| 4 Jul 2026 |
 hexa | it's funnny how little the python 3.14 migration made an impact | 13:23:32 |
| hexa | thanks for making my work look great ig :p | 13:23:47 |
 Grimmauld (any/all) | i mean, not hard if it gets overshadowed by literally everything else | 13:23:57 |
| Grimmauld (any/all) | lol | 13:24:00 |
| Grimmauld (any/all) | in any other cycle i am sure py 3.14 would have broken significant things, and now we just don't care | 13:25:20 |
 emily | the mysterious missing dependencies are just the fixed GC issues, no? | 13:37:10 |
| *  K900 sighs in paperless | 13:37:28 |
| emily | I can do a little looking at Darwin I guess if things are still bad on that front | 13:37:51 |
| emily | but running out of steam for this cycle a bit, I still need to get those update script fixes out | 13:38:04 |
| hexa | pikepdf looked bad | 13:38:06 |
| K900 | https://hydra.nixos.org/build/334188458 is another blocker | 13:38:31 |
| emily | yeah, that's one of the ones I have no idea about | 13:38:31 |
| emily | it looks like the libc++ update | 13:38:39 |
| hexa |
/nix/store/knrkjgyayyqs91jbjadbh15pd2r07s2z-python3.14-nanobind-2.12.0/lib/python3.14/site-packages/nanobind/include/nanobind/nanobind.h:35:10: fatal error: 'cstddef' file not found
| 13:38:46 |
| hexa | so weird | 13:38:51 |
| emily | yeah | 13:39:00 |
| emily | it's using clang-scan-deps | 13:39:05 |
| emily | which is IIRC unwrapped and kind of weird | 13:39:08 |
| emily | but presumably it was working before | 13:39:11 |
| emily | we did update darwin.libcxx with the SDK update, so it might be that, there's some other fallout that looks like that | 13:39:20 |
| emily | but idk why it'd be happening | 13:39:25 |
| emily | shion:~
❭ fd cstddef result/
result/include/c++/v1/__cstddef/
result/include/c++/v1/__cxx03/cstddef
result/include/c++/v1/cstddef
| 13:39:55 |
| Grimmauld (any/all) | @hexadoes python3Packages.nanobind.tests work? It breaks in that include, not pikepdf itself. Which is why i am wondering whether the issue may lie with nanobind. We don't do checks on that by default. | 13:49:38 |
| emily | ok, so pikepdf got bumped | 13:51:04 |
| emily | and started to use clang-scan-deps | 13:51:14 |
| emily | so it's a pikepdf update, not a Darwin change, that broke it most likely | 13:51:27 |
| emily | looks like they switched build systems | 13:51:32 |
| hexa | quickly fetching half the ml world for nanobind tests 🫠 | 13:51:58 |
| emily | oh they didn't even use nanobind before | 13:52:48 |
| hexa | nanobind.tests works | 13:53:06 |
