reflex-frp
Dependencies on multiple versions of haskell-src-exts
On nixpkgs master (at least as of 395bcc0b27ce0d3c8a9af09737ba076a089a8dd6), I can't get reflex-0.6.1 to build using because it's indirectly depending on two different versions of haskell-src-exts ( https://pastebin.com/BnHagELV ). I am not sure if this is reflex's fault, or how to diagnose this. If you can give me pointers I can go digging, or refile this issue somewhere better.
It would be very cool to be able to get an 0.6-series reflex through straight nixpkgs, and we're almost there.
Is there any chance of pushing a metadata revision for the 0.6.1 release, allowing haskell-src-exts < 1.22? That would go a long way to fixing this. I think you'd be able to close the reflex issue after that, and I can take the rest up with nixpkgs.
We recommend using https://github.com/reflex-frp/reflex-platform. Nixpkgs shouldn't be doing overrides that break package-version coherence, so if there is some override there that just affects one of haskell-src-meta or reflex that should be changed to affect the whole package-set.
Nixpkgs seems to override haskell-src-exts in three places: hlint, hoogle and lambdabot-plugins. The reason it affects package coherence here is because of the reflex test suite that wants the codebase to be hlint-clean.
Ah. hlint isn't used as a library right? It should be a build-tool-depends. That might help (should be separate package db).
I don't think there's a major problem with us changing the way we invoke hlint, if that is helpful.
On Tue, Jun 11, 2019 at 8:54 PM endgame [email protected] wrote:
Alas, no: https://github.com/reflex-frp/reflex/blob/develop/test/hlint.hs
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/reflex-frp/reflex/issues/309?email_source=notifications&email_token=AAI2KYEQC2DHAFMNS47TYQLP2BCM3A5CNFSM4HNICF5KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXO5RWA#issuecomment-501078232, or mute the thread https://github.com/notifications/unsubscribe-auth/AAI2KYDM6T735DJWEKXGDADP2BCM3ANCNFSM4HNICF5A .
I think it would be. Seems like the right thing to do is make hilnt a build-tool-depends, and have the hlint test invoke hlint in a subprocess?