Todd Vierling wrote:
> On Tue, 21 Mar 2006, Johnny Lam wrote:
> 
> 
>>You can force all packages that directly include ghc/buildlink3.mk to build
>>with ghc to have a dependency on gmp by modifying ghc/buildlink3.mk so that
>>gmp/buildlink3.mk is included outside of the BUILDLINK_DEPTH guards, i.e. move
>>that inclusion to the end of the file.  Should I go ahead and do this?
> 
> 
> Sounds reasonable.  We don't usually do this, but perhaps it's a concept we
> should employ where it is appropriate.  There are times when an indirect
> dependency really should be registered because its ABI has a direct impact
> on the final dependent.  This may have repercussions for pkgtools/revbump; I
> don't know for sure.

I think we're okay here.  finddepends(1) already looks for indirect 
dependencies, so it doesn't need to do anything different to handle this 
case where we are essentially promoting an indirect dependency to a 
direct one.  I think we should be wary of doing this all over pkgsrc, 
but using this technique does seem to solve this problem here for ghc.

> If this is done, we need to bump PKGREVISION of all *direct* dependencies of
> ghc, of course.  :)

Luckily, these were the packages that Thomas already directly added 
gmp/buildlink3.mk to and bumped the PKGREVISION, so the proposed change 
wouldn't force us to do it again.

I will go ahead with this change.

	Cheers,

	-- Johnny Lam <jlam@pkgsrc.org>