| 19 Sep 2025 |
matthewcroughan | Redacted or Malformed Event | 17:08:10 |
matthewcroughan | pkgsMusl.zig_0_14 seems to build fine, so it's just LLVM causing it | 17:22:30 |
matthewcroughan | found https://www.reddit.com/r/Zig/comments/18x1wce/is_it_true_that_zig_wants_to_move_away_from_llvm/ 🤔 | 17:22:57 |
| 21 Sep 2025 |
matthewcroughan | Tristan Ross: When I compile a fully LLVM system, the vm attribute produces something that says this | 00:14:44 |
matthewcroughan | unable to handle EFI zboot image with "zstd" compression
qemu-system-aarch64: could not load kernel '/nix/store/fiwxkwxq0nsbmlsbi44cj3jmv0c175nl-nixos-system-nixos-25.11.20250919.8eaee11/kernel'
``
| 00:14:56 |
matthewcroughan | * unable to handle EFI zboot image with "zstd" compression
qemu-system-aarch64: could not load kernel '/nix/store/fiwxkwxq0nsbmlsbi44cj3jmv0c175nl-nixos-system-nixos-25.11.20250919.8eaee11/kernel'
| 00:15:01 |
matthewcroughan | However..
boot.kernelPatches = [
{
name = "config-enable-zboot";
patch = null;
structuredExtraConfig = {
EFI_ZBOOT = lib.mkForce lib.kernel.yes;
KERNEL_ZSTD = lib.mkForce lib.kernel.yes;
RD_ZSTD = lib.mkForce lib.kernel.yes;
};
}
];
| 00:15:19 |
matthewcroughan | so I'm not sure what's happening there | 00:15:24 |
matthewcroughan | https://grep.app/qemu/qemu/master/hw/core/loader.c?q=unable+to+handle+EFI+zboot+image+with#L887 | 00:15:51 |
matthewcroughan | Ah okay I see, I'm producing a linux kernel with vmlinuz.efi as the target, which isn't supported by qemu on arm64 I guess | 00:17:34 |
matthewcroughan | * Ah okay I see, I'm producing a linux kernel with vmlinuz.efi as the target, which isn't supported by qemu on aarch64 I guess | 00:17:38 |
matthewcroughan | whereas for musl I'm just setting crossSystem, here I'm setting nixpkgs.hostPlatform and some linux-kernel args leaked in from my other experiments | 00:18:00 |
Tristan Ross | In reply to @matthewcroughan:defenestrate.it Ah okay I see, I'm producing a linux kernel with vmlinuz.efi as the target, which isn't supported by qemu on arm64 I guess Just do an uncompressed kernel | 01:12:34 |
matthewcroughan | I did, that worked fine | 01:12:43 |
matthewcroughan | I have an llvm, llvm-musl, llvm nixos config and the minimal nix files to change that now | 01:12:57 |
matthewcroughan | it's not too much, but it was a lot of research | 01:13:34 |
matthewcroughan | when I get my m2 ultra I'm gonna put up some CI on github for it | 01:15:00 |
| Jannik Wolff joined the room. | 08:04:08 |
SomeoneSerge (back on matrix) | py_mod.find_installation() with strictDeps = true seems to pick up meson's interpreter, rather than the python-env wrapper from buildInputs | 16:47:51 |
SomeoneSerge (back on matrix) | FWIW I tried to trace the chain from find_installation to find_program in hopes to understand how does one even specify which platform's python they want, and my conclusion was "fuck if I know" | 16:48:54 |
SomeoneSerge (back on matrix) | mesonbuild matrix says it's supposed to look for the host python, so something must be wrong on our side | 21:52:00 |
rosssmyth | What are you working on? If this is running at build-time it will probably sniff around PATH and find the meson interpreter or something. Add python to depsBuildBuild or nativeBuildInputs. buildInputs is for a python interpreter than runs on the host machine the final executable runs on, so at build time it will not find it. | 22:14:16 |
SomeoneSerge (back on matrix) |
this is running at build-time
It's not, it's for the host. The find_installation method is, according to upstream chat, for the host. It's a meson-packaged gtk app written in python o_0. Native/strictDeps=false works and builds fine, but like why would I want strictDeps=false this is cringe
| 23:44:02 |
SomeoneSerge (back on matrix) | *
this is running at build-time
It's not, it's for the host. The find_installation method is, according to upstream chat, for the host. It's a meson-packaged (nb: not meson-python) gtk app written in python o_0. Native/strictDeps=false works and builds fine, but like why would I want strictDeps=false this is cringe
| 23:44:53 |
| 22 Sep 2025 |
SomeoneSerge (back on matrix) | Mhmmm update: it's basically the same disaster as cmake, they look to run python from PATH on BUILD, but want its sysconfig to return cross stuff | 01:11:07 |
SomeoneSerge (back on matrix) | So I guess nvm it's meant to be broken | 01:11:21 |
rosssmyth | Maybe. You might be able to add a path to one of these options https://mesonbuild.com/Builtin-options.html#python-module but who knows. If not then patching your cross-file may be required with substituteInPlace or replaceVars | 01:28:56 |
rosssmyth | * Maybe. You might be able to add a path to one of these options https://mesonbuild.com/Builtin-options.html#python-module but who knows. If not then patching your cross-file may be required with substituteInPlace or replaceVars | 01:29:03 |
| 23 Sep 2025 |
| estrange joined the room. | 00:18:31 |
| kenji changed their display name from a-kenji to kenji. | 10:42:22 |