I am forwarding response from Alex.
-------- Pu*vodní zpráva -------- Pr(edme(t: Re: Provides: rubygem(name) automation Datum: Wed, 27 Apr 2011 12:14:15 +0300 Od: Alexander Kurtakov akurtako@redhat.com Spolec(nost: Red Hat Inc. Komu: Vít Ondruch vondruch@redhat.com Kopie: ruby-sig@lists.fedoraproject.org
On 12:07:29 PM Wednesday, April 27, 2011 Vít Ondruch wrote:
Dne 26.4.2011 23:55, Mo Morsi napsal(a):
On 04/21/2011 05:29 AM, Vít Ondruch wrote:
Hello everybody,
Today I was contacted by Aleksandar Kurtakov and he proposed, that he would help us to autogenerate the RPM provides. This functionality is allowed by RPM 4.9 [1], it means for F15+. This would help Aleksandar for better RubyGem integration into rpmstubby [2] and at the end into Eclipse Fedora Packager [3].
Basically we would need to place rubygem.attr (see [4] for maven example) file into /usr/lib/rpm/fileattr and script which would generate the provides (see [5] for maven example). Since this files are required during build, they should be probably part of rubygems package, although it has some cavities.
In theory, this approach could also be used to generate Requires and BuildRequires from gem spec file.
Any thought are highly appreciate.
Vit
Looked this over and I like the direction this is going, bringing more automated dependency checking and analysis into RPM is a good idea, and in conjunction with some additional Fedora policies (this is acceptable to start integrating into Fedora correct?), it seems that alot of developer effort would be alleviated.
The biggest concern that I have at this point is the dependency version analysis, especially with the Ruby packages.
For gems we can parse version information out of the gemspec, for bundler we can parse it out of the Gemfile, but we'll probably run into some cases which this will not be enough, whether the dependencies are not (or cannot) be fully represented or where there is no dependency list per-se (in which case we might be able to get away with parsing the 'require's, there is a few ideas on how to do this here [1]).
One additional question I had is how easy it would be to override the dependency generator. Lets say there is a mistake in the generator or a use case that it does not work for and/or generated incorrect dependencies for, will ruby packagers have to wait until that is resolved to submit their packages. Will explicit dependencies in the spec override the dependency checker?
* You can disable the dependency generator entirely. See http://rpm.org/wiki/PackagerDocs/DependencyGenerator * You can add additional provides manually and let the automatically generated provides just stay as is. It's not a problem if your manual provide is the same as the automatically generated one. You would not have to wait because everything you put in the spec file manually will simply be there no matter what the dependency generator has done.
In my understanding, there will be always both dependencies listed. One set of dependencies will be generated automatically by some dependency generator and there will second set of dependencies entered explicitly in .spec file.
That's not exactly true. Ideally after reaching critical mass with automated provides/requires generated you will not need to add anything manually in most cases unless there is some obscure case.
P.S. I'm not subscribed so the moderator should approve it or someone resend.
Hope that helps, Alex
-Mo[1] http://rubyforge.org/pipermail/gem2rpm-devel/2010-September/000008.html _______________________________________________ ruby-sig mailing list ruby-sig@lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/ruby-sig
On 04/27/2011 06:42 AM, Vít Ondruch wrote:
I am forwarding response from Alex.
One additional question I had is how easy it would be to override the dependency generator. Lets say there is a mistake in the generator or a use case that it does not work for and/or generated incorrect dependencies for, will ruby packagers have to wait until that is resolved to submit their packages. Will explicit dependencies in the spec override the dependency checker?
- You can disable the dependency generator entirely. See
Ya I read through this but did not see the bit about disabling the generators, did I miss it?
- You can add additional provides manually and let the automatically generated
provides just stay as is. It's not a problem if your manual provide is the same as the automatically generated one. You would not have to wait because everything you put in the spec file manually will simply be there no matter what the dependency generator has done.
Tru, though I'm little concerned about dependencies represented in both the automatically generated list and manually in the spec file, eg where the dependency generator thinks its one thing but the packager knows its something else. (see the examples in my previous email).
That being said, from what I've seen I think the majority of cases should work with this and those that don't aren't are good candidates for upstream bug reports. IMO I agree this should save some packaging time overall.
In my understanding, there will be always both dependencies listed. One set of dependencies will be generated automatically by some dependency generator and there will second set of dependencies entered explicitly in .spec file.
That's not exactly true. Ideally after reaching critical mass with automated provides/requires generated you will not need to add anything manually in most cases unless there is some obscure case.
Agreed, though just want to make sure we can handle those obscure cases.
-Mo
ruby-sig@lists.fedoraproject.org
-
Mo Morsi -
Vít Ondruch