| 15 Feb 2026 |
|  indy joined the room. | 06:54:27 |
|  amadaluzia[tde] changed their profile picture. | 22:03:28 |
| amadaluzia[tde] changed their profile picture. | 22:04:34 |
| 16 Feb 2026 |
 magic_rb |
romes: The first Stage3 cross-compiler (i.e. a cross-compiled compiler) built by our CI:
https://gitlab.haskell.org/ghc/ghc/-/jobs/2421476/artifacts/browse
I'm testing it now.
They managed to get a riscv ghc crossed from im assuming x86. Could be used for bootstrap on our side 🎉
(Cross post from #GHC:matrix.org)
| 11:29:59 |
 Alex | We already have a working setup for native RISC-V via x86 cross (which I think is in the official Nixpkgs cache now?), but an official RISC-V build could make it less fragile.
But I suspect that the new compiler is only viable for the latest GHC, which would take a while to become the default GHC? | 12:05:05 |
|  pneumatic changed their profile picture. | 15:17:52 |
| 17 Feb 2026 |
|  @jboy:utwente.io joined the room. | 12:45:25 |
| @jboy:utwente.io left the room. | 12:53:18 |
| 18 Feb 2026 |
|  -> @n:rab.lol changed their display name from niko ⚡️ to -> @n:rab.lol. | 23:28:25 |
| 19 Feb 2026 |
 woobilicious | what's the simplest way to bump a package version? sbv package is broken and that version is marked as deprecated upstream. I guess overrideSrc is best here? | 04:26:53 |
 peddie | out of curiosity, isn't sbv usually broken because its golden tests hardcode what they expect to be emitted for the name/path of the SMT solver, but in nixpkgs it emits the full path rather than just the binary name? did he finally fix this upstream? | 04:40:41 |
| woobilicious | Never used it before, so can't comment on it, wouldn't we want to just disable tests instead? | 04:43:53 |
| woobilicious | wait hold up we have a newer verision but its called sbv_12 hmm | 04:50:17 |
| woobilicious | but that is also marked as broken lol | 04:52:16 |
| woobilicious | note the only suppposed working version on hackage is v13.5 | 04:53:04 |
| woobilicious | well, looks like the reason why our version is 11.7 is that that is what stackage lts uses. | 05:22:52 |
| woobilicious | peddie: you might be right there, we don't mark it dontCheck by the looks. checks are also extremely slow. I've disabled now because I'm impatient lol | 05:33:03 |
| peddie | I looked at it a couple times and it was always only that issue, and I don't think Levent has a lot of time to support "unusual" platforms like nixpkgs :) | 05:34:34 |
| peddie | I guess one option would be to patch the golden test files when we build! | 05:36:52 |
| woobilicious | I guess we could fix it for him, or just disable tests. | 05:37:46 |
| peddie | well the tests compare exact string equality with the golden output files. IIRC the ones it ships with just contain the SMT solver name, like z3 or cvc5, but when we run tests in nixpkgs we get the full path /nix/store/blahblahblah/bin/z3, so they fail. but it doesn't make any sense to insert any full path upstream | 05:39:05 |
| woobilicious | I mean the test itself shouldn't care about fullpath v not, so there must be some way to fix this false negative without a "hacky" solution like inserting hard coded paths. | 05:44:45 |
| peddie | yes, this is really a spurious failure. we could just mask out the initial line that contains the solver name from both the output and the reference file, or something like that | 05:45:29 |
| woobilicious | do we have a link to the failing test? I assume it's on hydra somewhere | 05:46:01 |
| woobilicious | lol, why is hydra's UI so confusing, do periods break hydra search? wtf | 05:51:57 |
| woobilicious | Ahh it hasn't been tested since v10.9 in 2023 | 05:59:01 |
| woobilicious | Okay fixed: https://github.com/LeventErkok/sbv/pull/772 | 06:56:25 |
| peddie | the hero we don't deserve! nice one! | 07:09:40 |
| woobilicious | yeah was easy find :D | 07:09:51 |
| woobilicious | I also like feeling productive instead of wasting time on the actual code I was gonna write lol | 07:10:19 |
