(tl;dr: slightly more detailed version of what Julian said) On Tue, Jun 10, 2014 at 04:56:52PM +0100, Diarmuid O Briain wrote: > I was asked as part of a course I am doing to look at DNF/Libsolv as is > part of the migration path for YUM in Fedora and if Debian based OS apt-get > / Aptitude needs replacement or adaption to incorporate hawkey/libsolv. I hate to do your homework (at least it sounds a bit like it), but well: In a way we are way ahead on this as we don't limit ourself to one solver or throw our ecosystem out of the window just because we want to try out another. You can use any¹ APT tool already with any² solver you might like. ¹ all expect aptitude basically, as all others from apt to softwarecenter delegate their needs to libapt (over multiple layers like e.g. also packagekit) which abstracts away all the glory details of talking … ² via EDSP (see doc/external-dependency-solver-protocol.txt) This is still in active development (MultiArch support was added just minutes ago in the 1.0.4 upload of apt – all my fault, could have been months ago…), but you can already try it on a singlearch stable version of Debian if you feel like it without disrupting your workflow or invalidating your warranty (if you had one). So in short: No need for a replacement, but we would welcome the addition of a new contender in the game. > I was wondering what are the thoughts of Debian developers in this ? What > is the development roadmap for apt-get etc... The general roadmap in this area is to explore which wonders the various available solvers can provide and improve our interaction with them. If we have winner(s) there is nothing really stopping use from binding them closer or in the long run replace our homegrown inbuilt heuristic with it entirely, but at least I am not in a hurry: There is e.g. the question if SAT is actually a good enough answer for this problem or if e.g. ASP is a better one. And even if they aren't working in 100% of the cases, our heuristics tend to be pretty good. For the lack of a better one I refer to comparison [0] which I have lengthy critics written e.g. in [1], but I still like to read it as a confirmation of "heuristics ought to be enough for anyone" (for now) in the most common real world scenarios. > I would appreciate the thoughts of someone who knows something about this. I have cc'ed Stefano in case he wants to chirp in on EDSP some more. You can also look up CUDF on http://www.mancoosi.org/cudf/ for more on this topic. I am not following "the other side" of the package universe as close as I would like to, so I am not sure if there is some support there, but I at least haven't seen it. Beside that, you have already found the right place as deity@ is the mailinglist were all the cool people hang out if it comes to APT. I am not sure if people interested in cupt are lurking here, so if you want to talk to them as well, you might be better of inviting them, too. In case you actually want to discuss anything further of course. Best regards David Kalnischkies [0] http://www.mancoosi.org/measures/packagemanagers/2012/ [1] http://mancoosi.org/~abate/package-managers-comparison-take-2#comment-326
Attachment:
signature.asc
Description: Digital signature