| 22 Apr 2025 |
alexfmpe | so now it's just about keeping it building | 12:14:32 |
alexfmpe | (reflex-platform still has a bunch of stuff for mobile) | 12:15:25 |
emily | if Stackage LTS QA is actually as minimal as claimed here then it seems like tracking nightly would be an easy way to avoid painful flag days / long-lived branches with less risk than I expected | 12:16:47 |
Teo (he/him) | I think more sharing between nixpkgs and head.hackage would be great. head.hackage used to use the nixpkgs infra to build stuff so it's not inconceivable that it could do that again. At some point I want to write a little cli for head.hackage to help track upstream PRs etc, since it's a huge pain atm, and often PRs are just not made to the upstream. Nixpkgs is much better in this regard | 12:18:06 |
alexfmpe | apparently I'm wrong, stackage site claims they do run test suites | 12:18:21 |
alexfmpe | they just don't patch things, | 12:18:40 |
alexfmpe | * they just don't patch things | 12:18:44 |
alexfmpe |
head.hackage used to use the nixpkgs infra to build stuff oh? why did they stop?
| 12:19:41 |
alexfmpe | *
head.hackage used to use the nixpkgs infra to build stuff
oh? why did they stop?
| 12:19:46 |
Teo (he/him) | See this commit message: https://gitlab.haskell.org/ghc/head.hackage/-/commit/1b616c6c48dc11dfe1e357d4250ad6882cd3224a | 12:21:21 |
alexfmpe | switching to nightly needs one eager flag day but personally I'm up for that pain if we belive it makes the long term better | 12:21:33 |
alexfmpe | * switching to nightly needs one eager flag day but personally I'm up for that pain if we believe it makes the long term better | 12:21:43 |
emily | it just seems like there's a cadence issue (from previous talk in the room) and IMO any situation where you're ending up with months-long branches off Nixpkgs is pain | 12:23:05 |
alexfmpe |
Nix's Haskell infrastructure doesn't handle flags, which can complicate building some packages
it doesn't ?
| 12:23:12 |
emily | "if it hurts, do it more often" etc. | 12:23:24 |
Teo (he/him) | Perhaps the issue here was that you wanted the solver to pick some flags and nixpkgs doesn't have a solver? Not sure exactly what he's alluding to tho | 12:24:27 |
| mightybyte joined the room. | 12:27:35 |
| @ihar.hrachyshka:matrix.org joined the room. | 12:33:31 |
alexfmpe | * But it's not an orthogonal layer we can just be agnostic to. If they have some package deep in the dependency chains that is broke but only the tests expose it, stackage will happily base other package versions around it and then we need to undo a part of that
EDIT: me wrong. Stackage does run (native) test suites, they just don't patch stuff | 12:37:03 |
alexfmpe | * It's insane to me they don't run test suites
EDIT: me wrong. Stackage does run (native) test suites, they just don't patch stuff | 12:37:51 |
Teo (he/him) | Redacted or Malformed Event | 12:44:39 |
alexfmpe | In reply to @alexfmpe:matrix.org apparently I'm wrong, stackage site claims they do run test suites Think I see where I got confused. They do run test suites and also install c libraries and stuff on the docker image for libs that do need that What they don't seem to do is spin up separate processes or other crazy stuff we get up to when running the test suites | 12:48:51 |
alexfmpe | e.g. postgres libs test suites are disabled | 12:49:05 |
alexfmpe | https://github.com/commercialhaskell/stackage/blob/master/build-constraints.yaml#L8694 | 12:49:08 |
alexfmpe | Oh heh TIL they also have our silly test-suite-induced-cycle issues | 12:50:33 |
emily | FWIW, AIUI we used to track Stackage nightly but then stopped. it came up a while ago. I think maralorn had things to say about it. | 12:55:17 |
maralorn | I am in favor of tracking nightly. We simply didn’t have the energy to frontload the flag day. | 12:59:57 |
emily | is the flag day b/c jumping from current LTS to current nightly will effectively be the work of an LTS bump? | 13:01:09 |
alexfmpe | I assume that becomes closer to the truth the closer nightly is to become new LTS? | 13:06:06 |
alexfmpe | Or is nightly just very much broken in the early days? | 13:08:44 |