[Rpm-maint] Reverse dependencies

Jan Zelený jzeleny at redhat.com
Fri Apr 12 12:22:43 UTC 2013

On 12. 4. 2013 at 14:00:25, Vít Ondruch wrote:
> Dne 12.4.2013 13:53, Vít Ondruch napsal(a):
> > We are able to ensure, the either one or both Ruby implementations
> > will be installed on the system using virtual provide, such as
> > "Requires: ruby(release)", but in RPM there is currently no way how to
> > specify, that rubygem-nokogiri-mri must be available for
> > rubygem-nokogiri to work properly, if MRI Ruby are installed on the
> > system. I can imagine, that something like "Requires: jruby &&
> > rubygem-nokogiri-jruby" and "Requires: ruby && rubygem-nokogiri-mri".
> I'd like to fork this thread immediately to pledge for Reverse
> dependencies mentioned in the presentation (slide 6/7 [2]). They are
> more advanced case for AND dependencies IMO. Let me give you and example.
> I want to get into Fedora another Ruby implementation, such as Rubinius
> and I want to share all the gems. However, there is this
> rubygem-nokogiri, which will not work, unless it will be properly
> modified. However, it there would be reverse dependencies supported, I
> could prepare package rubygem-nokogiri-rubinius, which would provide the
> bindings for Rubinius, if both, rubygem-nokogiri and rubinius packages
> would be installed on my system. I would not need to bother
> rubygem-nokogiri package maintainer to get support for Rubinius.
> It seems that libsolv can handle such cases, so what is reason for RPM
> to not support the reverse dependencies metadata?

The reason is that it's a lot of work to implement proper support of these. 
Both were discussed very closely at our hackfest at DevConf. So you don't need 
to worry, we are well aware of both requests and are planning to address them 
in the future.


More information about the Rpm-maint mailing list