search

haskell / ghcup-metadata

GHCup metadata repository
33 stars 30 forks source link

Recommending GHC 9.6.x #220

Open hasufell opened 5 months ago

hasufell commented 5 months ago

Related discussions:

andreasabel commented 1 month ago

Stackage LTS 22 for GHC 9.6 still has not closed (LTS 22.39 released last week), but soon will: https://github.com/commercialhaskell/stackage/issues/7462#issuecomment-2436984765

9.8.3 is now in Nightly so we should be able to cut/release lts-23 soon based on it.

Once LTS development moves to GHC 9.8 I think 9.6 will be as stable as it gets and should get the status of the recommended GHC (or it will never get it).

hasufell commented 1 month ago

The FreeBSD bugfix has not been backported.

If GHC HQ doesn't do it, then I will. Which will take some time.

juhp commented 4 weeks ago

I meant to write this elsewhere (in Discourse), but I think it would be reasonable to have different default versions for different platforms when necessary. It doesn't make sense to hold back tier 1 platforms indefinitely, because of tier 2 or lower. Basically "everyone" considers ghc-9.6 stable at this point.

hasufell commented 4 weeks ago

I meant to write this elsewhere (in Discourse), but I think it would be reasonable to have different default versions for different platforms when necessary. It doesn't make sense to hold back tier 1 platforms indefinitely, because of tier 2 or lower. Basically "everyone" considers ghc-9.6 stable at this point.

I have explained this in some detail a couple of times in my talks: That is specifically against the philosophy of ghcup.

One of the major use cases of the 'recommended' tag is a university course of ~200 students doing some exercises. You definitely do not want different versions across platforms. The default installation should deliver the same toolchain on all platforms.

So the 'recommended' tag isn't meant to be a "this release is good on some platforms", but a "this works equally well across most platforms".

If GHC cannot deliver a consistent experience across platforms, then that's definitely a problem for us, but it also allows us to push back and say "go fix this".

hasufell commented 2 weeks ago

@AndreasPK what is the status of the RTS bugfix backport? Otherwise I will have to consider another downstream release (since it's a runtime bug... fixing in-place won't cut it).

https://gitlab.haskell.org/ghc/ghc/-/merge_requests/13276

bgamari commented 2 weeks ago

@hasufell that issue is, as far as we know, fixed in master and 9.8.3, as reflected in my comment on the merge request. Nevertheless, I would urge you not to pursue a downstream release before we discuss.