[Haskell-community] haskell.org download page

Sylvain Henry sylvain at haskus.fr
Tue Aug 30 14:23:11 UTC 2016


Stack ensures that the Haskell environment stays as specified. I find it 
very useful that GHC is included in this spec as we don't necessarily 
(want to) control the external installations. For example, when GHC 8.0 
has been pushed in the official repos of Arch Linux, stack just 
(re)installed the old one transparently and I didn't had to deal with it 
in my projects, nor rollback the Arch Linux package.

If new users try to use Haskell on such distros, they would better use 
stack with the latest LTS (implying GHC 7.10.3) than use GHC 8.0 from 
the official repo and deal with broken dependencies during the 
transition period. Moreover it is still possible to use "ghci" and "ghc" 
for small scripts and interactive sessions.

I agree that stack should be the recommended build tool, hence:
1) For the download page, I would say the "minimal install" (or "net 
install") is stack and should be on top.
2) On Linux distros where the HP is just a meta-package, as long as it 
brings the latest haskell-stack package, it shouldn't do no harm (but by 
default stack may not even use the provided GHC, depending on the 
current LTS and on the GHC version in the repo).
3) For bindist HPs, if the GHC version doesn't match the one used by 
stack, I don't see the point of including stack in the HP (or using the 
HP at all) because the first thing stack will do is to download another 
GHC. Maybe it would be possible to provide a default stack.yaml in the 
HP that would force stack to use a resolver associated with the provided 
GHC and libraries, in which case it would make more sense. Maybe bindist 
HPs could be automatically generated to match Stackage's LTS releases? A 
"HP Full" release could provide more packages from the LTS.
4) A "get-started" example using stack should be added on haskell.org

Cheers,
Sylvain


On 30/08/2016 12:23, Simon Marlow wrote:
> The choice boils down to whether you want stack to manage your GHC 
> installation or not.
>
> I personally find it distasteful.  This has been the biggest blocker 
> for me using stack, it wants to control more of my workflow than I 
> want to give it, leading to an overlap of responsibilities.  (I do use 
> stack, but only with external GHC installations, and I often get into 
> a mess when it tries to download another GHC)
>
> Having said that, is it better for new users to delegate the GHC 
> installation to stack?  I don't know.  It certainly has the downside 
> that you can't just type "ghci" and get a prompt.
>
> The world seems simpler when it consists of
> - GHC installations
> - build tools that use your GHC installations and manage local package 
> building
>
> But when my build tool manages my GHC installations, there's now a 
> layer of abstraction in the way of GHC and I can't figure out how to 
> interact directly with GHC any more.  Also I can't use cabal (which I 
> often do).
>
> So, I'd argue for HP minimal to be the default download option. By all 
> means recommend stack as the default build tool - I'm sure it's less 
> problematic for most people to get Stackage by default, and cabal 
> isn't set up to use Stackage out of the box.
>
> Can't we get rid of HP Full?  I don't see a use for that any more.
>
> Cheers
> Simon
>
>
> On 29 August 2016 at 16:29, Nicolas Wu <nicolas.wu at gmail.com 
> <mailto:nicolas.wu at gmail.com>> wrote:
>
>     Hello,
>
>     I think having multiple options is confusing to beginners, and so
>     I'd like to see a single download option on the download page.
>
>     For me it's important that we have a way for beginners to use
>     tools like ghc and ghci on the command line directly in order to
>     run small throw-away programs.
>
>     The decision about how to manage projects and their dependencies
>     should be open and isn't for beginners, whether that be using
>     stack or cabal: both have their merits, and I don't want to push
>     one over the other. The default installation should provide both
>     of these as well as other tools core to building ghc.
>
>     As such, I'm in favour of having the HP as the only option.
>
>     Nick
>
>     On Sat, Aug 27, 2016 at 5:50 AM Jason Dagit <dagitj at gmail.com
>     <mailto:dagitj at gmail.com>> wrote:
>
>         Hello all,
>
>         I just realized that the Minimal installer listed first on the
>         Downloads page (https://www.haskell.org/downloads
>         <https://www.haskell.org/downloads>) is deprecated and "dead".
>         This creates an unfortunate situation where our top suggested
>         way to get haskell immediately tells the user it's dead.
>
>         I think that we should remove mention of the minimal installer
>         ASAP on the grounds that the HP now comes in minimal and full
>         variants.
>
>         Furthermore, I would like to make the recommendation that we
>         list the HP above other methods as even the minimal HP
>         installer ships with stack (at least on windows it does).
>
>         Between the two changes, I think the first one is crucial and
>         the second one is merely reasonable.
>
>         Thanks,
>         Jason
>
>
>         _______________________________________________
>         Haskell-community mailing list
>         Haskell-community at haskell.org
>         <mailto:Haskell-community at haskell.org>
>         http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-community
>         <http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-community>
>
>
>     _______________________________________________
>     Haskell-community mailing list
>     Haskell-community at haskell.org <mailto:Haskell-community at haskell.org>
>     http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-community
>     <http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-community>
>
>
>
>
> _______________________________________________
> Haskell-community mailing list
> Haskell-community at haskell.org
> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-community



More information about the Haskell-community mailing list