FS#16816 - Maybe it's time to move the Haskell Platform to Community?
Attached to Project:
Community Packages
Opened by Skottish (skottish) - Saturday, 24 October 2009, 02:22 GMT
Last edited by Vesa Kaihlavirta (vegai) - Friday, 11 March 2011, 07:00 GMT
Opened by Skottish (skottish) - Saturday, 24 October 2009, 02:22 GMT
Last edited by Vesa Kaihlavirta (vegai) - Friday, 11 March 2011, 07:00 GMT
|
Details
Description:
Arch is becoming one of the premiere Linux platforms for Haskell development thanks to the people working on Arch-Haskell. There's a reoccurring theme on their mailing list about bringing the Haskell Platform to Community, and I feel that it's a great opportunity for Arch. I understand that it doesn't have hundreds of votes, but since there's a group of people who are more than qualified to do it, have a vested interest in Haskell's future, and are currently generating high quality PKGBUILDs and install files, I think that it would be good for Arch and Haskell hackers alike to make this move. |
This task depends upon
Closed by Vesa Kaihlavirta (vegai)
Friday, 11 March 2011, 07:00 GMT
Reason for closing: Fixed
Additional comments about closing: haskell-platform-2011.2.0.0
Friday, 11 March 2011, 07:00 GMT
Reason for closing: Fixed
Additional comments about closing: haskell-platform-2011.2.0.0
Anyway, what packages would need to be brought into [community]? More importantly, who will maintain them? One of the Arch Haskell leads is a dev so if he wanted to put them in [community] there is nothing stopping him apart from apparently not wanting to.
If we had an autobuilder bot, then this would be trivial.
I see two solutions for this:
1. Make major changes to the GHC 6.10 package and to HP to make sure the situation above is avoided.
2. Wait until HP requires GHC 6.12; this version of GHC doesn't ship with the base set of libraries. (I'm all in favour of early packaging of the next version of HP, as soon as there is a preliminary list of what it contains.)
On this request: My concern was that the idea may get reject on the basis that the platform doesn't roll incrementally like everything else in Arch. In this case it doesn't seem important to me. When xmonad-light (or whatever it's going to be called) hits the repos, I think that we'll see a giant drop off in GHC downloads; Nearly everyone else left will be coder of some sort. From there every state-of-the-art package will most likely be interesting to those that are deep enough into Haskell to get something out of them.
On the topic of rolling release: I personally see no difficulty in satisfying both camps, HP has strict depencencies on specific versions of packages, there is of course nothing that prevents those packages to also be avilable in newer versions on AUR (albeit they have to have different names, similar to how some bleeding edge packages are available as X-git or X-svn, maybe we should have haskell-X-hackage).
Also, the HP release schedule calls for a new version every few months, so it's unlikely to lag VERY far behind the bleeding edge of individual packages (except in cases where it makes sense, like parsec2 vs parsec3).
The best way to get this done is to convince a dev/TU to bring the packages in. I guess Vegai would be the one to ask. Otherwise, apply to be a TU and do it yourself.
Closing. Not a bug.
One problem is that arch users are eager to report newer versions of libraries, and arch devs/tus are often eager to build+upload those newer versions. And since we don't keep older versions, this will break things like HP easily.
1) Leave things the way that they are and keep the regular rolling release model.
2) Push the platform into community and allow anyone that needs alternate versions of libraries to use cabal (multiple or cutting edge versions) or pacman (cutting edge versions) to manage their own repos.
3) Add functionality to pacman to allow for the installation and management of development libraries allowing for users of any language that supports multiple versions to use it to manage their development libraries.
The latter would be ideal and if I had the skills I would submit a patch for review. The middle is certainly acceptable and probably a very good solution seeing that the majority of Haskell downloads in Arch is surrounding xmonad. The former is working at the moment and is certainly the easiest from a distro-package management point of view.
.cabal/
|-- config
`-- packages
`-- hackage.haskell.org
|-- 00-index.tar
|-- 00-index.tar.gz
`-- X11-xft
|-- 0.2
| `-- X11-xft-0.2.tar.gz
`-- 0.3
`-- X11-xft-0.3.tar.gz
e.g. have "haskell-platform" provide "x11-xft=0.2" and then provide a x11-xft-0.3 package for those that want something up-to-date.
Or something like that... much like perl does with the system-perl/vendor-perl stuff.
Implemented so that extra has all the packages that make up the platform, and haskell-platform itself is an empty package that depends on those specific versions.