| 4 Aug 2025 |
 Randy Eckenrode | In reply to @emilazy:matrix.org
nor do build inputs go after postPatch (there is no strict convention for all of this, but if you are going to do this kind of churn, explicitly making things comply less with the guidelines in https://github.com/jtojnar/nixpkgs-hammering/blob/b0817c8269798a830bc42f58a01c7543e7c5268a/explanations/attribute-ordering.md is hard to justify) Is this documented anywhere as the convention to use? I’ve been grouping by phases (so build phase stuff together, etc) …. | 23:24:34 |
 emily | I don't think there's any real consensus, but I think the nixpkgs-hammering ordering is something close to what's most common (i.e. probably many packages violate it in some way or another but if you were going to write down the most consistent set of rules you could based on patterns of relative ordering it would be close to that) | 23:25:41 |
| emily | I think some things are pretty strong consensus (passthru and meta last, pname, version, src, and usually outputs and patches before other stuff) | 23:26:31 |
| emily | and everything in between is a little fuzzy | 23:26:35 |
| 5 Aug 2025 |
|  Axel Karjalainen joined the room. | 17:39:37 |
| Axel Karjalainen | Was there any reason for libcCrossChooser to use targetPackages? See also my comment here: https://github.com/NixOS/nixpkgs/pull/409851#discussion_r2254887700
Is this the right Matrix room for this?
| 17:51:17 |
 vcunat | #cross-compiling:nixos.org also exists, but I don't know what's the best place. | 17:54:58 |
| 6 Aug 2025 |
|  vidariondr set a profile picture. | 18:31:02 |
| 8 Aug 2025 |
|  problems changed their profile picture. | 01:54:56 |
|  0x4A6F joined the room. | 06:32:02 |
|  André Lima joined the room. | 18:41:24 |
| 12 Aug 2025 |
 connor (burnt/out) (UTC-8) | If you were trying to add a stdenv for a new compiler which doesn’t support the same arguments as GCC and Clang, is there an easy way to filter out unsupported arguments?
As an example, suppose -frandom-seed wasn’t supported (which is used by https://github.com/NixOS/nixpkgs/blob/0300b04b2e48edec1f9934cda0fb6d4a2a2c1fbd/pkgs/stdenv/generic/default.nix#L83) — what can be done? | 20:51:30 |
| emily | I think we just conditionalize those as appropriate, but for something out-of-tree or where it would be invasive, you could drop them in a wrapper I suppose | 20:52:27 |
| emily | (of course ideally it would support the flags) | 20:52:36 |
| connor (burnt/out) (UTC-8) | Any references for prior art for how to do that, or recommendations for where such a wrapper would go? | 20:54:34 |
| emily | are you using cc-wrapper already? | 20:58:25 |
 Grimmauld (any/all) | there is already flags being filtered, e.g. -march flag will be dropped | 21:01:45 |
| 17 Aug 2025 |
|  Benno Fünfstück joined the room. | 17:51:29 |
| 19 Aug 2025 |
| emily | does anyone know if configurePlatforms = [ "host" ]; by default would break as much stuff as [ "build" "host" ] | 21:33:56 |
 Alex | In reply to @emilazy:matrix.org
does anyone know if configurePlatforms = [ "host" ]; by default would break as much stuff as [ "build" "host" ] Wouldn't this break cross for most packages?
Looking at the autoconf documentation, it seems like it won't try to guess the build platform if you set --host. | 22:39:43 |
| emily | we pass [ "build" "host" ] when doing cross | 22:40:01 |
| emily | I'm just talking about the default for the native case | 22:40:05 |
| emily | which should be [ "build" "host" ], but isn't because some stuff breaks | 22:40:15 |
| emily | though maybe it's about time to just throw that in a cycle and deal with the fallout | 22:41:53 |
| emily | I forget what Artturin said last time I was talking about a good ordering for these things | 22:42:06 |
| emily | not sure if https://github.com/NixOS/nixpkgs/pull/181724 ended up resolved | 22:43:07 |
| emily | ok, yeah, https://github.com/NixOS/nixpkgs/issues/178802 | 22:52:39 |
| emily | I think adding wrapped prefixed compilers was the first sensible step | 22:52:45 |
| emily | not sure if --host without --build would trigger such problems though | 22:52:53 |
| emily | oh, it looks like we no longer re-export the prefixed ones from the wrapped packages | 22:59:19 |
