On Thu, 2004-04-22 at 09:43, Gene C. wrote: > > Should kdenetwork be split so kmail is not included but instead moved to > Extras? This has to be a management nightmare. > For end-user apps (i.e. those in the menus), it will almost always be desirable to maintain a 1 .desktop file to 1 RPM mapping. And in fact we should be syncing the name of the package displayed in the package tool (including translations) with the .desktop file name. Or even making package management happen in terms of .desktop files. From a single-user standpoint, "menu editing" (at least in terms of add/remove items) and "package management" really have no reason to be different. The ideal user experience might be to effectively refcount each RPM by the number of users that have it in their menus, and when nobody has the item in their menus the package gets removed. And similarly you'd install a package by choosing what items to add to your menus. Unclear if there's a sane way to implement this, but this is a nice way for a desktop user to see things. Obviously it only works for desktop-app type of packages. You need another approach to deal with servers, though you could perhaps imagine a similar approach (when you enable the service, it automatically gets the required packages and punches the firewall to make the service go). Also, even if package add/remove were combined with menu editing, you still need an "update" or "get patches" kind of UI... though judging by the number of Windows viruses out there, perhaps the default should be to run this thing automatically every night unless the user opts out ;-) Anyway, we really should think about the issue much more broadly and not assume that the task at hand is "write a single tool that does package management." How can we get it smoothly integrated into the desktop? Maybe there are multiple tools for different tasks or kinds of user. There are related problems too, such as making it really easy for third parties to provide a set of RPMs with comps file on a CD or in an http directory, and handle that really nicely with an install wizard. Just some brainstorm ideas. Let's start with what the user should see though, and then figure out how to implement our closest approximation. Havoc
