Mark I see where you are coming from but I don’t see the strong connection between the process of creating the tarball and carrying out the installation -- they may sometimes happen at the same time but not in general. I would have expected the former to work in as wide a range of environments as is practicable while the latter (of course) must be highly controlled (hence the checks that the core packages are present, etc.). I would also like to offer some input into the proposed reworking of the platform packaging. I am guessing that we all have quite mature build systems for carrying out the packaging of the platform. For me the key interface is the tarball that gets inserted into the build process -- what we distribute through the HP website. If that is the general case then keeping the way the tarball works stable should be a principal aim of any reworking of the generic platform packaging. It seems to me to work quite well and not be in need of much change. *** That said, I do think a DESTDIR parameter for the build process would be useful. I have worked out a scheme for doing this by just modifying scripts/install.sh in a backwards compatible way (i.e., no DESTDIR parameter and the script behaves just as it does now). I will raise a ticket on the revived issue tracker… Chris From: Mark Lentczner [mailto:mark.lentcz...@gmail.com] Sent: 20 May 2012 00:44 To: Chris Dornan; Andres Löh; haskell-platform@projects.haskell.org Subject: Re: Early trial builds of 2012.2.0.0 In summary - cabal-install 0.14 outputs more info than the build scripts expect under some circumstances. The proposed fix simply made the build scripts ignore the extra, and should be benign... However, after thinking about it, I think we should *not* make that change in this release: The extra information is indication that the packages about to be will likely cause havoc vis-a-vis the package database cabal is running against. If you are packaging packages for a distribution of Haskell Platform, you should be doing this in an environment where that doesn't happen. In particular, you should be using a package database that starts with only exactly that which GHC 7.4.1 provides. See src/macos/Makefile through the target "build" for how to do this. For the next HP (2012.4.0.0) I intend to re-write this whole part of the build process. It will still rely on cabal to do the build ordering (which is what this part of the build is doing), but I'm not sure where that code will end up - and will certainly be re-written. At that time I'll take into account the extra info - probably issuing warnings or errors. If I had my druthers (and if I find the skillz) I'd like to be able to make the whole hermetic build process that the mac build does be generic. - Mark
_______________________________________________ Haskell-platform mailing list Haskell-platform@projects.haskell.org http://projects.haskell.org/cgi-bin/mailman/listinfo/haskell-platform
