[Haskell-cafe] Minimal Haskell Platform

[Tillmann Rendel]

Hi,

Simon Hengel wrote:
> Where would something like the HP actually make sense?

In my experience, the Haskell Platform makes it significantly easier to 
use Haskell on Windows.

On Windows and/or for the typical Haskellers on Windows, it is hard to 
build a package that depends on external C libraries or uses a configure 
script, however trivial. This is because on Windows, there is no way to 
manage installed C libraries without understanding C development. But 
most Windows users are not C programmers, and some Windows Haskellers 
even learn Haskell as their first programming language. In extreme 
cases, ghc and cabal are the first command-line tool they learn.

So on Windows, before the Haskell Platform was introduced, `cabal 
install some-random-package` usually failed for some dependency that 
required a C library. With the Haskell Platform, `cabal install 
some-random-package` usually works. I guess there are two effects:

  1. The Haskell platform seems to bundle some packages that require 
external C libraries or configure scripts. The network package comes to 
mind.

  2. The existence of the Haskell platform encourages non-Windows 
developers to avoid dependencies that are not in the Haskell platform. 
Because of effect 1, this translates to non-Windows developers avoiding 
dependencies that are hard to install on Windows.

So from the perspective of using Haskell on Windows, the Haskell 
Platform is very successful, and installing it is good advice. If 
anything, it should be bigger, not smaller. And it should focus on 
libraries with external dependencies.


So could we make a Haskell-on-Windows-Platform and drop the Haskell 
Platform for the other platforms? This would give Haskellers on Windows 
the first effect, but it would disable the second effect. I'm not sure 
how much the first effect is worth without the second.


Why should we care about Haskellers on Windows? I think there are some 
important classes of new Haskellers that are on Windows, at least 
initially, and if we want to convert them to long-time Haskellers, we 
need to support them in their context. For example, here are three 
reasons to start your Haskell experience on Windows:

  1. You might be a student at a university where the lab computers run 
Windows. A minimal installation without excellent support for libraries 
would be fine for following a lecture on Haskell, but you will not take 
Haskell serious if it doesn't give you easy access to libraries. In 
fact, installing Haskell libraries can be very easy, but only if you get 
the external dependency issue out of the way.

  2. You might be a not-yet-programmer who wants to use a embedded DSL 
for your domain, and you're on Windows "by default" or because of the 
tradition in your field. We like to claim that Haskell is good for 
embedded DSLs, so we should support this usage scenario.

  3. You might be a not-yet-programmer who wants to use an external DSL 
for your domain, and you're on Windows as above, and you want to become 
a power-user of the DSL. So maybe you start reporting bugs and are asked 
to  `cabal install` the development version to check fixes. Or maybe you 
want to copy-and-adapt a little Haskell program that uses a library 
interface to the DSL for scripting purposes. We see both of this 
happening for pandoc, for example.

Also, Haskell is a high-level language, and it would be somewhat silly 
if Haskell programs would not be easily portable due to low-level issues.

   Tillmann