Bong | i did, NixOS still pulled doc outputs
weird thing about Nixpkgs' outputs in general: they're all reachable from one another (meaning ffmpeg.dev is the same as ffmpeg.lib.bin.dev.bin.dev)
my workaround was to use symlinkJoin, but that looked kinda convoluted | 16:20:50 |
 linj | cannot reproduce. closure size of my cachix is only 155MiB
nix path-info -Sh $(nix build github:nixos/nixpkgs/nixos-unstable#cachix --print-out-paths)
/nix/store/k77ik2kcv82y2qvv3lg6x4r2cfwjbix2-cachix-1.9.1-bin 155.8 MiB
| 16:25:44 |
 ghpzin | You might disable documentation.doc.enable, if you do not want to pull them for any package. | 16:25:51 |
| ghpzin | You might disable documentation.doc.enable, if you do not want to pull them for any package in environment.systemPackages | 16:26:06 |
| Acid Bong | the way to reproduce is to have environment.systemPackages = [pkgs.cachix]; | 16:27:44 |
| Acid Bong | it does help, but I want to have doc output from other packages | 16:28:24 |
| Acid Bong | * it does help, but I want to have doc output from other packages, so not perfect | 16:28:31 |
| linj | interesting, let me try | 16:28:58 |
| Acid Bong | it'll start like this:
these 279 paths will be fetched (160.57 MiB download, 1693.50 MiB unpacked):
/nix/store/i6kmc2cg12acc5kx55vpkxwb1a07a17s-Diff-1.0.2-doc
/nix/store/1aqpij0r645krf360sp91755gbr23g6k-OneTuple-0.4.2-doc
/nix/store/nfnj3wiv2hmiybkywz72749b7vy17hpj-QuickCheck-2.15.0.1-doc
/nix/store/19yjvfpkjxxcvi8kyvj21wk8qlf1qfy9-SHA-1.6.4.4-doc
/nix/store/lr0vzbpjqzxli3i255y6rxxmihh75ljx-StateVar-1.2.2-doc
/nix/store/0y1dvczs1hhnhq21fhx6gpv8jdswxcd3-adjunctions-4.4.3-doc
/nix/store/qi4r1j0mf1qrfxkzgdcd402gkm8g1dz5-aeson-2.2.3.0-doc
/nix/store/91hz94r3z28al926pj1z49nwd3a7l7xk-aeson-pretty-0.8.10-doc
/nix/store/1qzz56rc62qbcaz6735nfrz68llmzmn9-algebraic-graphs-0.7-doc
/nix/store/njypyw5wkzqjz8yg5h5jsc7fprhjacfq-amazonka-2.0-unstable-2025-04-16-doc
/nix/store/mhr4sj0rv8amf2df1bqa1p23r2zxwa4v-amazonka-core-2.0-unstable-2025-04-16-doc
...
| 16:31:40 |
| Acid Bong | actually, this seems more like a buildEnv issue than Haskell (it's just only Haskell that has doc outputs for every library out there) | 16:35:47 |
| linj | Acid Bong: Yeah, you are right. | 17:05:07 |
| linj | Maybe we should add a variant of cachix that uses justStaticExecutables instead of enableSeparateBinOutput | 17:05:47 |
| Acid Bong | i seem to be not alone in this: https://github.com/NixOS/nixpkgs/issues/316768 | 17:07:32 |
| Acid Bong | maybe do that for all top-level standalone haskell packages? | 17:07:58 |
| Acid Bong | although Nixfmt is fine already | 17:09:12 |
| linj | No, I do not think they are the same issue. Actually, from a certain perspective,
your "issue" is not an issue. It works as expected: you have documentation.doc.enable = true so cachix doc is pulled in | 17:10:44 |
| linj | because it uses justStaticExecutables already | 17:11:39 |
| Acid Bong | not cachix's doc, but all its dependencies' docs | 17:12:27 |
| Acid Bong | * not only cachix's doc, but all its dependencies' docs | 17:12:37 |
| Acid Bong | and the second part is exactly undesirable | 17:12:48 |
| linj | haha, its doc depends on its dependencies' doc | 17:15:01 |
| linj | You can go to its dependencies' doc by clicking hyperlinks from its doc | 17:16:39 |
| Acid Bong | ah, I thought buildEnv pulls docs recursively | 17:16:41 |
| linj | It's a feature! | 17:16:44 |
| 29 Nov 2025 |
 maralorn | linj: re: justStaticExecutables. This is interesting. I have been advocating to replace justStaticExecutables with enableSeparateBinOutput because I was not aware of any downsides of the later (when it works. It can cause build errors because of problems with cabal). And I always found it more elegant to reuse the library output instead of just deleting it.
No I am wondering whether that was a bad direction.
| 00:09:44 |
| maralorn | * linj: re: justStaticExecutables. This is interesting. I have been advocating to replace justStaticExecutables with enableSeparateBinOutput because I was not aware of any downsides of the later (when it works. It can cause build errors because of problems with cabal). And I always found it more elegant to reuse the library output instead of just deleting it.
Now I am wondering whether that was a bad direction.
| 00:09:49 |
| maralorn | It is also kinda doubtful that the library docs of cachix are actually usefull docs for users of the executable. | 00:10:33 |
| Acid Bong | In reply to @maralorn:maralorn.de
It is also kinda doubtful that the library docs of cachix are actually usefull docs for users of the executable. yeah, cuz if they want Cachix the library, they can get haskellPackages.cachix (or shellFor {packages = h: [h.cachix];}) | 05:27:15 |
 sterni (he/him) | Wolfgang Walther: I could make a cabal2nix release for the next iteration, so end users on stable cabal2nix can benefit from _type etc. Not a big priority since it's somewhat fringe, but would be good to do eventually | 14:46:16 |
 Wolfgang Walther | No objections! | 18:03:35 |
