| 16 Feb 2025 |
 ma27 | In reply to @irenes:matrix.org
I'm on nix (Lix, like Nix) 2.93.0-dev-pre20250120-5f1782a which I realize may be slightly old IIRC this was indeed a regression on master and the date sounds like you could still be affected. So it probably makes sense to try upgrading. | 09:40:24 |
 Irenes | hmm thanks! | 09:42:11 |
|  aloisw changed their profile picture. | 10:14:30 |
 tea | In reply to@piegames:flausch.social
Though I'd probably stringify the enum tag instead of using a boolean, especially in combination with symbol optimization (strings in the symbol table compress down to u32) i am now afraid to ask, how are booleans stored? | 11:32:12 |
 piegames | In reply to @noob_tea:matrix.org
i am now afraid to ask, how are booleans stored? Maybe I was a bit unclear: I'd prefer strings over booleans for usability reasons, and there are ways to do so without eating performance | 11:33:57 |
| tea | In reply to@piegames:flausch.social
Maybe I was a bit unclear: I'd prefer strings over booleans for usability reasons, and there are ways to do so without eating performance ah okay, I thought you implied that booleans would be larger than an u32 (which would be terrible, but not that surprising, knowing nix) | 11:35:46 |
| piegames | I mean they are, at least in the sense that all Nix values are 24 byes (IIRC), but string are still slower because of more allocations and bookkeeping (string context etc.) | 11:38:05 |
|  gretj joined the room. | 16:11:33 |
 Charles | In reply to @piegames:flausch.social
Though I'd probably stringify the enum tag instead of using a boolean, especially in combination with symbol optimization (strings in the symbol table compress down to u32) I've just realized that you could just get rid of the bool entirely and just go off the length of the list | 16:41:49 |
| Charles | In this particular case anyway. But yeah you'd need a string tag or something in the general case | 16:42:16 |
| Charles | Also this (the general case) is pretty much the same as what I came up with for python a while back: https://charles.pages.gitlab.computer.surgery/blog/python-has-sum-types.html | 16:48:14 |
| piegames | In reply to @charles:computer.surgery
In this particular case anyway. But yeah you'd need a string tag or something in the general case Yeah, this exactly. I'd probably prefer making it a more general sum type system | 16:50:33 |
| Charles | is there a name for the <nixpkgs> syntax? newer versions of the cppnix manual seem to call this "path lookup", but i don't see that term in the lix manual anywhere | 18:32:36 |
 antifuchs | Believe that’s just called a path, at least that’s what the nix manual calls them | 18:41:23 |
| antifuchs | From https://nix.dev/manual/nix/2.18/language/values#type-path | 18:41:36 |
| Charles | your manual is too old | 18:41:45 |
| Charles | https://nix.dev/manual/nix/2.25/language/constructs/lookup-path | 18:42:03 |
| antifuchs | Fine fine https://docs.lix.systems/manual/lix/stable/language/values.html#type-path | 18:42:39 |
| Charles | that doesn't give the construct a name | 18:42:59 |
| Charles | it just says you can use that syntax, but it doesn't give it a name | 18:43:19 |
| Charles | i'm looking for a name | 18:43:23 |
| antifuchs | Ah, for the syntax; got it | 18:43:38 |
| Charles | yeah | 18:43:44 |
| Charles | semantically paths and lookup paths behave the same way so for a type system they should probably both just belong to a single path type, but i think it would be useful to have a term for both actual paths and the angle bracket syntax | 18:44:20 |
| antifuchs | Hm, yeah, maybe something to put on the value’s meta info (: | 18:49:26 |
| Charles | relatedly, here's some half-baked thoughts on a system for doing type annotations in nix-the-language: https://hedgedoc.computer.surgery/s/59ni6v-oA# | 19:37:35 |
| Charles | if anyone has ideas for how to solve the problem mentioned with callPackage towards the end there i'd be interested in hearing it | 19:46:06 |
| Charles | basically, callPackage will add attributes if the thing you're callPackageing returns an attribute set, otherwise it won't bother | 19:47:01 |
| Charles | the type in the reference doesn't even indicate this behavior: https://nixos.org/manual/nixpkgs/unstable/#function-library-lib.customisation.callPackageWith | 21:27:59 |
 jyrama | I somehow managed to get my nix-env oom-killed while running comma. Not sure why comma even ran nix-env in the background. Anyone has pointers for gathering information on the process so the underlying problem could be investigated after I start reproducing it? | 21:38:28 |
