20 Oct 2024 |
K900 | Which is a 64-bit package with a separate lib32 output | 20:25:23 |
Atemu | Can't we make the outputs dependent on hostPlatform bitness? | 20:26:54 |
K900 | I have no idea how the Nvidia package works | 20:27:40 |
K900 | That's part of the problem | 20:27:43 |
Atemu | I assume it runs the installer into some prefix | 20:28:00 |
Atemu | Hm, I don't know if we can make it do that in a 32bit stdenv though | 20:28:28 |
K900 | Like ideally we'd just have hardware.graphics.extraPackages = pkgs: [ hardware.nvidia.package.userspaceBits ] | 20:28:29 |
K900 | But I have no idea what the package does | 20:28:55 |
Atemu | And I don't think we can change the system attr on a per-output basis | 20:29:06 |
K900 | There's a kernel module involved that's definitely native bitness only | 20:29:08 |
K900 | We can't | 20:29:12 |
K900 | Well in theory we can | 20:29:20 |
K900 | With cursed shit | 20:29:25 |
K900 | But I don't want to do cursed shit to avoid other cursed shit | 20:29:33 |
K900 | And I definitely don't want to touch Nvidia stuff with no hardware to test it on | 20:30:58 |
Atemu | I think what we want is either a special attr in the 64bit drv that houses the 32bit output or a "cross-compiled" package from 64 to 32bit in pkgsi686Linux | 20:31:03 |
K900 | And no desire to acquire further cursed knowledge | 20:31:13 |
K900 | In reply to@atemu12:matrix.org I think what we want is either a special attr in the 64bit drv that houses the 32bit output or a "cross-compiled" package from 64 to 32bit in pkgsi686Linux The special attr is effectively what's happening since it's an output | 20:31:26 |
K900 | But that means we have to special case it somehow | 20:31:34 |
K900 | In hardware.graphics | 20:31:40 |
K900 | And I don't want to | 20:31:42 |
K900 | I think the actually correct way is to do what Arch does | 20:32:17 |
K900 | And just have separate packages | 20:32:27 |
K900 | For the kernel bits and the userspace bits | 20:32:33 |
Atemu | Isn't that what we have too? | 20:32:55 |
Atemu | We have kernelPackages.nvidia and the nvidia packgae | 20:33:06 |
K900 | It's the same package with multiple outputs | 20:33:07 |
K900 | And I just realized it's not going to work anyway | 20:33:13 |
K900 | Because the kernel module needs to exactly-match the userspace | 20:33:21 |
K900 | So we're back to pulling config.hardware.nvidia.package | 20:33:32 |