14 Oct 2024 |
fgaz | hellwolf: I recently wrote https://sr.ht/~fgaz/get-tested4nix/ to automate this kind of setup, you might find it useful | 19:35:48 |
emily | In reply to @hellwolf:matrix.org
Also, this is my new observation:
(haskell.compiler.${ghcVer}.override { enableDocs = false; }) mises the cache -> I guess the nix expression is not optimal in that docs disabled shouldn't cause a rebuild ghcVer from the previous ghcVer.
Downloading more from cache is still better than rebuilding.. so I fallback
what you want would require ca-derivations at the least | 19:40:20 |
emily | or, hm, maybe i misunderstand | 19:40:30 |
hellwolf | In reply to @fgaz:matrix.org hellwolf: I recently wrote https://sr.ht/~fgaz/get-tested4nix/ to automate this kind of setup, you might find it useful Very cool, I will check it out. | 19:47:00 |
hellwolf | In reply to @emilazy:matrix.org what you want would require ca-derivations at the least what I meant is that I expected the docs and the binary are independent derivations, and the parent derivation simply combines them. Hence, when I don't want docs, it shouldn't trigger a rebuild of the ghc binary. | 19:48:29 |
emily | right | 19:49:12 |
emily | I don't know how easy GHC's build system makes that | 19:49:18 |
maralorn | In reply to @alexfmpe:matrix.org guess 12.2 is the first plausible "uh oh" ? Huh, I guess we need to lobby ghchq to bump to 10 before 9.20. 😄 | 20:07:25 |
alexfmpe | I just want non associative versioning | 20:08:46 |
alexfmpe | How hard can it be to add _ or whatever delimiter | 20:09:09 |
maralorn | Well, no that hard. But probably harder than convincing upstream that it would be pain. 😄 | 20:09:43 |
maralorn | No, we should probably just do it. I actually even had a reasonable replacement strategy. It just would need to be implemented. | 20:10:40 |
maralorn | * Well, not that hard. But probably harder than convincing upstream that it would be pain. 😄 | 20:10:48 |
hellwolf | what would make GHC 10 instaed of GHC 9? | 20:12:12 |
hellwolf | maybe that would be the right timing for cleaning up this technical debt? | 20:12:27 |
emily | In reply to @alexfmpe:matrix.org How hard can it be to add _ or whatever delimiter FWIW, pkg-name_ver_sion is the Nixpkgs standard | 20:13:39 |
emily | but of course many things do not follow it | 20:13:44 |
emily | it's documented in the manual | 20:13:46 |
emily | so ghc_8_10 is "the right thing" | 20:13:58 |
hellwolf | ghc10 is probably when ghc is basically agda | 20:14:36 |
emily | In reply to @maralorn:maralorn.de Huh, I guess we need to lobby ghchq to bump to 10 before 9.20. 😄 (hey, another reason to ditch those old GHCs before they collide, no? :p ) | 20:14:41 |
maralorn | In reply to @emilazy:matrix.org (hey, another reason to ditch those old GHCs before they collide, no? :p ) I think even having the same attribute used for a completely different version in living memory would be a sin. | 20:15:47 |
maralorn | Just in: New deprecation policy remove old versions when the version number collides with a new release. | 20:16:50 |
maralorn | "Version number, you had on job!" 🤦 | 20:18:11 |
maralorn | * | 20:18:23 |
maralorn | I mean I guess incorrect uses of Monoid instances is kinda a theme in the Haskell world... | 20:19:28 |
Tristan Ross | Just don't package multiple patch releases heh, it's what we do for Flutter so we can have 3.24 be `flutter324`. | 20:22:52 |
alexfmpe | 3.24 or 32.4 ? | 20:49:14 |
alexfmpe | Even if humans can tell by context, that will wreak havok on automation | 20:50:12 |
maralorn | Well, we are already kinda paying that price. otoh its not super terrible, because normally in code you can just access ghc.version which is properly formatted | 20:57:54 |