Re: Strange CONFLICTS definitions render pkgin full-upgrade impossible

[nia <nia%NetBSD.org@localhost>]

On Fri, May 21, 2021 at 09:02:49AM +0200, Aleksej Lebedev wrote:
> On Fri, May 21, 2021, at 08:37, nia wrote:
> > On Thu, May 20, 2021 at 05:03:08PM +0200, Aleksej Lebedev wrote:
> > > However I checked if there are other packages with that kind of definition left in pkgsrc and I found some:
> > > 
> > > $ find textproc -name Makefile | xargs grep ^CONFL | grep ruby
> > > textproc/ruby-bluecloth/Makefile:CONFLICTS+=	ruby[1-9][0-9]-bluecloth-[0-9]*
> > > textproc/ruby-rttool/Makefile:CONFLICTS+=	ruby[1-9][0-9]-rttool-*
> > > textproc/ruby-amrita/Makefile:CONFLICTS+=	ruby[1-9][0-9]-amrita-*
> > > textproc/ruby-maruku/Makefile:CONFLICTS+=	ruby[1-9][0-9]-maruku-*
> > > 
> > > We happen not to use those but I am quite sure they will cause the same problem with pkgin upgrade.
> > > 
> > > My question: is it just a mistake or is there some kind of good indent behind those CONFLICTS defs?
> > 
> > These packages are self-conflicting between different ruby
> > versions (e.g. ruby26-bluecloth and ruby25-bluecloth can't be
> > installed at once) because they install an unprefixed/suffixed
> > binary to bin/.
> 
> Thanks, this makes sense. However the package should not conflict with itself built against the same version of ruby.
> Which means that either CONFLICTS should be fixed (I don't see how) or, alternatively, pkgin can be patched so that it ignore that kind of conflict.
> 
> It seems to me that the second option is more viable.

No. pkgin is correct.

The package is wrong and the CONFLICT should be removed correctly -
either by renaming the bin/ to include a ruby version suffix
(and using ALTERNATIVES as you suggested), or splitting off the bin/
component into a different package.