| 11 Nov 2025 |
 vcunat | (with work stuff) | 11:35:36 |
| vcunat | So either way I won't have capacity for longer discussion about the changes. | 11:36:53 |
 ElvishJerricco | https://github.com/NixOS/nixpkgs/pull/460639 | 11:39:43 |
 winston | I was looking into the arrow-cpp test failure on darwin, strangely the binary of the tests that fail end up with two versions of protobuf in the dynamic linker paths
v31, which is what we want in the derivation, v33 coming from somewhere else
whats stranger is that master also shows v32, but it doesn't have the same runtime error
I compiled it with -fsanitize=address and that pointed me into this direction: https://github.com/protocolbuffers/protobuf/issues/5290
https://gist.github.com/nekowinston/f6cbea976ba4605f318b9329d96a0544 | 11:52:38 |
| winston | * I was looking into the arrow-cpp test failure on darwin, strangely the binary of the tests that fail end up with two versions of protobuf in the dynamic linker paths
v31, which is what we want in the derivation, v33 coming from somewhere else
whats stranger is that master also shows v31+v32, but it doesn't have the same runtime error
I compiled it with -fsanitize=address and that pointed me into this direction: https://github.com/protocolbuffers/protobuf/issues/5290
https://gist.github.com/nekowinston/f6cbea976ba4605f318b9329d96a0544 | 11:55:11 |
 Wolfgang Walther | PostgreSQL releases are due this week, fixing two CVEs. I have currently split them up as:
- All server updates in https://github.com/NixOS/nixpkgs/pull/460644 targeting master.
libpq update in https://github.com/NixOS/nixpkgs/pull/460645 targeting staging-next. One CVE affects libpq, so not targeting staging here. Rebuild count is currently 4549 per Linux, 2167 per Darwin.
Rebuilds for the server part have increased by a few hundred over the last couple of months again, we're now at 1085 per Linux and 819 per Darwin. I can split the default attribute (postgresql) into a PR towards staging-next, too - or depending on the timing just retarget that whole PR, if any of these are preferred?
Any opinions?
| 12:12:18 |
 hexa | right now I think what goes where depends on the state of staging-next | 12:29:26 |
| vcunat | I'd merge staging-next if cachix is fixed, as it's a channel blocker. | 12:32:10 |
| Wolfgang Walther | fyi: I had already merged https://github.com/NixOS/nixpkgs/pull/460645 into staging-next, assuming the only question would be around the PR currently targeting master. | 12:37:51 |
| hexa | not a problem | 12:39:14 |
| Wolfgang Walther | Is anybody working on fixing that / aware of the problem already? I posted it in the haskell channel as a starter. | 12:48:15 |
 ivy | it looks like very odd error | 12:49:18 |
| ivy | like it’s failing to link its own compiled haskell modules | 12:49:42 |
| ivy | because the symbols its looking for are internal it seems | 12:50:25 |
| ivy | unless someone forgot to expose a c output or c import right | 12:51:00 |
| Wolfgang Walther | That doesn't look like loading haskell modules to me: dlopen(/nix/store/anfq16cq9qki033324cl26w80ngn0ska-nix-2.28.5/lib/libnixexpr.dylib, 0x0005): Symbol not found: | 12:51:59 |
 Sergei Zimmerman (xokdvium) | In reply to @wolfgangwalther:matrix.org
That doesn't look like loading haskell modules to me: dlopen(/nix/store/anfq16cq9qki033324cl26w80ngn0ska-nix-2.28.5/lib/libnixexpr.dylib, 0x0005): Symbol not found: Oh, it’s Darwin libc++ bananza? | 12:53:17 |
| winston | it builds successfully when the grpc dependency is overridden to use the same version of protobuf, i.e. grpc.override { protobuf = protobuf_31; } 🤔 | 12:54:27 |
| winston | protobuf_33 came from there | 12:54:42 |
| vcunat | O | 13:13:38 |
| vcunat | * I'm not aware. | 13:13:44 |
| vcunat | I just mentioned Domen, as the maintainer. (who has business built around cachix, I think) | 13:14:25 |
 Grimmauld (any/all) | why do we make peoples for-profit business stuff channel blockers anyways? If they want it, they need to get their crap together themselves. Unless they pay us, i see no good reason to carry their stuff and make it slow down the rest of the ecosystem | 13:35:24 |
| vcunat | As long as they fix it fast, etc. I don't mind. (and this case seems reasonable to me) | 13:40:19 |
| vcunat | I rather have issues with blockers that have no .meta.maintainers. | 13:40:35 |
| vcunat | Because... what do I do with them? Like, fix them myself always? | 13:40:48 |
| Wolfgang Walther | Well, that for-profit stuff also provides free caching for open source projects. And Nixpkgs CI is using it as well - it even does on darwin runners for the build jobs. | 13:40:54 |
| Wolfgang Walther | So with the state outlined above, what's the conclusion to draw?
PR is ready now. Merge to master or re-target? | 14:44:01 |
| hexa | depends on how quick staging-next lands, because then unstable evals will get cancelled | 14:44:42 |
| hexa | I'd say staging-next is probably the safe choice right now | 14:45:07 |
