| 13 Aug 2025 |
 Tristan Ross | Awesome, that GCC stuff will make it easier to move to GCC NG. | 19:25:41 |
 emily | LLVM 18 can probably go away for 26.05 depending on Zig ecosystem and stuff. LLVM 20 is probably imminently droppable if we ship 21 | 19:25:46 |
| emily | 19 might be droppable depending on how much stuff we need to pin after 21 | 19:25:53 |
| Tristan Ross | Zig should be getting 0.15 in a year | 19:26:08 |
| Tristan Ross | At least that's how I see it lol | 19:26:27 |
| emily | Ghostty uses a 0.13 which needs LLVM 18, e.g. | 19:26:34 |
| emily | * Ghostty uses 0.13 which needs LLVM 18, e.g. | 19:26:37 |
| Tristan Ross | Oh, surprised that uses 0.13 | 19:27:02 |
| emily | oh actually it's not because of damned CUDA | 19:27:05 |
| emily | but conceivably we could ship 26.05 with like 2–3 LLVM versions if we're lucky | 19:27:19 |
| emily | that would be nice | 19:27:22 |
| Tristan Ross | I agree and now that I'm back from DEF CON as of Monday night, I can start catching up on things. | 19:27:51 |
| emily | I am still testing LLVM 21 btw | 19:28:16 |
| emily | on Darwin | 19:28:20 |
| emily | it ran into some blockers | 19:28:23 |
| emily | I will get back to it but I diverted for a bit to remove all these compilers | 19:28:30 |
| Tristan Ross | Oh rip, I think there's a merge conflict on it or adjacent PR's. I can clean those up soon. | 19:28:55 |
 @rosssmyth:matrix.org | Ok I tried it, and the fact that it is a lie is a lie. My guess is this is a change since the stage 1 stdlib rework. It currently fails while building compiler_builtins. | 19:29:32 |
