[darcs-devel] ghc version range

[Ben Franksen]

Am 22.08.20 um 16:58 schrieb Ganesh Sittampalam:
> As we discussed on http://bugs.darcs.net/patch2017, let's agree the
> minimum GHC version we want to support on HEAD from now.
> 
> Debian stable (buster) has GHC 8.4:
> https://packages.debian.org/buster/ghc
> 
> The next Debian release (bullseye) will have at least 8.8:
> https://packages.debian.org/testing/ghc
> 
> Judging by the timelines for buster, I would guess that it will come out
> mid-late 2021:
> https://wiki.debian.org/DebianBuster#Before_the_release
> https://wiki.debian.org/DebianBullseye#Before_the_release
> 
> We don't necessarily need to actually bump the minimum right now, but
> maybe we can agree on a version that is an acceptable minimum if some
> change requires it.
> 
> I would propose 8.8. That said, if we think the next major release is
> likely to be in less than a years' time, then maybe we should just say
> it's 8.4.

It's very hard to say. Getting out 2.16 took much longer than I
expected. That doesn't necessarily mean the next release will be delayed
similarly. It also depends on what we want to achieve with that release.

What Debian does or not does should inform our choice but should not be
the only criteron. For instance, the bug reports we got from packagers
lately were from gentoo and from homebrew (which is, I think, a MacOS
packaging system). We still have Debian 7 and even Debian 6 on some
older servers where I work; these have so ancient ghc versions that in
order to compile darcs you first have to install a current ghc anyway.

What I want to say here is that we should keep supporting older ghc and
library versions as far as possible, unless we depending on features
that we can't or don't want to live without.

> If possible I would like to resurrect the bits of
> http://bugs.darcs.net/patch1883 that depended on GHC 8.6, but it's not
> that important.

It would certainly be good to get rid of having to manually define Show
instances. As I remember this patch would help to at least reduce that?
Also, ghc 8.6 is the first version with a haddock that can render our
docs. So I'd go for requiring 8.6 on a purely technical basis if and
when that actually helps us to make darcs better or easier to maintain.
At the moment I do not feel any strain from supporting multiple ghc
versions. Testing that everything works is easily automated and can be
done offline every once in a while. (Since I upgraded my machine to
16GiB memory I can run all the builds in parallel w/o any problems.)

What would help most in maintainance ATM would be to have an Appveyor
instance that can build and run the tests on Windows.

Cheers
Ben