On Wed, Jan 20, 2010 at 11:55 AM, Allan McRae <allan@archlinux.org> wrote:
As an aside, looking at holdver makes me realize just how stupid it is. It does not hold the checkout at the version specified, which I bet most people thinks it will. It just stop the pkgver being updated. e.g. I just build makepkg-git-19790101-1...
I know, it's completely broken. IIRC it only works with svn and cvs. That's just one of the many problems I have with this whole makepkg vcs stuff... Maybe that's the whole problem, the system might have been originally designed with cvs/svn in mind, and all the other version control were just added on top, not in an ideal way. The main complaints usually don't concern cvs/svn but all the others. Or at least the complaints that are not easily fixed. One of the last bug is actually about svn but seems like an easy fix : http://bugs.archlinux.org/task/17872 We need to have in mind the big differences between the different version control system : 1) ability to determine the last version remotely : only svn and bzr ? (I just remembered that cvs also just uses the current date) 2) ability to checkout the specified date/revision (a linear number which can be used in pkgver) : yes for cvs/svn, no for git/darcs I meant for a long time to have a closer and more precise look at the differences important for makepkg between all vcs. Just did not find the time/motivation for it. The second challenge is to have a system that is both flexible and factorize the most of code between two different PKGBUILD-vcs