| 22 Sep 2025 |
 emily | cross, perchance? | 19:54:59 |
| emily | or strictDeps? | 19:55:02 |
| emily | anyway it's probably https://github.com/mesonbuild/meson/commit/8b3c6c9fb4b27c1a5635d2b498275ed2eb326a8f | 19:55:12 |
 Grimmauld (any/all) | no, native build | 19:55:10 |
| emily | ran into this today | 19:55:17 |
| emily | Alyssa Ross: (I'll ask you to backport ^ on the bump, so might want to do it now) | 19:55:48 |
| Grimmauld (any/all) | https://github.com/NixOS/nixpkgs/blob/8eaee110344796db060382e15d3af0a9fc396e0e/pkgs/development/tools/language-servers/nixd/default.nix#L43
strict deps yes | 19:55:55 |
| emily | yes, so it's "LLVM 21" but actually it's Meson. very annoying | 19:56:15 |
 Alyssa Ross | In reply to @emilazy:matrix.org
Alyssa Ross: (I'll ask you to backport ^ on the bump, so might want to do it now) Too late, bedtime | 19:56:21 |
| emily | adding llvm to nativeBuildInputs will make it fall back to llvm-config | 19:56:32 |
| emily | or turning off strictDeps of course | 19:56:38 |
| emily | since technically you want that llvm-config | 19:56:39 |
| emily | we should fix it in Meson though | 19:56:46 |
| emily | but too late for this cycle | 19:56:48 |
 Lun | Problem: electron/chromium seem to be broken
Pinning an earlier LLVM builds for me but it looks like rust having a newer LLVM risks incompatible bitcode errors so isn't a correct fix
- earlier patch ensuring chromium always uses same LLVM version as rust https://github.com/NixOS/nixpkgs/commit/a71762621bb17c2c19a31fa50926c78007907d76
- patch to pin to 20 that's probably wrong and logs of the failure https://gist.github.com/LunNova/219767a842bdbda4e43d3766db199d4d
| 20:45:14 |
| Lun | * Problem: electron/chromium seem to be broken in staging-next
Pinning an earlier LLVM builds for me but it looks like rust having a newer LLVM risks incompatible bitcode errors so isn't a correct fix
- earlier patch ensuring chromium always uses same LLVM version as rust https://github.com/NixOS/nixpkgs/commit/a71762621bb17c2c19a31fa50926c78007907d76
- patch to pin to 20 that's probably wrong and logs of the failure https://gist.github.com/LunNova/219767a842bdbda4e43d3766db199d4d
| 20:45:21 |
| emily | we can probably backport a patch | 20:47:57 |
| emily | "chromium relies on these invalid enum values to represent histogram data" 🙃 | 20:48:12 |
| Lun | I think patching the enums used with the UMA_HISTOGRAM_ENUMERATION macro to have an underlying type makes it not UB but there are a lot of different enums used with UMA_HISTOGRAM_ENUMERATION. I failed to find a patch in the chromium repo so I may be missing something silly. | 20:48:15 |
| Lun | * I think patching the enums used with the UMA_HISTOGRAM_ENUMERATION macro to have a specific underlying numeric type makes it not UB but there are a lot of different enums used with UMA_HISTOGRAM_ENUMERATION. I failed to find a patch in the chromium repo so I may be missing something silly. | 20:48:48 |
| emily | no upstream report? | 20:50:23 |
| emily | cc emily | 20:50:29 |
| emily | didn't know Chromium was downstream of LLVM, gues we should be testing it along with Firefox | 20:50:44 |
| Lun | https://chromium-review.googlesource.com/c/chromium/src/+/6633292/2/base/metrics/histogram_macros_internal.h yay! | 20:53:30 |
| Lun | I was wrong about how hard it is to fix :phew: | 20:53:54 |
| emily | nice | 20:54:39 |
| emily | my impression is that the browsers try new compiler versions early | 20:54:55 |
 hexa | compiler updates are an obvious performance and security win, no? | 21:02:03 |
| emily | depends, they do get better at exploiting UB 😅 | 21:03:28 |
| Lun | i misdiagnosed it and it was just using an invalid enum value to represent the number of possible values
and let's be fair to them! it's only been UB since a 2014 wg meeting | 21:05:03 |
