I suggest adding 1 new item under words "Additionally, it is important to ensure that all the dependencies are complete for your package:" ==== added item start * If a package B is known to require a specific version of package A, a versioned dependency should be added, such as >=A-ver.subver ==== added item end This rule sounds obvious to me, but not to every gentoo developer. For example, libXrandr maintainters refuse to add a versioned dependency to libxproto, though libXrandr does not build with older libxproto: http://bugs.gentoo.org/356203 Alternatively, if the new rule above is wrong, and libXrandr does not require versioned dependency, there should be rule === alternative rule start Versioned dependency for package B on package A should be added if (...there goes your rule...) === alternative rule end Reproducible: Didn't try
This is not a devrel issue, but QA. Also if you want this issue rolling, please, post a message to gentoo-dev mailing list where this should be discussed first.
The maintainers are right. If you use ~testing tree you really need to use only testing packages. So why do you want to use an old ( probably stable ) libxproto against libXrandr? I strongly disagree with your proposal and I think that this bug should close as INVALID
(In reply to comment #2) > The maintainers are right. If you use ~testing tree you really need to use only > testing packages. So why do you want to use an old ( probably stable ) > libxproto against libXrandr? Since discussion stated here personally I think that it's better to use versions. This will make our tree even more fool proof and this really will help in case you need to mix stable/unstable/obsolete packages (and sometimes this is really needed). But again I think this is good topic for -dev.
Peter, by doing that we encourage people to use mixed trees. This should be avoided and honestly we can't really do that. There is no way to test every package against every version of its dependences. Having said that, this topic should be moved to -dev hence I am closing this bug to avoid long long discussions
(In reply to comment #4) > Peter, by doing that we encourage people to use mixed trees. This should be > avoided and honestly we can't really do that. There is no way to test every > package against every version of its dependences. Same logic applies to USE flag combinations - it's impossible to test them all. Should we ban packages with number of USE flags more then N, where N is really small number of testable configurations? > Having said that, this topic should be moved to -dev hence I am closing this > bug to avoid long long discussions Agree.