| 18 Oct 2025 |
 Alyssa Ross | or it does work with musl, but doesn't work with cross compilation to musl, which again isn't something anybody has generally bothered to test | 12:22:19 |
 matthewcroughan | Okay but in this case, do you think they're shipping mesa with valgrind? | 12:22:36 |
| Alyssa Ross | easy to find out | 12:22:53 |
| matthewcroughan | Our mesa is super heavy | 12:23:05 |
| Alyssa Ross | but I suspect probably not because Alpine tries to be small | 12:23:07 |
| Alyssa Ross | and that means less optional dependencies enabled by default | 12:23:18 |
| matthewcroughan | Yes so I think there's a happy balance between conditionals and patches, and that not everything has to be an upstream patch | 12:23:47 |
| matthewcroughan | like in this case valgrind is pretty heavy and maybe not all platforms should include valgrind | 12:24:04 |
| Alyssa Ross | if there's a principled criteria for that, sure | 12:24:18 |
| Alyssa Ross | but there's no fundamental reason we should include valgrind for GCC builds and not LLVM ones | 12:24:44 |
| matthewcroughan | Sure, my discovery has highlighted two things:
- An issue with valgrind that needs to be fixed upstream
- Maybe valgrind is really heavy and we don't need it in mesa for all platforms
| 12:25:44 |
| matthewcroughan | The reason Alpine has things working with mesa across the board is because they're compiling with less deps in general, so we're probably not going to have working musl for years unless we did the same | 12:26:24 |
| Alyssa Ross | we have a "working musl" | 12:26:36 |
| matthewcroughan | Not always, I know my issue here is LLVM related only, but I've had similar circumstances | 12:26:57 |
| Alyssa Ross | yes, packages are broken sometimes | 12:27:21 |
| matthewcroughan | Like let's say we want to compile Blender with Musl, alpine seems to have that working, but we probably won't due to our enablement of every dep and feature | 12:27:24 |
| matthewcroughan | And it may be that upstream patching is needed, but it's also the case that we could just compile a more minimal blender. | 12:28:13 |
| Alyssa Ross | which you can do with overrides | 12:28:31 |
| matthewcroughan | Ah okay that's the solution then | 12:28:39 |
| matthewcroughan | if something is for some reason not optional, enough to make it able to be compiled with musl, then that is the bug | 12:28:54 |
 dramforever | we don't have to do everything in upstream nixpkgs | 12:28:57 |
| Alyssa Ross | as a user, you probably don't actually want silent differences in feature support on different platforms for non-fundamental reasons every time somebody didn't feel like fixing something properly | 12:29:08 |
| dramforever | like this is the thing you need to lay off the gas on | 12:29:08 |
| matthewcroughan | A lot of what I have found is not optional and not fixable outside of nixpkgs | 12:29:33 |
| matthewcroughan | So in that case it does need to be done | 12:29:42 |
| dramforever | you can just maintain a fork of nixpkgs that works for musl | 12:29:51 |
| matthewcroughan | https://github.com/NixOS/nixpkgs/pull/452738 | 12:29:58 |
| dramforever | and then push stuff either into nixpkgs or upstream | 12:30:01 |
| dramforever | as maint goes on | 12:30:07 |
| dramforever | example: https://github.com/loongson-community/nixpkgs | 12:30:23 |
