[Haskell-community] Haskell Platform as the default recommendation considered harmful

[Michael Snoyman]

On Mon, Sep 21, 2015 at 6:06 PM, John Wiegley <johnw at newartisans.com> wrote:

> >>>>> Christopher Allen <cma at bitemyapp.com> writes:
>
> > 1. ghc-pkg
> > 2. The global vs. user package databases
> > 3. Cabal
> > 4. What Platform does to their build environment
>
> > New Haskell users are least equipped to understand errors caused by:
>
> > 1. Package version conflicts
> > 2. Something innate to their install, rather than something wrong with
> their
> >    package constraints.
>
> Unless I am mistaken, every one of these points relates to the usage of
> cabal
> after installing the platform, rather than to the Platform itself?
>
> Once Stack is in the platform, do you agree it will rectify every concern
> you
> mentioned? If so, I'd rather not abandon the platform, only to be seen
> re-adopting it after a few months.
>
> John
> _______________________________________________
>

In theory: yes, using Stack with the Haskell Platform will solve this
problems. However, theory does not always add up to reality, for various
reasons we may not be able to predict. I'll give a few examples:

* There's no clear timeframe for the next HP release including Stack
* It's unclear what exactly will be in the global database at that time
* It's unclear how MSYS2 will be handled on Windows
* There are a few bugs that have been reported to the Stack and conduit
repos recently[1][2] about bad interactions with the global packages from HP
* Non-unique install package IDs can lead to shadowing with HP[3]
* There's no talk of the upgrade story around HP

We can find all of these things out, but making decisions today based on a
theoretical future state makes no sense to me.

Chris has done a great job here of collecting a lot of the reasons why so
many people have been advocating against the Haskell Platform. I do not
believe the current wording on the downloads page, nor any of the
discussions we've had via email or issue trackers, comes close to
reflecting what the community actually believes is best practice today.

My recommendation: we look at the current state of all tooling, and decide
what will be the best choice for most new users to Haskell. We don't need
to cater to experienced users, since we all know about the choices. The
downloads page should be explicitly about first impressions with new users.
And we should assume those first impressions will be happening today, not a
month, six months, or two years from now.

Michael

[1] https://github.com/commercialhaskell/stack/issues/980
[2]
http://stackoverflow.com/questions/32444762/cabal-repl-wont-run-cant-load-so-dll-for
[3] https://github.com/haskell/cabal/issues/2830
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.haskell.org/pipermail/haskell-community/attachments/20150921/66f348b1/attachment.html>