On 09/02/2016 06:57 PM, Stephen Gallagher wrote: > On 09/02/2016 12:54 PM, Robbie Harwood wrote: >> Stephen Gallagher <sgallagh(a)redhat.com&gt; writes: >> >>> On 09/02/2016 07:14 AM, Igor Gnatenko wrote: >>>> >>>> * Weird obsoletes (broken) >>>> "krb5-server" has "Obsoletes: krb5-server-1.14.3-8.fc26.i686". >>>> Basically it will not obsolete anything because it's threated as >>>> package name (and we definitely don't have such package name). >>> >>> This definitely looks odd... Robbie? >> >> This is part of something I was requested to add (from the RHEL >> packaging where we have lines like `Obsoletes: >> krb5-server-1.11.3-49.el7.i686`, added by a previous maintainer) wherein >> the 64-bit versions of packages need to obsolete the 32-bit versions >> because we run into problems if both are installed. >> >> If what's in the spec file is not the correct way to accomplish that, >> what is? I am unable to find documentation for any of this. >> > > Is that because some machines at one point did have both installed? That's kind > of a mess. I'd recommend taking that discussion to > packaging(a)lists.fedoraproject.org and see what they recommend. I think the issue here is just that the syntax is wrong. Instead of what's right now, Obsoletes: krb5-server-1.11.3-49.el7.i686 it should be: Obsoletes: krb5-server <= 1.11.3-49.el7.i686 ... or something along those lines. Right now the problem is that dnf considers the whole string "krb5-server-1.11.3-49.el7.i686" to be a package name while the package name is really "krb5-server". This makes the obsoletes just plain not do anything right now since they don't match any package name.

DNF has nothing to do with Obsoletes. It's up to RPM how to handle it.

On 09/02/2016 12:54 PM, Robbie Harwood wrote: > Stephen Gallagher <sgallagh(a)redhat.com&gt; writes: > >> On 09/02/2016 07:14 AM, Igor Gnatenko wrote: >>> >>> * Weird obsoletes (broken) >>> "krb5-server" has "Obsoletes: krb5-server-1.14.3-8.fc26.i686". >>> Basically it will not obsolete anything because it's threated as >>> package name (and we definitely don't have such package name). >> >> This definitely looks odd... Robbie? > > This is part of something I was requested to add (from the RHEL > packaging where we have lines like `Obsoletes: > krb5-server-1.11.3-49.el7.i686`, added by a previous maintainer) wherein > the 64-bit versions of packages need to obsolete the 32-bit versions > because we run into problems if both are installed.

> If what's in the spec file is not the correct way to accomplish that, > what is? I am unable to find documentation for any of this.