Packages that split hackage

[Duncan Coutts]

On Sun, 2009-11-15 at 14:51 -0500, Isaac Dupree wrote:
> Duncan Coutts wrote:
> > I've been trying to make a consistent set of packages to test releases
> > of Cabal and cabal-install.
> > 
> > What I've noticed is that there are a small number of packages that
> > almost cut hackage down the middle. 
> > 
> > At the moment the main culprit is
> > QuickCheck 1 vs 2
> > Whichever version you pick, you exclude a large swath of packages.
> 
> Can't you allow both 1 and 2 simultaneously?

Not if another package would indirectly depend on both.

> QuickCheck, at least, rarely involves re-exporting QuickCheck's types
> from packages that depend on QuickCheck, so allowing both is unlikely
> to cause build errors.

That's true, unfortunately Cabal does not have enough information to
know that.

> > haskell-src-exts, lots of packages depend (directly or indirectly) on
> > various incompatible 1.x versions.
> 
> parsing libraries are also relatively likely to be an internal-only 
> dependency, but who knows...

Right, same issue. What we need is a way to provide Cabal with that
extra information, eg via private build dependencies. We would have to
ask ghc to check the property that no types from private packages are
re-exported from exposed modules.

Duncan