| 17 Sep 2025 |
 Katalin 🔪 |
GNUstep Objective-C Runtime 2.3
The blocks runtime from swift-corelibs-libdispatch can now be used instead of
the runtime included in libobjc2.
Interesting. no idea if it’s actually worth it but might look into enabling this once the Swift stuff is updated. at least as an overridable option
| 00:44:49 |
 emily | for Linux? | 00:46:00 |
| Katalin 🔪 | yeah | 00:46:12 |
| Katalin 🔪 | https://github.com/gnustep/libobjc2/pull/293 | 00:47:18 |
 Tristan Ross | haskellPackages.networking doesn't build on one of my work darwin machines lol | 00:52:20 |
 Randy Eckenrode | Intel? | 00:52:40 |
| Tristan Ross | network> Network/Socket/Types.hsc:1162:33:
network> 1) Network.Socket.bind successfully binds to a unix socket
network> uncaught exception: ErrorCall
network> pokeSockAddr: path is too long in SockAddrUnix "/nix/var/nix/bs/nix-build-network-3.2.7.0.drv-1442-3341045394/haskell-network-995a3dc6c606cea3/socket-file", length 106, unixPathMax 104
network> CallStack (from HasCallStack):
network> error, called at Network/Socket/Types.hsc:1162:33 in network-3.2.7.0-7CEyG0JEJ4490t7C71AZAi:Network.Socket.Types
network>
network> To rerun use: --match "/Network.Socket/bind/successfully binds to a unix socket/" --seed 828889125
| 00:52:43 |
| Tristan Ross | No | 00:52:48 |
| Katalin 🔪 | wow, this is actually an issue? yeah there was some issue related to unix socket paths with the new build directory | 00:53:24 |
| Katalin 🔪 | that big of an issue, I mean | 00:53:31 |
| Randy Eckenrode | Oh. Fallout from moving the build folder to the store. The path is longer than the domain socket path length limit. | 00:53:32 |
| Tristan Ross | Yeah | 00:53:39 |
| Tristan Ross | Which we're calling this build from GitHub Actions and I can't make it be /build because it's RO | 00:54:07 |
| Tristan Ross | It also doesn't like /tmp because other perm stuff | 00:54:43 |
| Tristan Ross | So likely gonna have to remove that test or rename that socket | 00:55:10 |
| emily | please don't, both Lix and Nix have fixes in the pipeline for this | 01:03:09 |
| emily | it causes regressions in many other things already | 01:03:19 |
| emily | you can set a shorter build-dir for now | 01:03:29 |
| emily | hacky Nixpkgs patches are not the solution | 01:03:44 |
| Tristan Ross | But that doesn't work with the constraints I have | 01:22:40 |
| Tristan Ross | I'm constrained to /nix/var/nix pretty much and /nix/var/nix/bs is as short as I can get. I could drop 1 more character but I need to drop 2 so it goes to 104 in that test. | 01:23:39 |
| emily | GHA gives you root, no? | 01:24:56 |
| Tristan Ross | It does but sudo mkdir /build says read-only | 01:25:18 |
| emily | /tmp is what build-dir was previously, so if it worked in Nix < 2.30 it should work still | 01:25:25 |
| emily | yes, you need to use /etc/synthetic.conf to make a firmlink | 01:25:36 |
| Tristan Ross | It doesn't like /tmp/build because /tmp is world-writable | 01:25:41 |
| Tristan Ross | mmm, I'll poke it tomorrow. | 01:26:06 |
| Randy Eckenrode | What do the plan to do? | 01:27:31 |
| emily | /nix/var/nix/b/<128 bits of base32> rather than including the derivation name which is what reliably screws things up | 01:28:26 |
| emily | the paths are shorter on average than when it was in /tmp, especially when stuff resolves that symlink to /private/tmp | 01:28:36 |
