On Wed Dec 31 2014 at 4:25:12 PM Herbert Valerio Riedel <hvriedel@gmail.com> wrote:

Hello Michael,

On 2014-12-31 at 15:10:37 +0100, Michael Snoyman wrote:
> tl;dr: Now that ghc (the library) doesn't depend on Cabal (the library),
> can we remove Cabal from the global package database installed with GHC?

[...]

> For both of these reasons, I think we should limit the number of packages
> included in the global package database. One seemingly small step we could
> do on that front is not include extraneous packages. In GHC 7.10rc1, that
> includes Cabal and xhtml:

btw, haskeline and terminfo should be considered as well, according to
your argument.

Yes, I didn't mean to imply that my list was exhaustive, those were just two examples that jumped out at me.

> both packages are in the global package database, but could just as
> easily be removed from there and installed by users. The motivation
> for that would be to avoid problem (1) above.

However, as for xhtml, haskeline and terminfo, we had to register/expose
them in the global pkg DB due to

https://ghc.haskell.org/trac/ghc/ticket/8919

IIUC, this comes down to the fact that distros want to install all packages into the global database, whereas the pain point I'm describing here is from users wanting to *avoid* having things in the global database. Perhaps there's a simple solution here: if end-users non-distro installations of GHC unregistered those packages, would it cause them any problems? Perhaps having those packages in the global database is something that *only* distros need.

and as for Cabal, I have been told (maybe Duncan can weigh in...?) that
Haskell implementations need to provide it (together with a `*hc-pkg`
executable) in order to conform to the CABAL specification[1].

If that's the case, I would request that we modify the CABAL specification. I wouldn't want us to make a bad technical decision because a spec told us to do so.

Michael