On Mon, Jul 16, 2007 at 11:16:28AM +0200, ngaba@petra.hos.u-szeged.hu wrote:

Well, it needs further investigation and testing, but I don't see why that provdiff stuff needed in add.c: upgrade is a "remove then add" process, remove_commit and add_commit should manage requiredby fields properly. However, see my next e-mail today.

I told Dan this part looked rather messy. The problem I think is that we lose all works we do in deps.c For example, we have this in checkdeps : 282 if(alpm_depcmp(pkg, depend)) { 283 _alpm_log(PM_LOG_DEBUG, "checkdeps: dependency '%s' has moved from '%s' to '%s'", 284 depend->name, alpm_pkg_get_name(oldpkg), alpm_pkg_get_name(pkg)); 285 satisfied = 1; 286 break; 287 } We already compute this information here, it's stupid to compute it again in add_commit . The problem imo is that we need to store it somehow, and only register it to the database when the transaction has been commited. But I think it's a more general problem we have in all the dependencies code (as you already noticed for sortbydeps / checkdeps / resolvedeps) that we lose all informations we compute (but maybe that general problem concerns also the whole pacman/libalpm code).