On Wed, Aug 11, 2010 at 5:04 PM, Peter Simons <simons@cryp.to> wrote:

Hi Evangelos,

 > I can see that three out of the five packages you list were last  > updated in May and are no longer marked out-of-date. This means that  > for those three packages at least, there's no reason for disowning.

right, the ones that are still out-of-date (and have been for 4 months) are:

 haskell-funcmp        http://aur.archlinux.org/packages.php?ID=18122  haskell-hsdns         http://aur.archlinux.org/packages.php?ID=17745

The other three packages I mentioned have been updated. Still, it's interesting to compare the times of the update on Hackage and on AUR:

 haskell-hopenssl      http://aur.archlinux.org/packages.php?ID=17743

   Updated on Hackage:          2 Mar 2010    Updated on AUR:             22 May 2010

 haskell-hsemail       http://aur.archlinux.org/packages.php?ID=17746

   Updated on Hackage:         25 Mar 2010    Updated on AUR:             30 May 2010

 haskell-hsyslog       http://aur.archlinux.org/packages.php?ID=17751

   Updated on Hackage:         26 Feb 2010    Updated on AUR:             22 May 2010

This looks to me like the minimum time frame in which an AUR package gets updated by Don Stewart is 2 months or more. I don't think that's particularly impressive. IMHO, the Arch Linux team should re-consider whether it's a wise choice to grant one person a monopoly on all Haskell-related packages on AUR.

 > Also, you mention that you were referred to the Arch-Haskell mailing  > list. If you posted there, can you provide a link to your opening  > message? I'd be interested to see what replies you got.

At <http://www.haskell.org/pipermail/arch-haskell/2010-May/thread.html>, you can see the thread "Can I help to maintain Haskell packages on AUR?", which consists of the replies I got. My original messages was this one:

 | From: Peter Simons <simons@cryp.to>  | Subject: Can I help to maintain Haskell packages on AUR?  | To: Don Stewart <dons@galois.com>  | Date: Tue, 25 May 2010 18:35:49 +0200  | Message-ID: <87vdac0wii.fsf@write-only.cryp.to>  |  | Hi Don,  |  | the AUR packages haskell-{hsdns,hsemail,funcmp} are out-of-date for quite a  | while now. According to gour, the reason is that the ArchHaskell team lacks  | developers to perform the updates.  |  | I would be happy to contribute some time and effort to the project. Can I help  | in some way?  |  | Take care,  | Peter

Don told me to "stay tuned", but nothing happened.

 > I apologize if your requests seem to go unfulfilled, but --to me at  > least-- the situation with arch-haskell and the AUR is a bit unclear,  > so I'm not sure what action should be taken.

Well, when a Haskell package is obviously out-of-date on AUR and someone offers to take over maintenance, then disown it and let that person handle the package. That's the way it's done for everything else, and the procedure seems to work fine. It's a mystery to me why the ArchLinux team deviated from that procedure for Haskell packages in the first place.

Take care, Peter

I agree with your last paragraph. I've disowned all five packages since it makes sense to have their maintainer on Hackage also maintain them on the AUR. Go ahead and adopt them. Thanks. :)

On Wed, 2010-08-11 at 16:04 +0200, Peter Simons wrote:

Well, when a Haskell package is obviously out-of-date on AUR and someone offers to take over maintenance, then disown it and let that person handle the package. That's the way it's done for everything else, and the procedure seems to work fine. It's a mystery to me why the ArchLinux team deviated from that procedure for Haskell packages in the first place.

Take care, Peter

Wasn't there a Haskell update script that automatically did the updating?

On 08/11/10 18:44, Ng Oon-Ee wrote:

On Wed, 2010-08-11 at 16:04 +0200, Peter Simons wrote:

...

Well, when a Haskell package is obviously out-of-date on AUR and someone offers to take over maintenance, then disown it and let that person handle the package. That's the way it's done for everything else, and the procedure seems to work fine. It's a mystery to me why the ArchLinux team deviated from that procedure for Haskell packages in the first place.

Take care, Peter

Wasn't there a Haskell update script that automatically did the updating?

I thought so... I think the reason we (Haskellers) got this special case for the Haskell-Hackage packages is we thought we had a cool tool that would do it all automatically, and everyone (Arch and/or Haskell I guess) was excited about it. But evidently something in the technology or the process stopped/didn't work quite that well (yet) (And until/unless it does, using the usual AUR rules on ownership makes sense to me.) -- I'm not sure what sort of obstacles remain currently: probably Don knows but is busy with other work... -Isaac

Hi Ng,

Wasn't there a Haskell update script that automatically did the updating?

yes and no. Haskell packages have build instructions in "Cabal" files, and there is a tool that converts Cabal files into PKGBUILD format. That conversion in automatic, and the result of the conversion can be uploaded to AUR more or less automatically. However, Cabal and PKGBUILD differ quite a lot, and there is no obviously correct mapping between the two. For example, the Haskell package manager can have any number of versions of the same package installed simultaneously. Thus, packages can specify fairly sophisticated dependencies that don't translate easily into something Pacman will understand, like this one: base (3.*), bytestring (≥0.9), containers (≥0.1), directory (≥1), extensible-exceptions, filepath (≥1.1), HTTP (≥4000.0.5), mtl (≥1.1), network (≥2), old-time (≥1), parsec (≥2.1), pretty (≥1), process (≥1), random, texmath (≥0.3), utf8-string (≥0.3), xhtml (≥3000.0), xml (≥1.3.5 & <1.4), zip-archive (≥0.1.1.4) or base (4.*), bytestring (≥0.9), containers (≥0.1), directory (≥1), extensible-exceptions, filepath (≥1.1), HTTP (≥4000.0.5), mtl (≥1.1), network (≥2), old-time (≥1), parsec (≥2.1), pretty (≥1), process (≥1), random, syb, texmath (≥0.3), utf8-string (≥0.3), xhtml (≥3000.0), xml (≥1.3.5 & <1.4), zip-archive (≥0.1.1.4) Furthermore, it is possible to install packages that have contradicting dependencies, i.e. a package X can depend on "foo 1.0" and Y can depend on "foo 2.0" -- yet both X and Y can be installed at the same time. This is not possible in ArchLinux, though. Last but not least, Cabal files can require other non-Haskell libraries that need to be linked to the generated executable. Now, those libraries exist in ArchLinux, too, but it's not always possible to derive the proper ABS package name from the name of the *.so file. One could run "pacman -Qo" to figure that out, but this would require the person who runs the "cabal2arch" utility to have all those libraries installed in their system, which is not practical. So, yes, the cabal2arch utility can generate PKGBUILD files automatically from Cabal. However, for non-trivial builds the generated PKGBUILD will not work without further editing by a human being. In cases where package have contradicting dependencies, the situation is even more complicated. Personally, I'd call the update procedure semi-automatic. Still, even that semi-automatic procedure needs to be run. To the best of my knowledge, the only person who can do that is Don Stewart, because he seems to be the only one who has access to the 'arch-haskell' account on AUR. I (and other people) have asked for access to that account so that we can help updating AUR, but unfortunately Don appears to be too busy to do anything towards that end. Take care, Peter