| 31 Oct 2025 |
 mio | I noticed that gtk packages like trayscale cannot find icons | 01:32:03 |
| mio | 
Download CleanShot 2025-10-31 at 12.32.19@2x.png | 01:32:32 |
| mio | gnome-test-editor | 01:32:33 |
|  saige leah changed their profile picture. | 07:11:33 |
|  twistedttea changed their display name from bartholomew to twistedttea. | 10:08:26 |
 niklaskorz | so that wasn't it | 13:05:42 |
| niklaskorz | but I have a theory | 13:05:46 |
| niklaskorz | apparently cross compiling Rust crates that try to compile vendored C code are using the wrong compiler | 13:06:17 |
| niklaskorz | no that's not it, minimal reproducer using the cc crate correctly uses x86_64-unknown-linux-gnu-gcc | 13:25:03 |
| niklaskorz | but it's sufficient to add the liblzma-sys crate to a minimal application to reproduce | 13:29:44 |
| niklaskorz | * but it's sufficient to add the liblzma-sys crate to a minimal application to reproduce the linker failure | 13:30:15 |
 Randy Eckenrode | Is that crate doing something wrong in its build.rs that breaks with cross-compilation? | 13:30:24 |
| Randy Eckenrode | * Is that crate doing something in its build.rs that breaks with cross-compilation? | 13:30:33 |
| niklaskorz | I'm trying to find out, I suspect so | 13:30:46 |
| niklaskorz | but it's not only that crate | 13:30:50 |
| niklaskorz | also the ring crate that is a very common dependency | 13:30:58 |
| niklaskorz | notably the failing linker invocation has these args that I would suspect should not be there when producing a linux elf binary: | 13:31:52 |
| niklaskorz | "-liconv" "-lSystem" "-lc" "-lm" "-arch" "arm64" "-mmacosx-version-min=11.0.0"
| 13:31:54 |
| niklaskorz | oh wait, what's failing is compiling the build script itself :o | 13:37:07 |
| niklaskorz | ok so then it makes sense that is being compiled for the build platform | 13:37:26 |
| niklaskorz | hm but in that case emily's original suggestion is likely the solution and I just added it in the wrong place | 13:38:30 |
| niklaskorz | at least I'm now certain it fails for cc crate with feature parallel enabled | 13:52:59 |
| Randy Eckenrode | Shouldn’t the build platform CC have the right SDK and stuff? It should be using the fallback SDK. | 13:55:02 |
| Randy Eckenrode | Oh, but it’s not finding the propagated inputs probably. If you add just libiconv, does it work? | 13:56:53 |
| Randy Eckenrode | I hate how we do dependencies so much. | 13:57:02 |
| Randy Eckenrode | Like with a bunch of wrapper stuff and Bash to find propagated dependencies. | 13:57:50 |
| niklaskorz | no, already tried that :/ | 13:58:44 |
| Randy Eckenrode | Maybe we should symlink the dylibs or add text-based stubs for them instead of propagating them. I can do that for 26.05. It seems too risky for 25.11. | 13:59:08 |
| Randy Eckenrode | I don’t think adding an SDK for the build platform is the right solution. | 13:59:31 |
| Randy Eckenrode | * | 13:59:46 |
