Hi! On Thu, 2014-09-04 at 23:10:30 +0200, Petter Reinholdtsen wrote: > [Guillem Jover] > > Hope that explains. > > Thank you, it does. And after giving it some thoughts, I believe the > background for my surprise is that I expect recommends to behave like > depends when APT::Install-Recommends is enabled.
Well, that only tells apt that it should try to install all Recommends, it does not tell it to change the semantics of the field. I don't think it will fail or complain much if say, one of the Recommends is uninstallable or not available for example, because it's still a Recommends. > Handling it like > something else break the priciple of least surprise to me. Handling it like anything other than a Recommends, breaks it for me pretty hard, in a way that makes it not make any sense as a different field than a Depends. > When I added > 'Recommends: apache2' to sitesummary, I expected it to behave as if it > was a dependency in the default case (APT::Install-Recommends enabled) > when installing sitesummary, while allowing sitesummary to be installed > without apache2 (or apache2 to be removed after installation) without > also having to remove sitesummary. > I realise now that is not how apt and dpkg really work. Perhaps they > should work like that? Am I the only one surprised by this? Policy is > not mentioning anything about expected postinst ordering for recommends, > but perhaps it should? As I mentioned before, I think that would be a bad idea, it effectively turns Recommands into Depends, and I'm not sure how dpkg would be able to distinguish them apart. It either has to consider them always to be like Depends, which would mean all Recommends need to be present (and on given version if specified), honored on configuration, etc. Or get a new option (like --recommends-as-depends) which would not fly, because some Recommends might have been removed by the user and that should be allowed, or the frontend might have decided it's easier to install them in a later run or whatever, etc. In any case, I think you might want to ask the apt team to add actual documentation for that option, as it seems to only be mentioned in passing on the example configuration file. But coming back to the solutions to your problem, I think you either need a trigger in apache2, which would make sitesummary perform the the actual configuration. Or one from sitesummary, that would tell apache2 to perform any required configuration activation like Jonathan suggested. Thanks, Guillem -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
