[arch-haskell] Re: habs updates

[Magnus Therning]

On Mon, Nov 1, 2010 at 14:28, Peter Simons <simons at cryp.to> wrote:
> Hi Magnus,
>
>  >> * packages provided by GHC are no longer listed as dependencies
>  >
>  > I strongly disagree with this one, I'd rather see the dependency list
>  > continue to include the meta-packages provided by the ghc package.
>
> Remy has put time and effort into developing that feature, and it's been
> in Git for almost two weeks. Now, you disagree strongly. Could you
> please describe how you intend to resolve the current situation? Would
> you like Remy's patches revoked? And if that is the case, exactly which
> ones? What do you suggest we do?

Well, I didn't know he was doing that.  Which is my fault, I haven't
been reading through all the patches that has gone into
cabal2arch/archlinux.

So, this is how I understand the situation:

In the past, when cabal2arch saw a dependency on 'base', then it would
insert a dependency on 'haskell-base'.  The 'ghc' package then states
that it provides 'haskell-base', and pacman (and other package
managers) could use this to resolve dependencies correctly.  It was
then the job of the 'ghc' package maintainer to correctly state all
the packages that are provided.  (I should also note this behaviour
was added with commit d4107f03211c79e2dba4,
http://github.com/archhaskell/cabal2arch/commit/d4107f03211c79e2dba4304ac9dadaa7da1b06e1).

The situation now is that 'cabal2arch' (or rather 'archlinux') again
knows what packages are provided by the 'ghc' package.  Resulting in a
list that has to be maintained by a person who isn't intimately
involved in packaging 'ghc' (like this commit:
http://github.com/archhaskell/archlinux/commit/4ce5f0415ebc070640b29439b036e8a1eeedfdd9).
 AFAICS this also forces a re-release of cabal2arch when the list of
core packages changes.

Now, I'm convinced that depending on the core packages through the
pacman mechanism, and therefore relying on proper 'provides' in ghc,
is the correct thing to do.  Feel free to convince me otherwise.

As to how we resolve this, well, this is why we have a versioning
system.  Practically, backing out Remy's changes is perfectly doable.
However, before we discuss the practicalities I'd like to know why
Remy felt it was desirable to revert to the old behaviour at all.

/M

-- 
Magnus Therning                        (OpenPGP: 0xAB4DFBA4)
magnus＠therning．org          Jabber: magnus＠therning．org
http://therning.org/magnus         identi.ca|twitter: magthe