| 7 Apr 2026 |
Randy Eckenrode | system_cmds doesn’t really need libdispatch’s private headers. The APIs it uses are actually in the public headers. I’ll have to vendor the header for now to make this a no-rebuild transition (to separate the source releases from the SDK). | 02:07:32 |
EsperLily [she/her] | does using a newer apple-sdk package require actually being on that OS to build? the nixpkgs darwin docs don't actually say (and I'm already on macOS 26 or I'd just test) | 02:56:09 |
Randy Eckenrode | It shouldn’t as long as you don’t increase the default deployment target. Clang is set up by default to error when APIs are used from a newer deployment target without putting them in an availability check. | 03:01:58 |
EsperLily [she/her] | ok so why do we default to an older SDK then? | 03:07:26 |
Randy Eckenrode | The Clang change is a trial run for having only the latest SDK. | 04:08:03 |
Randy Eckenrode | There are some other concerns (building Swift, source incompatibility in newer SDKs, etc), but availability checks were the big one. We currently default to the 14.4 SDK with a 14.0 deployment target. It seems okay so far. | 04:11:04 |
| insipx joined the room. | 14:21:41 |
viraptor | This is not Mac specific, but clang will hit us way more often with this. Somehow I ran into two packages now which use -fmodules which causes writes to $home/.cache
I wonder if that would warrant a fix in some generic builder to set CLANG_MODULE_CACHE_PATH to a different path. We're only going to hit that one more often in the future. | 15:55:55 |
emily | yes, I think we should likely set that in stdenv | 15:58:50 |
Randy Eckenrode | emily, did you want to take a look at https://github.com/NixOS/nixpkgs/pull/507136 before I merge it later today? | 16:04:11 |
Randy Eckenrode | This is part 1 of separating the source releases from the SDK versions, so we can do SDK bumps as soon as they happen. | 16:04:48 |
emily | will try to take a look later, but I'm ok not being blocked on | 17:08:48 |
emily | btw haskell room did say the GHC thing can be dropped | 17:08:54 |
Randy Eckenrode | I’ll probably leave it since the PR does have one approval already. | 17:11:02 |
Randy Eckenrode | I have considered a few possibilities, but I think what I am going to do is move the sourceRelease function to darwin and refactor mkAppleDerivation to use it. | 22:19:02 |
Randy Eckenrode | That avoids polluting the package set with unbuildable packages. | 22:19:39 |
Randy Eckenrode | It can’t be internal because there’s at least one package outside of the darwin package set that needs access to it. | 22:20:27 |
| 8 Apr 2026 |
Randy Eckenrode | I updated https://github.com/NixOS/nixpkgs/pull/506470 after I merged the by-name migration. | 00:22:25 |
| radu242 joined the room. | 01:55:10 |
ankarhem | anyone had issues with fish shell lately? Cannot enter fish after rebuilding. I think it’s the same as this: https://github.com/NixOS/nixpkgs/issues/507531 | 10:58:17 |
Randy Eckenrode | Maybe related? https://github.com/NixOS/nixpkgs/issues/208951#issuecomment-4195559375 | 10:59:34 |
Randy Eckenrode | https://github.com/NixOS/nix/issues/6065#issuecomment-4203079922 | 11:00:04 |
emily | this seems confused. since nix-build --check clobbers signatures by itself. | 11:06:59 |
ankarhem | Idk I know nothing about this but maybe?
ankarhem@mbp:~/ > codesign -vvv --verify $(which fish)
/etc/profiles/per-user/ankarhem/bin/fish: invalid signature (code or signature have been modified)
In architecture: arm64
| 11:07:40 |
emily | and then the fix is a dreadful layering violation that probably has nothing to do with the fish thing. | 11:07:40 |
emily | if it "fixes" the transient Hydra code signing issues then it's just going to fix them because rebuilds usually fix them | 11:08:11 |
emily | left a comment on the PR for Claude to read 🙃 | 11:13:06 |
ankarhem | So should i just disable fish -> clean store -> rebuild until it works? | 11:19:03 |
emily | nah, you'll pick up the broken build from Hydra :/ | 11:28:25 |
emily | what will probably work is making a trivial change to fish in Nixpkgs | 11:28:37 |