On Sun, Oct 19, 2014 at 01:34:13PM +0200, Holger Levsen wrote: > cc:ing the apt maintainers to get their opinion on making this the default... [Disclaimer: I have written the APT part of it. I might be biased.] Hell no – as this isn't the point of the implementation. It is intended to help researchers and developers alike to experiment with resolvers in "normal" situations rather than fabricated snapshots as you can't jump to conclusion about the "greatness" of a resolver based on one single test – and experimentation is basically the opposite of what you would want the default resolver be: The multiple levels of indirection in the design are e.g. great for playing, but horrible from a speed point of view. It would also move a lot of stuff onto every debian system… I guess I don't have to tell you what it means to pull in prio:extra packages from a build-essential prio:important (and for all practical proposes nearly essential:yes) package. It's also not complete yet. The CUDF protocol for example learned very recently what MutliArch is and how to not explode if its encountered, I wouldn't exactly call this path exceptional well tested as a result. [Disclaimer: I have written MultiArch in APT as well, so I flipped between both for quiet a while… not fun, so biased again]. It isn't even clear yet which technique is strictly superior to what we have at the moment as our naturally grown heuristic-solvers do pretty well in real world scenarios. If you don't trust me on that: http://mancoosi.org/~abate/package-managers-comparison-take-2 which talks about http://www.mancoosi.org/measures/packagemanagers/2012/ [Disclaimer: I have written a lengthy comment there which should give you a hint why I come to this conclusion … bias the third] (And yes, while I consider it a bug that apt isn't able to figure this one out without a little help, I don't really consider this case an important realworld scenario. In the end, the times I will change init systems is hopefully far below the GR proposal count for that. Not only because I am lazy, but because it would mean that everyone would have done a pretty crappy job making Debian jessie the best release ever if no init works reliably [totally unbiased on this one] ) Best regards David Kalnischkies
Attachment:
signature.asc
Description: Digital signature