hercules-ci-agent: Fix build on aarch64-darwin - #463879
Conversation
d8a57d8 to
544c4e8
Compare
|
Is there anything else to be done? |
|
It's quite unsatisfactory to have such a targeted and unprincipled hack without any path forward to remove it anytime soon. If older compilers are not affected by this, can't we just build |
|
I agree that the solution isn't pretty necessarily, but loading a Haskell program + RTS + nix libraries into a single process global symbol namespace works fine, as demonstrated by
This would require hydra to build a significant portion of this package set, whereas currently it only has to build ghc itself at this version. |
9.8.4 is also affected. I believe 9.8.2 is the latest in the 9.8 series without the linker changes that cause the build failure. |
|
Why was this working fine with 9.8.4 then? |
@sternenseemann it actually worked fine with both 9.8.4 and 9.10.3. |
cc @NixOS/darwin-core, maybe you have some input to this as well. |
|
The only thing we build from dyld are a couple of tools ( |
|
I'm not convinced, yet, that this the right thing to do. Imho, this should be fully bisected. I started that, but didn't finish. I have it down to: git bisect view --oneline |
|
Thank you @wolfgangwalther. I haven't quite figured that one out either. The bad commit produced a different problem, and some more recent commits actually built successfully somehow. The good news is I've found a new workaround that doesn't require any changes to GHC or anything like that. |
|
hercules-ci-agentstill needs the fix by @sandydoo in order to build.I've made it conditional to nullify the possible impact elsewhere.
Long term I'd like to get rid of the TemplateHaskell that's triggering this problem, but that's blocked on a number of issues.
So a workaround is needed.
Things done
passthru.tests.nixpkgs-reviewon this PR. See nixpkgs-review usage../result/bin/.Add a 👍 reaction to pull requests you find important.