Hi all, Today, I noticed that mock build root prepared by DNF is significantly larger then prepared by YUM (see attached logs). Owners of packages installed into minimal buildroot probably wants to review their dependency chain. I also reported the issue against DNF [1] in case DNF guys wants to improve this situation. Vít [1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634

>>Today, I noticed that mock build root prepared by DNF is significantly >>larger then prepared by YUM (see attached logs). Owners of packages

The cause might be soft dependencies. dnf installs Recommends, while yum ignores them. I think some packages in base buildroot are already using them (gdb).

Am Dienstag, den 18.08.2015, 14:54 -0400 schrieb Matthew Miller: > Hmmm. I think that for the buildroot case, we probably want DNF to > _not_ installer Recommended packages. That makes sense to me. -- Regards, Heiko Adams -- devel mailing list devel(a)lists.fedoraproject.org https://admin.fedoraproject.org/mailman/listinfo/devel Fedora Code of Conduct: http://fedoraproject.org/code-of-conduct

Hi all, Today, I noticed that mock build root prepared by DNF is significantly larger then prepared by YUM (see attached logs). Owners of packages installed into minimal buildroot probably wants to review their dependency chain. I also reported the issue against DNF [1] in case DNF guys wants to improve this situation. Vít [1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634

Dne 18.8.2015 v 16:26 Vít Ondruch napsal(a): > Hi all, > > Today, I noticed that mock build root prepared by DNF is significantly > larger then prepared by YUM (see attached logs). Owners of packages > installed into minimal buildroot probably wants to review their > dependency chain. > > I also reported the issue against DNF [1] in case DNF guys wants to > improve this situation. > > > Vít > > > > [1] https://bugzilla.redhat.com/show_bug.cgi?id=1254634 > This change does not look to be reflected on Koji, so I opened https://fedorahosted.org/rel-eng/ticket/6330 Vít

So it appears this thread was probably not enough. Which keeps us with interesting state where mock by default does not install weak dependencies where Koji installs them. It causes interesting issues already.

Not sure if I should prepare my packages to be ready or not ....

mån 2016-01-25 klockan 18:09 +0100 skrev Marcin Juszkiewicz: > W dniu 25.01.2016 o 17:03, Vít Ondruch pisze: >> So it appears this thread was probably not enough. Which keeps us with >> interesting state where mock by default does not install weak >> dependencies where Koji installs them. It causes interesting issues already. > mock/koji not installing weak dependencies == anything wanting ruby > being broken.

> > Reason: "ruby" suggests "rubypick" which suggests "ruby". > > Packages buildrequire "ruby" but do not get "rubypick" installed (or if > they are lucky they get) so are unable to find Ruby because there is no > "/bin/ruby" executable. If ruby needs ruby-pick to work, then ruby-pick must not be a weak dependency of ruby, but a hard one.

>> > So it appears this thread was probably not enough. Which keeps >> > us with interesting state where mock by default does not install >> > weak dependencies where Koji installs them. It causes >> > interesting issues already. >> >> mock/koji not installing weak dependencies == anything wanting ruby >> being broken. >> >> Reason: "ruby" suggests "rubypick" which suggests "ruby". >> >> Packages buildrequire "ruby" but do not get "rubypick" installed >> (or if they are lucky they get) so are unable to find Ruby because >> there is no "/bin/ruby" executable. > > If ruby needs ruby-pick to work, then ruby-pick must not be a weak > dependency of ruby, but a hard one. > > The koji buildroot really should only install hard dependencies. The > buildroot is supposed to be the minimal possible set needed to build > the package. If a package that would be installed as a weak > dependency of one of the build dependencies is needed to build the > package, that packa is a build dependency too. That's incorrect. EG a package has optional Ruby bindings, it doesn't need ruby to build but if ruby isn't present the ruby binding sub package is empty. The buildroot should have all that is needed to build the desired functionality, nothing more nothing less.

tis 2016-01-26 klockan 10:18 +0000 skrev Peter Robinson: >>>> So it appears this thread was probably not enough. Which keeps us with >>>> interesting state where mock by default does not install weak >>>> dependencies where Koji installs them. It causes interesting issues already. >>> mock/koji not installing weak dependencies == anything wanting ruby >>> being broken. >>> >>> Reason: "ruby" suggests "rubypick" which suggests "ruby". >>> >>> Packages buildrequire "ruby" but do not get "rubypick" installed (or if >>> they are lucky they get) so are unable to find Ruby because there is no >>> "/bin/ruby" executable. >> If ruby needs ruby-pick to work, then ruby-pick must not be a weak >> dependency of ruby, but a hard one. >> >> The koji buildroot really should only install hard dependencies. The >> buildroot is supposed to be the minimal possible set needed to build >> the package. If a package that would be installed as a weak dependency >> of one of the build dependencies is needed to build the package, that >> packa is a build dependency too. > That's incorrect. EG a package has optional Ruby bindings, it doesn't > need ruby to build but if ruby isn't present the ruby binding sub > package is empty. The buildroot should have all that is needed to > build the desired functionality, nothing more nothing less. > > Peter I agree that it is a reasonable expectation that installing "ruby" should make a /usr/bin/ruby binary available. The fact that it currently doesn't when weak dependencies are not installed is a packaging bug in ruby, which should be addressed. It is not a reason to bloat every koji build root for every build of every package in Fedora with every weak dependency of every build dependency of the package being built.